Predicting Surface Water Critical Loads at the Catchment Scale Thesis submitted for the degree of Doctor of Philosophy University of London Martin Kernan Department of Geography University College London April 1998
Predicting Surface Water Critical Loads at the Catchment Scale
Thesis submitted for the degree of Doctor of Philosophy University of London
Martin Kernan
Department of Geography University College London
April 1998
ProQuest Number: 10011142
All rights reserved
INFORMATION TO ALL USERS The quality of this reproduction is dependent upon the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.
uest.
ProQuest 10011142
Published by ProQuest LLC(2016). Copyright of the Dissertation is held by the Author.
All rights reserved.This work is protected against unauthorized copying under Title 17, United States Code.
Microform Edition © ProQuest LLC.
ProQuest LLC 789 East Eisenhower Parkway
P.O. Box 1346 Ann Arbor, Ml 48106-1346
Abstract
Current applications of the critical loads concept are geared primarily towards targeting
emission control strategies at a national and international level. In the UK maps of critical
loads for freshwaters are available at 10km^ resolution based on a single representative site
in each grid square. These maps do not take variations of water chemistry within mapping
units into account and are therefore of limited use for application to non-mapped sites. This
thesis describes the development of an empirical statistical model, which uses nationally
available secondary data, to predict freshwater critical loads for catchments lacking the
appropriate water chemistry information.
A calibration exercise using data from 78 catchments throughout Scotland is described.
Water chemistry for each catchment has been determined and each catchment is
characterised according to a number of attributes. Multivariate statistical analysis of these
data shows clear relationships between catchment attributes and water chemistry and
between water chemistry and diatom critical load. The key variables which explain most of
the variation in critical load relate to soil, geology and land use within the catchment. Using
these variables (as predictors) in a regression analysis diatom critical load could be
predicted across a broad gradient of sensitivity = c.0.8). The predictive power of the
model was maintained when different combinations of explanatory variables were used. This
accords the model a degree of flexibility in that model paramaterisation can be geared
towards availability of secondary data.
There are limitations with the model. These relate to the nature of the predictor variables
and the ability of the model to predict critical loads for more sensitive sites. Nevertheless the
ability of the model to differentiate between sensitive and non-sensitive sites offers
considerable scope for environmental managers to undertake national inventories of
catchment sensitivity and specific assessments of individual catchments.
Acknowledgements
This research was funded by the Natural Environment Research Council (Grant GT4/92/17/P). Funding for water chemistry analysis was provided by the UGL Graduate School.
I would like to thank my supervisors Tim Allott and Rick Battarbee and co-supervisor Steve Juggins for their advice and encouragement throughout this research.
I was helped on fieldwork by Jon Cox, Chris Curtis and Dave Ryves, each of whom contributed very useful discussion towards many aspects of this thesis.
Analytical water chemistry was carried out at the Freshwater and Fisheries Laboratory, Pitlochry. I am grateful to Ron Harriman for this and for advice in this area. I would also like to thank Tony Osbourne (Dept of Geology, UCL) and Sarah James (Dept of Geology, RHUC) for help with the ion chromatograph and ICP facility respectively.
Much secondary data was made available to me and I am grateful for the co-operation and assistance of the following:
i) The CLAG Freshwaters sub-group for use of the chemistry and other data from the CLAG database.
ii) The Institute of Terrestrial Ecology who provided me with data from the Land Classification and Land Cover datasets. Permission for use of these data was granted by Bob Bunce and Robin Fuller respectively. I am grateful to Jane Hall, Helen Dyke, Jacquie Ullyet and Mike Brown for help with extracting these and other data held at ITE.
ill) ENSIS for permission to use the Acid Waters Monitoring Data and Mike Renshaw who extracted it for me.
iv) The British Geological Society for granting me a license to digitise geology maps.
v) The Macauley Land Use Research Institute for permission to digitise Scottish soil maps and for use of data. I am particularly indebted to Simon Langan for valuable help and advice with extracting and interpreting these data. Mark Hodson, also at MLURI, aided my interpretation of geological data.
I was helped along the steep GIS/Unix learning curve by Dave Allison (from the Remote Sensing Unit at UCL), Trevor Tsang, Su-min Shen and Ian Carson.
For advice on the application of, and dangers associated with, multivariate statistical techniques I am grateful to John Birks.
For help with putting the final product together I would like to thank Catherine Dalton.
Finally, advice, help and informed opinion across a wide spectrum of themes was readily forthcoming, throughout the course of my research, from Angus Beans, Nigel Cameron, Adrian Chappell, Roger Flower, Janet Hope, Viv Jones, Annette Kreiser, Anson Mackay, Don Monteith, Simon Patrick, Jane Reed, Neil Rose and Julian Thompson. Many thanks to all of them.
Table of contents
Abstract................................................................................................................................................... 2
Acknowledgements.................................................................................................................................. 3
Table of contents .................................................................................................................................... 4
List of tables........................................................................................................................................... 10
List of figu res ........................................................................................................................................ 13
List of appendices.................................................................................................................................. 15
Chapter 1 : Introduction
1.1 Background .................................................................................................................................... 17
1.2 The critical loads approach ............................................................................................................ 18
1.3 Prediction of catchment critical loads - the study rationale........................................................... 20
1.4 Structure of thes is ........................................................................................................................... 21
Chapter 2: Acidification
2.1 Introduction...................................................................................................................................... 23
2.2 Acid ra in ........................................................................................................................................... 24
2.3 Emissions of acidifying compounds ............................................................................................... 25
2.3.1 Sulphur emissions.............................................................................................. 26
2.3.1.1 Natural sources .......................................................................................... 26
2.3.1.2 Anthropogenic sources ................................................................................ 27
2.3.2 Nitrogen emissions.............................................................................................. 28
2.3.2.1 Natural sources............................................................................................. 28
2.3.2.2 Anthropogenic sources ................................................................................ 28
2.3.3 Other emissions ............................................................................................... 29
2.4 Atmospheric transportation and transformation of acidifying compounds...................................... 30
2.4.1 Dry phase transformation......................................................................................... 30
2.4.2 Aqueous phase transformations................................................................................... 31
2.5 Atmospheric Deposition................................................................................................................... 31
2.5.1 Dry deposition.................................................................................................. 32
2.5.2 Wet deposition .............................................................................................................. 33
2.5.3 Cloud droplet (occult) deposition.................................................................................... 33
2.5.4 Monitoring and mapping deposition patterns........................................................... 34
2.6 The links between deposition and surface water acidification....................................................... 36
2.7 Catchment sensitivity....................................................................................................................... 37
2.7.1 Geology and s o ils ......................................................................................................... 37
2.7.1.1 Introduction ................................................................................................. 37
2.7.1.2 Fundamental concepts - soil acidification..................................................... 38
2.7.1.3 Soil as a b u ffe r............................................................................................ 40
2.7.1.4 Geology as a buffer...................................................................................... 44
2.7.2 Land use and catchment management......................................................................... 45
2.7.2.1 Conifer afforestation...................................................................................... 45
2.7.2.2 Upland agricultural improvement .............................................................. 48
2.7.2.3. Catchment liming ........................................................................................ 48
2.7.2.4. Lake liming ................................................................................................. 49
2.7.3 Catchment morphology and hydrology ......................................................................... 49
2.7.3.1 Introduction................................................................................................... 49
2.7.3.2 Catchment morphology ............................................................................... 50
2.1.3.2 Hydrological pathways................................................................................. 51
2.8 An integrated approach : catchment characteristics as predictors of surface water chemistry . . . 52
2.9 Sum m ary........................................................................................................................................ 54
Chapter 3: The Critical Loads Concept
3.1 Introduction...................................................................................................................................... 55
3.2 Critical loads for freshwaters .......................................................................................................... 55
3.2.1 Diatom critical load ....................................................................................................... 57
3.2.2 Henriksen (steady state water chemistry) Critical Load .............................................. 60
3.2.3 The First Order Acidity Balance (FAB) Model .............................................................. 62
3.2.4 Dynamic modelling ....................................................................................................... 64
3.3. Critical load exceedances........................................................................................................... 65
3.4 Mapping freshwater critical lo a d s ................................................................................................ 66
5
3.5 Problems with mapping resolutions ............................................................................................... 67
3.6 Catchment scale critical loads: a predictive m odel......................................................................... 68
Chapter 4: Research Design and Methodology
4.1 Introduction...................................................................................................................................... 71
4.2 Site selection .................................................................................................................................. 71
4.3 Temporal variation in water chemistry............................................................................................. 75
4.4 Sampling techniques....................................................................................................................... 79
4.5 Analytical chemistry......................................................................................................................... 81
4.6 Secondary data sources ................................................................................................................ 82
4.6.1 Phase 1 secondary data ............................................................................................... 83
4.6.1.1 Catchment properties from the CLAG database.......................................... 83
4.6.1.2 Soil Critical Loads........................................................................................ 84
4.6.1.3 Land classification data... .............................................................................. 86
4.6.1.4 Land cover data ........................................................................................... 87
4.6.1.5 Site sensitivity............................................................................................... 88
4.6.2 Phase 2 secondary data ............................................................................................... 90
4.6.2.1 Introduction................................................................................................... 90
4.6.2.2 Catchment delineation.................................................................................. 91
4.6.2.3 Land use........................................................................................................ 91
4.6.2.4 Solid geology ............................................................................................... 92
4.6.2.5 Drift deposits................................................................................................. 96
4.6.2.6 S o il................................................................................................................ 96
4.6.2.7 Other attributes........................................................................................... 101
4.7 Statistical analysis ....................................................................................................................... 103
4.7.1 Ordination ................................................................................................................... 103
4.7.1.1 Introduction................................................................................................. 103
4.7.1.2 Indirect gradient analysis........................................................................... 105
4.7.1.3 Direct gradient analysis... ............................................................................. 106
4.7.1.4 Direct gradient analysis as a data reduction tool ....................................... 108
4.7.2 Variance partitioning ................................................................................................... 109
6
4.7.3 Multiple regression ..................................................................................................... 109
4.8 Discussion .................................................................................................................................. I l l
4.8.1 Land use da ta .............................................................................................................. I l l
4.8.2 Geology data .............................................................................................................. I l l
4.8.3 Soil d a ta ....................................................................................................................... 112
4.8.4 Deposition d a ta ............................................................................................................ 113
Chapter 5: Phase 1 - Preliminary study of model feasibility
5.1 Introduction.................................................................................................................................... 114
5.2 Analysis of the full dataset (954 CLAG sites)............................................................................... 115
5.2.1 Exploratory data analysis of response variables (water chemistry) ........................... 115
5.2.1.1 Summary statistics...................................................................................... 116
5.2.1.2 Correlation Structure .................................................................................. 117
5.2.1.3 Principal Components Analysis ................................................................ 119
5.2.2 Exploratory data analysis - explanatory variables ..................................................... 127
5.2.2.1 Summary statistics for continuous variables.............................................. 127
5.2.2.2 Correlation ................................................................................................. 128
5.2.2.3 Distribution of nominal/ordinal variables..................................................... 128
5.2.3 Direct gradient analysis - chemistry and catchment data .......................................... 131
5.2.3.1 Preliminary Redundancy Analysis (RDA) of the nominal/ordinal
variables................................................................................................... 131
5.2.3.2 RDA of chemistry and catchment variables.............................................. 132
5.2.3.3 Forward selection of explanatory variables .............................................. 138
5.2.3.4 RDA using DCL as a single response variable.......................................... 140
5.3 Analysis of a reduced dataset of more sensitive sites (Ca '^<20G|ieq I’’ ) ................................... 144
5.3.1 Redundancy analysis of water chemistry and catchment variables ........................... 144
5.3.2 Forward selection of environmental variables ........................................................... 148
5.3.3 RDA using DCL as a single response variable ......................................................... 149
5.4 Discussion .................................................................................................................................... 152
Chapter 6: Phase 2 - Model Development and Calibration
6.1 Introduction.................................................................................................................................... 155
6.2 Exploratory analysis - chemistry and catchment datasets............................................................ 157
6.2.1 Water chemistry data ................................................................................................... 157
6.2.1.1 Summary statistics...................................................................................... 158
6.2.1.2 Principal components analysis (PCA)- water chemistry d a ta .................... 162
6.2.2 Catchment data .......................................................................................................... 167
6.2.2.1 Summary statistics .................................................................................... 168
6.2.2.2 Correlation ................................................................................................. 171
6.2.2.S Principal Component Analysis (PCA) - catchment attributes.................. 174
6.3 Direct gradient analysis - chemistry and catchment variables..................................................... 178
6.3.1 Redundancy Analysis (RDA) on full catchment and chemistry datasets.................... 179
6.3.2 Forward selection of catchment variables .................................................................. 184
6.3.3 Redundancy analysis (RDA) using Diatom critical load (DCL) as a
single response variable........................................................................................ 190
6.3.4 Variance partitioning ................................................................................................... 196
6.4 Analysis of a reduced dataset of more sensitive sites (Ca^*<400|ieq T’) .................................... 203
6.4.1 Exploratory analysis of response (water chemistry) variables................................... 204
6.4.2 Exploratory data analysis - explanatory variables....................................................... 209
6.4.3. Direct gradient analysis............................................................................................ 212
6.4.3.1 Redundancy analysis (RDA) on sites where Ca ‘'<400|ieq r ’ .................. 212
6.4.3.2 Forward selection of environmental variables............................................ 215
6.4.3.3 RDA using DCL as a single response variable.......................................... 218
6.5 Sum m ary...................................................................................................................................... 223
Chapter 7: Phase 2 - Model calibration
7.1 Introduction.................................................................................................................................... 226
7.2 Regression results and model diagnostics .................................................................................. 226
7.3 Multiple Regression with more sensitive sites (Ca^"^<400|ieq I"’) ................................................. 236
7.4 Discussion .................................................................................................................................... 237
Chapter 8: Discussion and Conclusion
8.1 Introduction.................................................................................................................................... 239
8.2 Summary of results....................................................................................................................... 239
8.3 Implications of results for model development............................................................................. 241
8.4 Methodological Issues.................................................................................................................. 242
8.4.1 Sampling strategy....................................................................................................... 243
8.4.2 Water chemistry .......................................................................................................... 246
8.4.3 Catchment characterisation ........................................................................................ 247
8.5 Model representation of catchment processes............................................................................. 254
8.5.1 Geology......................................................................................................................... 256
8.5.2 Soil................................................................................................................................. 262
8.5.3 Land use....................................................................................................................... 276
8.4 Model evaluation........................................................................................................................ 286
8.7 Further research possibilities....................................................................................................... 288
8.7.1 Potential improvements in model paramaterisation..................................................... 289
8.7.2 Analysis of catchment data at different spatial resolutions ......................................... 293
8.7.3 Potential for predicting other measures of sensitivity and acid-base
status............................................................................................................................. 294
8.7.4 Model evaluation and further development using national d a ta ................................... 300
8.8 Implications of model improvement for catchment management................................................ 305
8.9 Conclusions .................................................................................................................................. 307
References.......................................................................................................................................... 310
Appendices ........................................................................................................................................ 332
Additional References......................................................................................................................... 396
List of tables
Table 2.1:
Table 2.2:
Table 2.3:
Table 2.4:
Table 4.1:
Table 4.2:
Table 4.3:
Table 4.4:
Table 4.5:
Table 4.6:
Table 4.7:
Table 4.8:
Table 4.9:
Table 4.10:
Table 5.1:
Table 5.2:
Table 5.3:
Table 5.4:
Table 5.5:
Table 5.6:
Estimates of global emissions of sulphur and nitrogen.(From Rodhe et a/.,1995)
Summary of chemical processes in neutralisation of rainfall acidity (from UKAWRG, 1986)
25
43
Soil buffering classes used by Catt (1985, in Hornung, 1990b) to mapsoil neutralising capacity in Wales. Terminology based on classificationby Avery (1980) 43
Buffering capacities of solid geology in Wales (from Hornung at ai, 1990b) 45
Precision, accuracy and detection limits of analytical methods, Freshwater Fisheries Laboratory (FFL) 81
Mineralogical and petrological classification of soil material and critical loads of soils (after Nilsson and Grennfelt, 1988 and Sverdrup and Warfvinge,1988 - modified by Hornung at ai, 1994) 85
Number of sites in each land classification class 86
Aggregated land cover classification (9 classes) 88
Aggregated land cover classification (6 classes) 88
Surface water sensitivity classes as defined by soil and geology classes (after Hornung at ai, 1995a) 90
Categories adopted for classification of the solid geology map (1:625,000)of the UK according to sensitivity to acidification (after Kinniburghand Edmunds, 1984) 94
Classification of individual map units of the solid geology map (1:625,000)of the UK according to sensitivity to acidification (after Kinniburghand Edmunds, 1994) 94
Soil series sensitivity classes (after Hornung at ai, 1995a) 99
Weathering rates for 17 major soil associations (afterLangan et a/., 1995) 101
Summary statistics for untransformed chemistry/response variables(n = 954) 117
Matrix of Pearson product-moment correlations for 14 transformed water chemistry determinands from 954 CLAG sites 118
Results of PCA on transformed water chemistry determinands (954 sites). 121
Ca^* and DCL values for seven sites along the first PCA axis 125
Key chemical values for selected outlying sites (t=|ieq l'\ *=p,Scm \BD=Below detection limits) 126
Summary statistics for untransformed catchment/explanatory variables (a7=954) 127
10
Table 5.7:
Table 5.8:
Table 5.9:
Table 5.10a:
Table 5.10b:
Table 5.11:
Table 5.12:
Table 5.13:
Table 5.14a:
Table 5.14b:
Table 5.15:
Table 5.16:
Table 6.1:
Table 6.2:
Table 6.3:
Table 6.4:
Table 6.5
Table 6.6:
Table 6.7:
Table 6.8a:
Table 6.8b:
Table 6.9:
Table 6.10:
Table 6.11:
Table 6.12:
Table 6.13:
Table 6.14:
Matrix of Pearson product-moment correlations between transformed catcfiment attributes for 954 CLAG sites 128
RDA of chemistry and selected classifications 132
Results of RDA on chemistry and environmental variables (954 sites) 134
Forward selection of environmental variables 139
RDA summary using variables identified in the forward selection procedure 139
Results of an RDA on DCL and catchment attributes (954 sites) 141
Forward selection with DCL as a sole response variable 141
Results of RDA on chemistry and environmental variables (469 sites) 145
Forward selection of environmental variables 148
RDA summary using variables from forward selection 148
Results of an RDA on DCL and catchment attributes (459 sites) 149
Forward selection with DCL as a sole response variable 150
Summary statistics for untransformed chemistry/response variables in the Phase 2 (calibration) dataset (n=78) 158
Results of PCA on transformed water chemistry determinands (n = 78) 163
Summary statistics for untransformed catchment/predictor variables (n=78) 169
Matrix of Pearson product-moment correlations for 31 transformed catchment variables (n = 78)
Results of a PCA on transformed catchment attributes (see Table 6.3 for full description of variables)
Values for dominant catchment variables for 5 sites along PCA Axis 1
Results of RDA on chemistry and catchment variables {n=78)
Catchment variables identified by the forward selection procedure
RDA summary using catchment variables identified by forward selection
Results of an RDA on DCL and catchment attributes
Forward selection with DCL as a sole response variable
Redundancy Analyses on Catchment Variable Components
Results of (partial) RDA (A=Soil, B=Geology, C=Land use, D=Extrinsic, AnBnc=Unique covariance between A,B and C)
Results of PCA on transformed water chemistry determinands for sites where Ca^ <= 400peq T’
Results of a PCA on transformed catchment attributes on sites with Ca + < 400eq I''
173
175
177
180
185
185
191
193
197
201
206
210
11
Table 6.15: Results of RDA on chemistry and environmental variables for siteswhere Ca ‘'<400neq T’ 213
Table 6.16a: Forward selection of catchment variables at sites where Ca ‘'<400^eq T’ 217
Table 6.16b: RDA summary using variables from forward selection at sites whereCa^+<400|ieq I'' 217
Table 6.17: Results of an RDA on DCL and catchment attributes (using forwardselection) for sites where Ca '" < 400p,eq 1-1 219
Table 6.18: Redundancy Analysis (with forward selection) on datasets of varyingsensitivity 221
Table 7.1: Multiple linear regression output with G1, G2, SOL, and LC2 aspredictors. 228
Table 7.2: Multiple linear regression output with G1, G2, hF and LG2 as predictors 232
Table 7.3: Multiple linear regression output with G1, G2, and LC2 as predictors 233
Table 7.4: Results of regression analyses on a variety of catchment attributecombinations 234
Table 7.5: Multiple linear regression output with SCL3 and LC2 as predictors -sites where Ca^*<400|ieq 1' 237
12
List of figures
Figure 1.1; Schematic illustration showing critical load variation 19
Figure 3.1: Critical and target loads concept (after Battarbee et al., 1994).The critical load for a site is exceeded at (a)\ (b) and (c) are critical loads for specific species. A target load (T) can be chosen to protect selected species as acid deposition declines in the future.Full recovery is represented by point (a) on the ’future’ curve. 56
Figure 4.1: Location of CLAG sites used in Phase 1 analysis 73
Figure 4.2: Location of Phase 2 model calibration sites 76
Figure 4.3: Calcium concentrations at minimum, mean and maximum flow for 11Acid Waters Monitoring Network sites 78
Figure 4.4: Flow diagram illustrating the soil variables available for characterisingcatchments 102
Figure 5.1: PCA correlation biplot of 15 water chemistry determinands for 954 CLAG sites(plotted using CALIBRATE - Juggins and ter Braak, 1993) 122
Figure 5.2: PCA biplot showing the position of sites relative to the first two PCA axes(vectors have been multiplied by three for clarity) 124
Figure 5.3: Scatterplot of DCL against PCA axis 1 site scores 126
Figure 5.4: Bar charts showing the distribution of sites for the nominal/ordinalexplanatory variables 129
Figure 5.5: RDA correlation biplot of chemistry and surrogate catchment data (954 sites)showing water chemistry (solid vectors), continuous catchment parameters (dashed vectors) and dummy variables (filled circles) 137
Figure 5.6: Box plots showing DCL values classed according to nominal explanatoryvariables 142
Figure 5.7 Scatterplots showing DCL against continuous catchment variables 143
Figure 5.8: RDA biplot of chemistry and environmental data (469 sites) 147
Figure 5.9: Box and whisker plots of distribution of site DCL according to classificationvariables - sensitive sites 150
Figure 5.10: Scatterplot showing DCL against continuous environmental variables -sensitive sites 151
Figure 6.1: Percentage of Diatom Critical Load classes for the Phase 1, Phase 2 andCLAG mapping datasets (Curtis at al, 1995) 161
Figure 6.2: PCA correlation biplot of Phase 2 water chemistry (plotted using CALIBRATE -Juggins and ter Braak, 1993) - vectors have been multiplied by three to aid clarity 165
Figure 6.3: Scatterplot of DCL against PCA Axis 1 site scores 167
Figure 6.4: PCA correlation biplot of Phase 2 water chemistry (plotted using CALIBRATE -Juggins and Ter Braak, 1993) - vectors have been multiplied by three for clarity. (For key, see Table 6.3) 177
13
Figure 6.5: RDA biplot of chemistry and catchment data showing water chemistry(solid vectors) and catchment attributes (dashed vectors). See Tables 6.1 (chemistry) and 6.3 (catchments) for key 183
Figure 6.6: RDA biplot of chemistry and catchment data showing water chemistry(solid vectors) and catchment parameters (dashed vectors) for soil map unit, the latter included following forward selection. 186
Figure 6.7 Scatterplots showing DCL against variables selected by the forward selectionprocedure 194
Figure 6.8: Schematic representation of the explanatory variable components and covariancesused in (partial) RDA on significant catchment variables and DCL. ATIB is the unique covariation between A and B, AFlBIlC between A, B and C etc., 200
Figure 6.9: Bar chart showing the results of (partial) RDA on catchment attributes andDCL 202
Figure 6.10: PCA plot of water chemistry determinands from sites whereCa + <=400peq r' 207
Figure 6.11: Scatterplot of Diatom Critical Load (DCL) against PCA Axes scores for siteswhere Ca^<400peq T’ 208
Figure 6.12: PCA correlation biplot of transformed catchment attributes on sites whereCa^"<400peq I'' 211
Figure 6.13: RDA biplot of chemistry and chemistry attributes for sites whereCa^^<400peq I ’ 214
Figure 6.14: RDA biplot using variables from forward selection at sites whereCa^"<400peq r' 218
Figure 6.15: Scatterplots showing diatom critical load against variables selected byforward selection procedure (sites where Ca<400peq 1' ) 220
Figure 7.1: Residuals plotted against predicted DCL 229
Figure 7.2: Distribution analyses of studentized residuals 230
Figure 7.3: Plot of observed DCL against predicted DCL 231
14
List of appendices
4.1; Sampling strategy originally adopted during Phase 2 332
4.2: Phase 2 site locations 339
4.3: Scatterplots of calcium concentration (peq 1-1) against
flow (cumecs) for selected Acid Waters Monitoring Network sites 343
4.4: Water chemistry for Phase 2 sites (including critical load
values) 349
4.5a: Summary of ITE Land classification system (from
Bunce et al., 1982) 352
4.5b: Hierarchical aggregations of ITE Land classification system
(from Bunce et a!., 1982), aggregated using TWINSPAN (Hill, 1979) 357
4.6: Classes from the Land Cover Map of Great Britain. Correspondence
between the 25 'target' cover types and 17 'key' cover types
(from Fuller and Groom, 1993a) 359
4.7: Percentage of each land cover class at each site (25m resolution,
6 class aggregation) 361
4.8: Percentage of each sensitivity class for geology in each
catchment 363
4.9: Percentage of each drift type in each catchment 365
4.10: Phase 2 catchment values for each soil variable 366
4.11 : Miscellaneous data for Phase 2 catchments 369
5.1 : Summary statistics for transformed chemistry/ response variables 371
5.2: Summary statistics for transformed catchment/explanatory variables 372
5.3: Exploratory data analyses of the sensitive sub-set (Ca2^"’<2GGeq 1-1) 373
15
6.1: Pearson product-moment correlations for 16 transformed water chemistry
determinands (full dataset, n = 78). 381
6.2: Comparative analysis of alternative sub-sets of the catchment data 383
6.3: Summary statistics for untransformed water chemistry variables
(sensitive subset) 392
6.4: Matrix of Pearson product-moment correiations for 16 transformed
water chemistry determinands (sensitive subset, n = 46). 393
6.5: Summary statistics for untransformed catchment attributes -
sites where Ca^* <=400peq 1' 394
6.6: Matrix of Pearson product-moment correlations for 28 transformed
catchment attributes (sensitive subset, n = 46) SCL2 not present
in any catchment). 395
16
CHAPTER 1 : INTRODUCTION
1.1 Background
The deposition of anthropogenicaiiy derived acidic precipitation onto terrestrial and aquatic
ecosystems is now generally recognised as a major environmental problem, particularly over
large areas of North America and Europe (Beamish and Harvey, 1972; Gjessing etal., 1976;
Thompson et al., 1980; Wright et a!., 1980; Cowling, 1982). The consequences for aquatic
ecosystems are well documented (Harriman and Morrison, 1982; Ormerod and Wade, 1990;
Muniz, 1991; Gorham, 1992; Havas and Rosseland, 1995). In response to this, a number
of international protocols have been signed seeking to limit and ultimately reduce the
emissions of acidifying compounds. In 1985 the sulphur protocol was signed by most
UNECE member states. This aimed to reduce national emissions of SOg by at least 30%
by 1993, based on levels in 1980. The subsequent EG Large Combustion Plant directive
(LCPD) requires an emission reduction of 60% from large combustion plants by 2003, also
based on a 1980 start year. This further stipulates that UK LCPD NO emissions be reduced
by 30% by 1998.
This kind of emission control strategy employs a blanket reduction approach. Clearly,
however, reductions need to be targeted at those countries where emissions are greater.
In addition, the amount of damage to ecosystems varies from region to region, as does the
potential for further damage. Consequently, to optimise reductions so that they are reduced
where most needed, these spatial variations need to be considered. Those areas with low
neutralising capacities are more susceptible to acidification than well buffered areas. Such
considerations led to the development and adoption of the critical loads approach in Europe
and North America.
17
1.2 The critical loads approach
The most commonly used general definition of critical loads is that proposed by Nilsson and
Grennfelt:
"the highest deposition of acidifying compounds that will not cause chemical changes leading
to long term harmful effects on ecosystem structure and function".
The development of the critical loads approach has been widely reviewed (e.g. Henriksen
etal., 1990; Bull, 1991; Brodin and Kuylenstierna, 1992; Kàmâri, eta!., 1992a; CLAG 1994;
UN EGE, 1994; Bull, 1995). The critical loads approach for freshwaters has resulted in the
production of national (CLAG, 1994) and international maps (Hettelingh et a!., 1991;
Downing etal., 1993; Posch etal., 1995) of critical loads values. Mapped data for deposition
of acidifying compounds used in conjunction with the critical loads maps provide a picture
of areas where critical loads are exceeded. Future deposition scenarios can be used to
assess the effects of emission reduction strategies. The approach has now been
incorporated into the second Sulphur Protocol, signed in 1994, which recommended
emission reductions both on the basis of environmental effects and the cost of control
strategies (UN ECE, 1994). In the UK, critical loads are now used as part of pollution control
policy (HMSO, 1990).
The European and UK mapping exercises are very much geared towards targeting emission
control strategies at regional, national and international levels. The UK freshwater critical
loads maps show critical loads for sulphur and where these have been exceeded throughout
the UK (CLAG, 1994). These are mapped in a grid form at lOkm^ resolution. However the
the critical load value for each square is not necessarily representative of the sensitivity of
all the lakes and streams in that square (Curtis etal., 1995) (Figure 1.1). As a consequence,
18
Figure 1.1: Schematic illustration showing critical load variation with mapping resolution. Although the 10km sguare exemplified has a mapped critical load of <0.2keg ha'^ y r '\ the critical loads for three sub-catchments within the sguare are much more variable.
Critical load (kea'ha/vr)
100km
< 0.2
n 10 - 2 . 0
10km
19
maps at this resolution, using a grid based approach, are of limited use for management and
assessment of specific catchments. Thus although the national critical loads map for the UK
is used by the Forestry Authority to identify where afforestation by trees might lead to
freshwater acidification, it is acknowledged that the map cannot be used to determine the
susceptibility of surface waters in individual catchments (Forestry Authority, 1993). The
critical loads approach, as currently applied, is inappropriate where catchment scale
assessments are necessary. In an applied context these are required by forestry
organisations (e.g. Forest Authority), conservation bodies (e.g. Countryside Commission for
Wales, Scottish Natural Heritage, English Nature) and pollution control organisations (e.g.
Environment Agency) to examine ecosystem response to changing land use and changing
industrial emission patterns, and to assess the likely consequences on catchment organisms
of increased acid deposition. In addition, understanding and prediction of the ecosystem
responses to anthropogenic acid loading is best approached at a catchment scale where
well defined boundaries enable assessment of the interactions between terrestrial and
aquatic systems to be made (Hornung et al., 1990a). As a consequence, there is a need
for an approach where the sensitivity of specific catchments can be gauged.
1.3 Prediction of catchment criticai loads - the study rationaie
Currently assessments of surface water critical loads can only be achieved by an analysis
of water chemistry. However, water chemistry data are not readily available at a national
level and, where the sensitivity of a large number of catchments across a wide geographical
area needs to be assessed, the only existing approach is to undertake costly water sampling
programmes. Data relating to catchment characteristics (e.g. soil, geology, land use) are
more commonly mapped and are available nationally. Given the integrated nature of the
terrestrial and aquatic systems within catchments it is likely the characteristics of the former
will influence the nature of the latter.
20
The overall aim of this thesis is to examine the character of the reiationships between
catchment attributes and surface water chemistry and to assess whether these can be used
to develop an empirically based model which will predict critical loads from quantified
catchment characteristics. In an applied context, it is hoped that the use of nationally
available catchment data to calibrate the predictive model will enable critical loads for
surface waters to be predicted for any site throughout the UK.
1.4 Structure of thesis
Chapter 2 introduces the concepts and processes of surface water acidification. It comprises
an integrated examination of the atmospheric emission of acidifying compounds, the transfer
and transformation of these compounds, deposition, chemicai reactions in the soil/vegetation
environment and the geochemical and hydrological processes operating between the soil
and surface water spheres. Within this context previous attempts at relating catchment
characteristics to aspects of surface water chemistry are examined. This chapter shows how
processes operating within the catchment dictate surface water chemistry and thus
determine sensitivity. The predictive model is based on the strength of these relationships.
The background, development and current issues relating to critical loads are reviewed in
Chapter 3. The use of modelling, at a variety of scales, to predict freshwater critical load is
introduced and discussed with regard to the requirements of catchment scale applications.
The methodology used to develop the predictive model is described in Chapter 4. The
origins and derivation of the data used to represent catchment characteristics are presented
together with the statistical techniques employed in model calibration. Chapter 5 presents
the results of a preliminary analysis of secondary water chemistry and a variety of surrogate
catchment characteristics. Multivariate statistical techniques are used to assess the feasibility
of using catchment data to predict surface water chemistry. This approach is subsequently
21
developed in Chapter 6 where a calibration water chemistry dataset is used in tandem with
catchment specific data to identify catchment variables which most explain variation in,
initially, water chemistry as a whole and, subsequently, critical load. Chapter 7 describes a
series of multiple regression analyses which use these key catchment variables as
predictors of freshwater critical loads.
Discussion in Chapter 8 initially focuses on the limitations and uncertainties of the predictive
model. A number of suggestions for improving model utility are presented. The value of the
model as a tool for catchment management is discussed both in its present form, and
following potential improvement.
22
CHAPTER 2 : ACIDIFICATION
2.1 Introduction
This chapter examines the processes which are responsible for the acidification of surface
waters and how these relate to the physical systems within which they operate. These
processes begin with the emission of acidifying compounds of sulphur (S) and nitrogen (N)
into the atmosphere from a wide variety of sources. Chemical transformation in the
atmosphere alters the nature of these compounds prior to deposition. Once deposited as dry
or aqueous media the compounds are subsequently modified, to varying degrees, by
interaction with vegetation, soil and geology and, via a variety of hydrological pathways,
reach streams and standing water bodies within the catchment. After introducing the
atmospheric processes which result in elevated acidity levels in precipitation, the primary
objective of this chapter is to examine the catchment processes which occur at the interface
between water and soil, geology and vegetation. The hypothesis here is that these
processes, because they are involved in modifying the chemical composition of incoming
precipitation, ultimately determine the chemistry of catchment surface waters. The factors
or attributes which most influence catchment sensitivity to surface water acidification will
need to be represented in a predictive model. More emphasis is placed on the role of
sulphur cycling within the catchment rather than nitrogen cycling because the model is
calibrated for critical loads for sulphur (see Section 3.6). Leaching of N species also leads
to acidification (or eutrophication). However the processes governing N cycling are much
more complex than 8 because the former is also involved in many biological reactions.
These processes are discussed more fully elsewhere (e.g. Gunderson and Bashkin, 1994;
Reynolds and Edwards, 1995).
23
2.2 Acid rain
’Acid rain’ as a physical phenomenon is not a recent, nor anthropogenicaiiy induced,
development. If pH values less than 7 are to describe acidity then rain is always acidic. At
equilibrium, unpolluted rain has a pH of approximately 5.67 (Kennedy 1992) a result of
naturally occurring acids and bases and atmospheric reactions, mainly with carbon dioxide
(COg) which forms carbonic acid (HgCOg). This is, in effect, a theoretical value as rain can
be contaminated by wind blown alkaline dusts which raise pH and by sulphur and nitrogen
from volcanic and biological sources which can lower it. Background pH values of 4.0 to 6.0
have been recorded in remote areas of the Southern Hemisphere (Galloway et al., 1982)
and it has been suggested that rainfall pH at pristine sites varies between 4.5 to 5.6 as a
result of temporal and spatial variations in the sulphur cycle (Charleson and Rodhe, 1982).
This range is proposed as the probable background level of acidity that would have
characterised the precipitation of pre-industrial Europe (Galloway at a!., 1982; Irwin and
Williams, 1988).
The term ’acid rain’ as used to describe rain polluted by anthropogenic means was first used
in connection with air pollution and its effect on the buildings and vegetation of urban
England following observations that rain approaching Manchester was found to contain
sulphuric acid proportional to its distance from the town (Smith, 1852). Subsequent research
contended that acid rain over the Lake District stemmed from air masses that were acidified
as they passed over the industrial regions to the south and east and that the deposition of
large amounts of sulphuric acid over a period of a hundred years had probably initiated
significant ecological changes in the bog pools and upland tarns of the region (Gorham,
1958). Ten years later it was argued that ’acid rain’ was acidifying lakes and killing fish in
Sweden and that the origins of this polluted precipitation were the heavily industrialised parts
of Britain and continental Western Europe (Oden, 1968). The claims that air pollutants from
24
Britain were responsible for acidified precipitation were largely ignored throughout the
1970’s. It was not until the 1980’s that the concept of long range trans-boundary air pollution
became widely accepted and import/export balances between countries of acidifying
pollutants such as sulphur dioxide (SO^ and nitrogen oxides (NOJ are now calculated on
an annual basis (Iversen et al., 1991 in Lovblad et al., 1992).
2.3 Emissions of acidifying compounds
Elevated acid levels in precipitation stem primarily from the emission of compounds of
sulphur and nitrogen oxides as well as ammonia (NH3) (Rodhe et al., 1995) although there
are a variety of other contributors including hydrochloric acid and volatile organic compounds
(Irwin and Williams, 1988). Table 2.1 summarises current estimates of global emissions of
S and N compounds.
Table 2.1: Estimates of global emissions of sulphur and nitrogen. (From Rodhe eta!., 1995)
Source Sulphur
(TgS yr-')Oxidised (NO,)
NitrogenReduced (NH,)
(TgN y r ‘)
Anthropogenic:
Fossil fuel combustion 70 - 80 20 .
Biomass Burning 0.8 - 2.5 6 2Fertilizers - - 2Domestic animals - - 22
Sub-total 71 - 83 26 30
Natural:
Soils and vegetation 0.2 - 4 4 5Volcanoes 7.0 - 10 - -
Lightning - 5 0Oceans 10 - 50 - 7Wild animals - - 2.5
Sub-total 17 - 64 9 15
Total 88 - 147 35 45
25
2.3.1 Sulphur emissions
2.3.1.1 Natural sources
Natural sources of atmospheric sulphur are derived, in order of importance, from biogenic
sources, sea-spray and geothermal activity. Sulphates derived from sea-spray do not directly
contribute to acid deposition as they occur in neutralized form (Rodhe etal., 1995) and need
not be considered here. It should be noted however that ion exchange processes,
specifically Na displacing acidic and AP , can depress the pH of runoff water following
precipitation inputs with high concentrations of marine salts (Langan, 1989; Harriman etal.,
1995a).
The most important natural source of atmospheric sulphur is the biological reduction of
sulphur compounds (Cullis and Hirschler, 1980). These are generated by the non-specific
reduction of sulphur in marine algae, soils and decaying vegetation (Rassmussen, 1974) and
by bacteria specifically reducing certain sulphur compounds (Hallberg et a!., 1976). It is
thought that the principal sulphur compound emitted biogenically is hydrogen sulphide (Cullis
and Hirschler 1980). However, the importance of organic sulphur compounds such as
dimethyl sulphide (DMS) and carbon disulphide, derived from organic sulphur in algae,
plants and animals as well as inorganic sulphate, has also been noted (Davison and Hewitt,
1992; Liss et a!., 1994; Tarrasson et a!., 1995) and it is argued that this is the primary
mechanism for the natural transmission of biogenic sulphur to the atmosphere (Rassmussen,
1974).
The overwhelming contribution to geothermal emissions is from volcanic eruptions which
produce significant amounts of sulphur dioxide and hydrogen sulphide (Kellog etal., 1972).
Although long range stratospheric transportation of volcanic sulphur compounds has been
26
found to occur (Cullis and Hirschler, 1980; Rampino and Self, 1982), sulphur bearing
compounds have a limited residence time in the atmosphere (Charleson and Rodhe, 1982)
and the acidifying effect of a volcanic eruption is liable to have a strong, short-lived local
bias (e.g. Letter and Birks, 1993).
2.3.1.2 Anthropogenic sources
The magnitude of anthropogenic emissions of sulphur has increased markedly over the past
century although the contributions from coal and petroleum combustion, petroleum refining
and the smelting of non-ferrous ores have changed relative to each other to a considerable
degree (Cullis and Hirschler, 1980).
The most abundant source of anthropogenically derived atmospheric sulphur remains the
combustion of coal and its by-products (Galloway, 1995). A substantial amount of the coal
used industrially and domestically contains over 2% sulphur, about half of which is present
as iron pyrite (FeSg), the remainder being organic (Kennedy, 1982). Sulphur dioxide is
readily produced when these elements are burned, for example;
4FeS2 + 11O2 => 2 Fe203 + 8SO2 (2 .1)
Petroleum products also contribute significantly to elevated levels of atmospheric sulphur.
Despite the fact that petrol consumption has expanded more rapidly than that of coal, levels
of sulphur emission have increased less rapidly than the total consumption of petrol (Cullis
and Hirschler 1980). Both recovery and desulphurisation processes have become more
efficient.
In recent years there has been growing pressure for industrialised nations to limit the
27
increase in sulphur emissions by adopting energy conservation policies and employing
desulphurisation technology. This, together with greater emphasis on gas as a fuel and a
decline in heavy industry has been reflected in a stabilisation of emissions. In fact,
anthropogenic sulphur emissions have declined in Europe and North America over the past
decade (Hultberg etal., 1995) although increases have been observed in Asia, particularly
China (Rodhe at a!., 1995).
2.3.2 Nitrogen emissions
Nitrous oxide (NOJ emissions (comprising both nitric oxide (NO) and nitrogen dioxide (NOg))
have a dual role in acid deposition. They are crucially involved in the photochemical
production of ozone and OH radicals, important factors in the atmospheric reactions leading
to acidification of precipitation and, more directly, as precursors of acidity (Irwin and Williams
1988). While, in the atmosphere, reduced N in the form of ammonia (NHg) neutralizes nitric
(HNOg) and sulphuric acid (HgSOJ (ApSimon etal., 1987). However, NH (NHg + N H / from
dry and wet deposition respectively) also has the capacity to cause acidification in
ecosystems (Van Breeman et al., 1982).
2.3.2.1 Natural sources
The primary sources of natural NO and NHg emissions are essentially those involving
ecosystem losses through dissimilation and denitrification. These include biomass burning,
ammonia oxidation, microbial activity and marine photolytic and biological processes. Other
sources of NG include lightning production and stratospheric input (Irwin, 1989).
2.3 2.2 Anthropogenic sources
28
The combustion of fossil fuels constitutes the main source of anthropogenic NO . This is
derived both from the nitrogen held in the fuel and from the oxidation of atmospheric
nitrogen. As a consequence NO emission is dependent on the combustion processes as
well as the properties of the fuel, making quantification more difficult than SOg emissions.
A significant proportion of the NO produced in this way stems from motor vehicle exhaust
(Williams, 1987) which has increased substantially since the 1940’s both absolutely and
relative to other sources of NO .
A second major source of anthropogenically derived NO emissions is biomass burning for
agricultural land clearance. Biomass burning also accounts for a significant amount of NHg
emissions although these are dominated by livestock wastes and fertilizer application (see
Table 2.1). Other minor sources include traffic exhaust, soil microbial activity, coal
combustion and human respiration (ApSimon al et al., 1987; Buijsman at a/.,1987).
Emissions of NHg have increased recently as a result of more intensive animal husbandry
(ApSimon etal., 1987).
Spatially, the global distribution of anthropogenic NO emissions are broadly comparable
with those of SOg, with high concentrations in Europe and North America. However, the
recent decreases in SOg emissions have not been mirrored by a decline in NO (Irwin,
1989). Approximately 90% of global NHg emissions originate in Asia where food production
contributes a higher proportion of N emissions than fossil fuel consumption (Galloway,
1995).
2.3.3 Other emissions
Volatile organic compounds (VOCs) are comprised of reactive hydrocarbons and
oxygenates, and are important in the formation of acid compounds in the atmosphere by
29
virtue of their involvement in the generation of oxidizing radicals (Inwin, 1989). VOC
emissions are extremely difficult to quantify as they arise from sources other than
combustion and agricultural activity, including evaporation and certain industrial and
commercial processes (Irwin and Williams, 1988).
2.4 Atmospheric transportation and transformation of acidifying compounds
When emitted from terrestrial and oceanic sources S is typically in an oxidised state. Both
oxidised and reduced N are common. These pollutants typically remain in the atmosphere
for only a few days before they are deposited. During that time SOg and NG may be
transported for hundreds of kilometres and undergo certain physical or chemical
transformations (UKRGAR, 1990). The chemical fluxes which characterise transformation
between the original compound and that which is ultimately deposited will vary according to
whether the compounds are subject to dry or aqueous transformations. The most important
interactions tend to be those with the oxidising species (particularly OH radicals) and NHg.
2.4.1 Dry phase transformation
In the gas phase, oxidation of S requires reactions primarily with the OH radical although
under certain pH conditions ozone and hydrogen peroxide (HgOg) become important
oxidising agents (Penkett etal., 1979). During gas phase transformations NOg will compete
with SO4 for OH radicals and the former will tend to oxidise preferentially . The reactions
between the OH radical and oxides of S and N produce HgSO and HNO3 respectively, both
strong acids but with substantially different dry deposition velocities (Irwin and Williams,
1988).
Ammonium sulphate aerosol is produced by the reaction of HgSO and NHg which occurs
30
very rapidly. Ammonium nitrate aerosols are formed either by the scavenging of HNO3 by
coarse particles (sea salt based in maritime air and alkaline soil particles in continental air)
or by the reaction of HNO3 with ammonia producing fine particles (Den/vent, 1987). The dry
deposition rates of these aerosols are likely to be fairly low and as a consequence they have
a relatively long residence time. The production rates of nitrate aerosol and nitric acid are
an important controlling mechanism in determining the balance between residence times
(and thus transportation distances) of sulphur and nitrogen in the atmosphere.
2.4.2 Aqueous phase transformations
SOg oxidises in rain and cloud phases, primarily through reactions with ozone (O3) and HgOg
(Penkett et al., 1979). The relative importance of these oxidising agents remains unclear.
Reactions with O3 are limited by ozone solubility and pH while HgOg is highly soluble and not
dependent on pH. Oxidation of SOg in the aqueous phase is of great importance and may
account for up to 70% of sulphate in precipitation (Scire and Venkatram, 1985). During
winter when photochemical activity, and thus OH concentration is low, it has been suggested
that such processes are the only means of oxidising SOg (Clark at al., 1984). There is no
significant aqueous phase transformation mechanism in the production of nitric acid.
The removal of both sulphur and nitrogen species in the aqueous phase involves scavenging
of both gaseous species and particulate aerosols (Irwin and Williams, 1988). With regard to
the former, SOg removal is limited by the poor solubility in water although aqueous phase
oxidations can increase the SOg removed from the gas phase.
2.5 Atmospheric Deposition
There are three means by which acidic species may be deposited onto terrestrial and
31
aquatic systems. These are termed dry deposition, wet deposition (through precipitation) and
cloud droplet deposition (droplet impaction onto vegetation surfaces).
2.5.1 Dry deposition
The dry deposition of gases and particulates involves a transfer from the boundary layer to
the vicinity of the surface, molecular diffusion and uptake at the surface by dissolution,
sorption or chemical reactions (Reynolds and Ormerod, 1993). Untransformed oxides are
deposited by adsorption and absorption while transformed gases fallout onto ground and
vegetation surfaces. These gases include HNO3, HCI and NHg. The rate of uptake is
governed by conditions at three levels. Above and within the forest canopy, deposition is
dependent on windspeed and the aerodynamic roughness of the vegetation surface. Tilled
soil, moorland and forestry are characterised by increasing surface roughness. At the
vegetation/atmosphere interface and within the stomata, rates of uptake are controlled by
molecular diffusion. At leaf surfaces uptake is governed either by chemical reactions
occurring between the leaf surface and the gas or by entry into the leaf via stomata pores
and subsequently by solution in the intercellular fluid. The dominant control over deposition
rates will depend on the reactivity of the gas deposited (Fowler et al., 1989). Other factors
can influence the uptake pathways of individual gases. At night uptake of SOg and NOg
takes place via chemical reactions on the surface of the leaf whereas, during the day, when
stomata pores are open, the uptake occurs through the pores and subsequently, in solution
in intercellular fluids (Fowler and Cape, 1985). The control here is stomatal opening which
is determined by changes in temperature and light. Surface wetness is also a factor which
influences chemical reactions on the leaf surface although, for many vegetation types, dry
deposition of SOg to wetted surfaces is not appreciably greater than uptake on dry surfaces
(Fowler and Cape, 1985).
32
2.5.2 Wet deposition
Wet deposition comprises the atmospheric acids (and bases) which are deposited onto
terrestrial and aquatic ecosystems via precipitation. This can occur, for example, when
H2 SO4 is incorporated into water droplets or ice crystals and falls to the ground as
precipitation (rainout). Alternatively, if in particulate form, H^SO^ or HNO3 can be removed
from the atmosphere by raindrop impaction (washout). This includes the seeder feeder effect
(Bader and Roach, 1977) which involves the scavenging of sulphuric and nitric acid held in
mist or low lying orographic cloud formed as part of frontal weather systems by precipitation
from overlying clouds (Bergeron, 1965). This tends to increase ionic deposition in cloud-
capped upland areas. The feeder cap cloud is generally characterised by higher
concentrations of acidic species than the precipitation from the seeder cloud above
(Carruthers and Choularton, 1984). The lower mountain cloud caps incorporate the higher
concentrations of ions in the atmospheric boundary layer whereas in the frontal clouds the
processes of raindrop formation are initiated by the vapour growth of snowflakes. This does
not efficiently incorporate the dissolved particulate cloud droplets and, as the snowflakes
melt at lower altitudes, raindrops are formed which scavenge cloud droplets in the feeder
cloud as a result of collision coalescence. Frontal weather systems are responsible for much
of the precipitation over upland areas (Fowler etal., 1995) as the moist boundary layer rises
over elevated terrain. Thus the seeder-feeder effect is primarily responsible for deposition
of acidifying compounds in these areas, a supposition supported by experimental work
(Fowler at a!., 1988; Dore et a!., 1992; Inglis et a!., 1995).
2.5.3 Crowd droplet (occult) deposition
Exposed vegetation in upland areas can directly intercept water droplets held in wind driven
cloud and fog. Concentrations of acidic species in cloudwater can be considerably greater
33
than in rainfall in the same area (Crossley etal., 1992) and it is suggested that cloud droplet
deposition in areas prone to low cloud could increase wet deposition estimates by up to 2 0 %
above that detected in rainfall gauges (Dollard et a!., 1993). Deposition loadings vary with
different vegetation communities (Ferrier etal., 1990). Estimates of wet and bulk deposition
should thus be modified to account for direct impaction in upland areas.
2.5.4 Monitoring and mapping deposition patterns
Measuring dry deposition presents particular difficulties as it tends to be governed by surface
properties and a wide variety of measurement techniques have been developed for this
purpose (Ross and Lindberg, 1994). Dry deposition rates of SOg onto vegetation and soil
for different surfaces throughout Great Britain have been calculated (Garland, 1978; Fowler
and Unsworth, 1979; Fowler and Cape, 1985). Rates are calculated from the product of
near surface concentration and an appropriate deposition velocity. This is inversely related
to the distance from the source. As a consequence, dry deposition tends to be greatest near
major emission source areas and contributes more than 75% of total deposition in Southern
and Eastern England (Cottrill et al., 1986). Estimated annual inputs of acidity from dry
deposition of SOg have been mapped showing that dry deposition in the industrial Midlands
and north of England is much greater than wet deposition while the converse is true in
western Wales and north Scotland (Fowler and Cape, 1985).
Wet deposition can be measured relatively easily by collecting precipitation and multiplying
the amount by solute concentrations. This precipitation weighting technique enables spatial
and temporal patterns to be identified. In spatial terms two aspects of wet deposition require
consideration, the concentration in precipitation of acidifying compounds and the amount of
acidity actually deposited (Irwin and Williams, 1988). UK Maps showing the concentration
of precipitation weighted non-marine sulphate and the amount deposited illustrate the
34
difference (Cottrill et al., 1987). The greatest concentrations are found in the east of Britain
where rainfall levels are lower while deposition is much greater in areas of higher rainfall in
North West England and North Wales. The relative contributions of HgSO and HNO3 in the
UK have been estimated as 71% and 29% respectively (Fowler etal., 1982) although the
latter is becoming increasingly important both in absolute terms (Skeffington and Wilson,
1988) and relative to the former (Galloway and Likens 1981, Rodhe and Rood, 1986). The
relative contribution of each to soil and water acidification is less easy to quantify due to the
mitigating effects that ecosystem interactions have on N species (Sutton and Fowler, 1992).
The concentration of acidic species in precipitation also exhibits seasonal variation with non
marine sulphate and nitrate maxima generally occurring in the spring or early summer (Irwin
and Williams, 1981). Variations in composition can also occur between and within
precipitation events (Coscio etal., 1982). At one site in Eastern England 30% of the annual
sulphate deposition occurred in five days (UKAWRG, 1986) while in Wales 30% of deposited
acidity falls on less than 5% of wet days (Reynolds, 1987). The implications for ecosystem
response of pulsed deposition episodes are discussed below.
In general terms non marine sulphate (/.e that not derived from sea spray) and nitrate have
fairly similar spatial patterns with lower concentrations in the north and west while those in
the East Midlands and East Anglia are up to a factor of 10 greater (Campbell et al., 1987).
For dry deposition, UK maps are not based on a monitoring network because of the
difficulties involved in obtaining accurate measurements. Estimates are based on semi-
empirical mathematical models which incorporate transport, transformation and removal
processes (Barrett and InArin, 1983). Maps are produced on a 20km^ grid basis using the
proportions of different land types in each square (UKRGAR, 1990). On a European scale
wet deposition maps are based on long-range transport modelling (van Leeuwen et al.,
35
1995). National maps are based more on measurement networks. In the UK, wet deposition
maps for S and N are based on a network of 38 monitoring sites together with the UK
Meteorological Office precipitation measurement network (Fowler et al., 1994). Maps have
been produced which incorporate the effects of orographic enhancement (UKRGAR, 1990;
Dore et a!., 1992). A modelling approach to deposition mapping has also been developed.
Initially, the Harwell Trajectory Model (HIM) coupled SOg, NO , NHg and HCI with simple
meteorology data (DenA/ent et a!., 1988; Metcalfe et a!., 1989). The Hull Acid Rain Model
(HARM) refines this approach although it presently concentrates on modelling 8 deposition
(both at current emission levels and under future emission scenarios). Using data on
emissions, rainfall, windspeed, trajectories, dry deposition and wet removal the HARM model
produces similar deposition patterns as those using measured data (Metcalfe and Whyatt,
1994)
It is has been shown that there is considerable variation in deposition levels onto different
landscape features and at different elevations. This variation is not fully incorporated into
maps at 2 0 km scale and precludes the use of these maps for identifying deposition at the
catchment scale. This has important implications for the application of the critical loads
approach at this scale where it is necessary to compare the sensitivity of the surface waters
for a specific catchment with the actual deposition loading to identify where critical load
exceedance may occur (Erisman, et a!., 1995). Acid loading onto individual catchments is
dependent on altitude, slope, aspect, vegetation cover and location, factors which can vary
substantially from catchment to catchment, even at a local scale (Ross and Lindberg, 1994).
The significance of these uncertainties on the development and application of a catchment
scale predictive model are discussed further in Chapters 3 and 8 .
2.6 The links between deposition and surface water acidification
36
Although the link between elevated levels of anthropogenically derived acid deposition and
the acidification of poorly buffered soils (Reuss and Johnson, 1986; Tamm and Hallbacken,
1988; Reuss and Walthall, 1990) and freshwaters (Oden, 1968, Battarbee et a i, 1985,
Henriksen et ai, 1988) is now almost universally accepted, it has been the subject of some
debate. The occurrence of acid waters in areas of acid soils has been seen as reason to
refute the acid deposition explanation in favour of one based solely on changes in the
terrestrial ecosystem (Rosenqvist, 1978; Krug and Frink, 1983). However there is strong
empirical evidence linking acid deposition to freshwater acidification (Reuss et a i, 1987).
The use of palaeolimnological techniques has been particularly prominent in this respect
(Battarbee, 1990; Battarbee et al., 1988, Patrick and Stevenson 1990, Fritz et a i, 1990).
This is corroborated by the results of dynamic modelling used to reconstruct historical trends
in acidification (Whitehead et a i, 1990). The role of organic acids, occurring naturally in the
soil, as precursors has also been recognised (Seip et ai, 1990).
On the basis of these studies it is assumed here that recent (post 1800 A.D.) surface water
acidification is primarily a result of acid deposition. Whether deposition leads to acidified
waters in individual catchments will depend on the level of buffering within the catchment
{ie., the catchment sensitivity). The attributes which are likely to determine sensitivity relate
to the soil, geology, hydrology and vegetation characterising the catchment.
2.7 Catchment sensitivity
2.7.1 Geology and soils
2.7.1.1 Introduction
An understanding of the consequences of elevated levels of acid deposition onto the soil
37
requires a recognition of the natural processes of soil development and acidification. The
chemistry of the soil determines the chemistry of the soil solution and thus the chemical
composition of surface water. The effect of anthropogenically derived acid precipitation on
the soil system and its relationship with vegetation is addressed here. Emphasis is placed
on soil as a buffer for acidic input, particularly the complex interrelationships between soil,
geology and surface water.
2.7.1.2 Fundamental concepts - soil acidification
The response of soil to increased inputs of acidic species is determined by soil chemistry.
Of crucial importance in this respect is the cation-exchange complex. This comprises
negative charges on clay minerals or soil organic matter (Reuss and Johnson, 1986). The
negatively charged exchange complex is dominated by base cations in alkaline or neutral
soils, aluminium species in acid mineral soils and in acid organic soils. Soil acidity is
therefore determined by the relative amounts of base cations and acid aluminium species
on the exchange complex. Acidification can occur when the number of negative charges
increases relative to base cations. This may result from an increased organic matter
accumulation or clay formation, or the removal of base cations by leaching. Conversely, an
increase in base cations relative to negative charges will increase alkalinity. Base cations
may be added via atmospheric deposition or from the weathering of soil minerals and a
reduction in negative charges can result, for example, from biomass burning.
Soil acidification can be quantified in a number of ways although it cannot be measured
using any single index (Reuss and Johnson, 1986). Soil pH can be used to define
acidification status although difficulties arise over the physical meaning of this concept and
its dynamic nature over time (Reynolds and Ormerod, 1993). An alternative measure is
provided by changes in base saturation (the extent to which the exchange complex is
38
occupied by exchangeable base cations) which decreases with increasing acidification.
Changes in capacity can also be used, which, in terms of acidity, generally refers either to
the storage of protons, AP" or base cations on the ion-exchange complex (or in weatherable
minerals). Within this context the deposition of acidic species will result in an increase in
exchange acidity and a decrease in exchangeable bases. This is a consequence either of
increased H"" inputs or from increased exchangeable aluminium following the reaction of H'"
with soil minerals. Exchangeable bases are reduced by the replacement of base cations on
the exchange complex by aluminium species. These are then leached from the system with
the strong anions (SO/' and NOg') associated with acid deposition.
Thus, a reduction in exchangeable bases leads to soil acidification. There are two major
processes by which this can occur, the first of which is base cation uptake by biomass. Base
cations are exchanged for H'' ions released from the plant roots. Plant growth is therefore
an acidifying process, although the base cations are returned to the soil following death and
decay. However harvesting vegetation results in a permanent depletion of base cations to
an extent determined by species, timing and type of harvesting (Hornung, 1985). The
second major process is leaching in soil solution. To maintain electric neutrality a balance
of positive and negative charges is required in solution. For base cation removal in solution
a mobile anion is required in association. The dominant anion in most neutral or slightly
acidic soil solutions is HCOg' (Reuss and Johnson, 1986). In acid precipitation however the
dominant strong acid anion accompanying H"' is often SO/'. Increasing concentrations of this
anion in soil solution can significantly increase the potential for base cation leaching.
These fundamental concepts are presented here so that the processes leading to an
increase in surface water acidification in response to acid precipitation are clarified. For a
more detailed review of soil and soil solution chemistry in response to acid deposition see,
for example, Cresser et al., (1986), Eriksson, (1988), Mulder and Cresser, (1994).
39
Water reaching surface waters must first pass through or over the soil. The interactions
between soil and water are then crucial in determining the chemical nature of the surface
waters. Precipitation that passes through the soil and reaches the surface water bodies as
baseflow or subsurface flow will be modified to a greater extent than surface run-off. From
the overview presented it is implicit that soils will tend to have a buffering effect on incoming
precipitation, reducing its acidity before it reaches the stream network. The extent of this
buffering capacity will depend on the nature of the soil and the underlying geology. The role
of NOg' is not considered here specifically due to the added complexity of N fluxes within
catchments, including uptake by vegetation, immobilisation and denitrification (de Vries and
LaTour, 1995). Similarly, although reduced N has the capacity to cause acidification in soils,
its role is not discussed in detail here (for a review see Galloway, 1995).
2.7.1.3 Soil as a buffer
The buffering capacity of a soil stems initially from mineral weathering of the parent material
as this is the primary source of cations. Thus the parent material (bedrock or subsequent
deposits) will be highly influential in the mineralogical composition of the soil.
Fundamentally, weathering is a combination of destruction and synthesis (Brady, 1990). As
parent materials are broken down physically the primary minerals are also subject to
chemical forces resulting in the synthesis of new or altered minerals. The most important
agents of chemical weathering are water and its dissolved salts and acids. Mineral
resynthesis is enhanced by the water driven processes of hydrolysis, hydration and
dissolution.
Changes in mineralogy are accompanied by reductions in particle size and the release of
soluble materials which are either leached out or combined into secondary minerals. The
40
secondary minerals can be broadly divided into silicate clays and very resistant iron and
aluminium oxide clays. The silicate minerals are still weatherable and through the action of
hydrolysis are able to assimilate ions from water percolating through the soil (Hornung
et al., 1990). As part of this process the H* ions are neutralised by OH ions. Thus the acidity
of incoming precipitation is neutralised by the presence of silicate minerals in the soil (Table
2.2, Equation 3). The solution of carbonate minerals derived from calcareous parent
materials has a similar effect (Equation 2.2) with COg ' neutralising the H"" ions. Weathering
rate will therefore be a major determinant of both terrestrial and aquatic ecosystems
acidification potential (Langan etal., 1994).
CaCOg (calcite) + H+ = Ca + + HCOg (2.2)
The neutralising capacity of the soil will thus be determined primarily by the amount of
carbonate and weatherable silicate. Related to these properties and equally important are
the cation exchange capacity and the base saturation.
The cation exchange capacity (CEC) of a soil can be defined as the sum of positive charges
of the adsorbed cations that a soil can adsorb at a specific pH (Foth, 1990). CEC is
determined by the relative amounts of different colloids (small mineral or organic particles)
in a particular soil and the CEC of each colloid present. Sandy soils will have lower CEC’s
than clay soils because the former are lower in clay and humus content. Soil colloids are
focal points for cation exchange reactions. Calcium and other metallic ions held on colloidal
surfaces may be exchanged with two H* ions in the soil solution (Brady, 1990). This will
have a neutralising effect on the solution. The reaction however is reversible and base
cations can be adsorbed onto colloidal surfaces instead of H" . Base saturation is simply a
measure of the extent to which the exchange sites of a soil's adsorption complex are
occupied by exchangeable base cations. It is expressed as a percentage of the total cation
41
exchange capacity. Once all exchange sites have been taken up by hydrogen Ions there Is
no longer any scope for base cation exchange. entering the soil will remain In solution
and thus contribute to surface water acidification. Where the pH of the solution surrounding
the mineral complex Is less than c.5.5, aluminium Is mobilised (Bache, 1986). As soils and
soil solution become more acid as a result of Increased acid loadings, conditions become
more favourable for aluminium mobilization. The Implications of Increased levels of dissolved
aluminium, which Is toxic to a variety of plant and animal species, are discussed below.
Sulphate adsorption capacity Is also an Important property. Soils are able to adsorb anions
such as s o / through fixing by aluminium or Iron oxides or hydroxides. The acidity of the
soil solution Is countered by the removal of SO/'. Once the sulphate adsorption capacity Is
exceeded SO /' will remain In solution thus Increasing the acidity of the drainage water
(UKAWRG, 1986).
The major neutralising reactions are summarised In Table 2.2. Generally, soils containing
free carbonates and high levels of weatherable silicate minerals or high base saturation will
tend to neutralise Incoming acid deposition. Conversely, In more acid soils with low base
saturation, fewer weatherable silicate minerals and no carbonate minerals, acid neutralising
capacity will be much reduced. Various combinations of these properties or surrogates
thereof have been used to classify soils on the basis of buffering capacity. These Include
texture (Lau and Malnwalring, 1985), CEC, and base status of subsoil horizons (Catt 1985,
Hornung et al., 1990b). Table 2.3 shows the classification used by Catt (1985) to produce
a map of England and Wales showing the distribution of soils with large, moderate and little
or no neutralising capacity. Soils were classed according to the base status of subsoil
horizons.
Carbonate and weatherable silicate mineral content, the cation exchange capacity and the
42
base saturation of a soil are determined by the nature of the parent material, age and
weathering and leaching history (Hornung etal., 1990b). Thus the underlying geology plays
an important role in catchment buffering capacity and this is discussed below.
Table 2.2 Summary of chemical processes in neutralisation of rainfall acidity (from UKAWRG. 1986)
1. Reaction of bicarbonate in water:
H+ + HCO,- ^ H P + CO;
2. Dissolution of carbonate minerals in soils and rocks:
2H+ + CaCO, H ;0 + CO; + Ca'+
Calcium leached from the system with strong acid anions
3. Hydrolysis of silicate minerals in soils and rocks (an example):
2H+ + 2KAlSi,0s + 9H;0 - 2K+ Al;(0H),Si;0, + 4Si(0H), oithoclase/feldspar kaolinite soluble silicate
Base cations are leached with bicarbonate or strong acid anions
4. Cation exchange with soils:
2H+ + Soil Ca Soil H; + Ca +
Calcium is leached with strong acid anions and the soil becomes more acidic. In mineral soils aluminium ions may eventually be released and occur as exchangeable or soluble cations.
Table 2.3 Soil buffering classes used by Catt (1985. in Hornung. 1990b) to map soil neutralising capacity in Wales. Terminology based on classification by Avery (1980)
Soils with little or no neutralising capacity
Soils with moderate neutralising capacity
Soils with large neutralising capacities
Holocene podzols; deeply weathered palaeo-argillic soils, thin rankers on non-calcareous Palaeozoic rocks; brown earths on siliceous gravels; gley soils on non-calcareous or pyritic clays and shales; raw bog, basin and blanket peat soils
Non-calcareous pelosols; brown earths on base-rich materials; gley soils on non-calcareous clay; fen peats
Little weathered soils on Holocene marine clays and calcareous sands; rendzinas; calcareous pelosols; brown calcareous earths; calcareous gleys formed during the Holocene or calcareous sediments.
43
2.7.1.4 Geology as a buffer
The main properties governing the neutralising capacity of geology are similar to those
controlling buffering in soils, principally, the content of carbonate and weatherable silicate
minerals. Solution and hydrolysis respectively, act on these minerals assimilating H" ions
(Hornung etal., 1990b). Where these minerals are in short supply it is possible for the pool
of available cations to be depleted faster than they are replenished by weathering (Paces,
1986). Other influences include contact area (greater in shattered, finely bedded or jointed
rocks) and residence time (longer residence times increase the likelihood that reactions will
reach equilibrium). Bedrock geology has been used to predict the occurrence of waters
sensitive to acidification (Likens at a/., 1979) and classifications of geology based on
buffering capacity have been produced similar to those for soils (Edmunds and Kinniburgh
1986). Table 2.4 shows a classification of the solid geology of Wales based on buffering
capacity (Hornung at a/., 1990b). The most sensitive surface waters tend to drain areas
overlying granitic or other highly siliceous geology with thin patchy soils (Reuss at a/., 1987).
Conversely, calcareous bedrock or unconsolidated calcareous sediments are able to buffer
even the most intense acid loading (Kramer, 1976).
Although soils and geology have been examined separately here, the neutralising capacities
of the two are obviously inextricably linked. However, it has been argued, that the dominant
control on surface water chemistry is bedrock chemistry (Miller and Drever, 1977; Norton,
1980) regardless of the nature of the soil cover (Bricker, 1986). Both form part of an
integrated system which is closely connected with a range of other catchment characteristics
including land use and hydrology. The contribution of these other factors to the chemistry
of surface waters is now considered.
44
Table 2.4 Buffering capacities of solid geology in Wales (from Hornung et al., 1990b)
Class Buffering capacity Rock types
1 Little or no buffering capacity
Granite and acid igneous rocks or metamorphic equivalents; granite gneisses, quartz sandstones and metamorphic equivalents; decalcified sandstones; most metasediments, including slates; non arkosic grits.
2 Low to medium buffering capacity
Sandstones, shales, conglomerates and metamorphic equivalents; coalmeasures, intermediate igneous rocks; high grade metamorphic felsic to intermediate volcanic rocks.
3 Medium to high buffering capacity
Slightly calcareous rocks, eg marlstones; basic and ultrabasic igneous rocks; Mesozoic mudstones; low grade intermediate to mafic volcanic rocks.
4 Infinite buffering capacity
Limestones, chalk, doloraitic limestones, highly fossiliferous sediments or metamorphic equivalents.
2.7.2 Land use and catchment management
It is apparent that the acid neutralising capacity in catchments is strongly influenced by soil
type and geology. Additionally, the response of surface water chemistry to different
vegetation types within catchments has been addressed by numerous studies (e.g. Johnson
and Swank, 1973; belong etal., 1990). In anthropogenically modified catchments in the UK
the influence of land management practices are particularly important. These can
significantly alter the chemistry of the soil as well as influencing deposition efficiency (Fowler
at a!., 1989). With regard to the acidification of surface waters two types of land use change
have major consequences, particularly in the more sensitive upland areas. These are conifer
afforestation and acid amelioration programmes based on the application of neutralising
agents (Reynolds and Ormerod, 1993).
2.7.2.1 Conifer afforestation
A number of authors have observed increased levels of acidity and toxic dissolved
aluminium in streams draining areas of conifer forest compared with those draining
45
equivalent moorland areas (Harriman and Morrison, 1982; Stoner et a!., 1984; Stoner and
Gee, 1985; Reynoids at al., 1986). This has also been supported by analysis of diatom
assemblages in the sedimentary records of lakes with and without coniferous forestry in the
catchment (Kreiser at a!., 1990). There are a number of mechanisms which may be
responsible for this and the impact of these will, to some extent, be dependent on the
species of tree (Hornung, 1985).
Increases in acid input can resuit from cloud droplet deposition. Compared to moorland
vegetation, the forest canopy provides more surfaces onto which acidic water droplets can
impact. This causes increased concentrations of solutes in throughfali (Fowler at a!., 1989).
Solute concentrations will also be increased by evaporation in the canopy (Reynolds and
Ormerod, 1993). Concentrations in throughflow and stemfiow to the forest floor may exceed
that of rain input by a factor of five (Cape at a/., 1987) although it is aiso suggested that
trees can reduce the acidity of water passing through the canopy layer (Miller at a/., 1987).
The increased levels of évapotranspiration also affects the annual runoff and baseflow levels
of afforested streams relative to moorland streams. The proportion of well buffered baseflow
in the streamflow of afforested catchments is reduced thus iowering the pH (Bird at a/.,
1990a).
Prior to planting, ploughing and drainage can alter the chemistry of the soil and influence
the hydrological regime of the affected catchment (Waters and Jenkins, 1992). The soil
becomes drier following drainage. This increases organic matter decomposition and aeration,
releasing sulphate which, in turn, increases the level of soil water acidity (Bache, 1984). The
furrows from ploughing and drainage channels also influence soil water hydrology so that
acidic runoff enters the stream network, bypassing the buffering potential of the soil system,
although a number of management techniques have been identified which may overcome
this problem (Miller, 1985). This latter effect becomes subordinate to that of soil drying as
46
the forest matures (Bird et al., 1990a).
The planting of coniferous forests on uplands previously characterised by moorland
vegetation is likely to cause an increase in base cation uptake. Following harvesting the
base cations are lost to the system thus decreasing the base saturation of the soil although
forest management practices such as reducing the intensity of the harvest (Hornbeck, 1992)
and the addition of fertilizer (Kreutzer, 1988) may lessen the impact. An understanding of
the long term effects of clearfelling is inhibited by the practice of replanting felled areas
(Reynolds etal., 1995). Short term streamwater responses are driven by factors which may
exacerbate or ameliorate acidity. Harvesting may impact on flow pathways with more water
flowing through acid upper horizons, reducing the potential for buffering (Neal et al., 1992;
Reynolds et al., 1992). Elevated concentrations of NOg', Ca^ and labile monomeric
aluminium, with concomitant decreases in pH, have been observed for two years following
whole tree harvesting (Lawrence, 1988), a result of increased soil nitrification and reduced
uptake. Conversely canopy removal is likely to lead to a reduction in dry and cloudwater
deposition of 8 and N species (Fuller etal., 1987; Adamson and Hornung, 1990). Generally,
the acidifying potential of forest growth is particularly problematic in sensitive catchments
(Sverdrup and Warvfinge, 1990).
There is considerable debate as to whether observed increases in stream acidity in forested
catchments is due to forest growth or as a result of acid deposition (Billet et al., 1990,
Eriksson et al., 1992). Recent applications of the dynamic MAGIC model to forested sites
suggest that a combination of afforestation and high sulphate deposition on acid sensitive
soils is responsible for acidification of soils and surface waters (Cosby et al., 1990; Jenkins
etal., 1990a; Jenkins etal., 1994).
47
2.T.2.2 Upland agricultural improvement
A variety of techniques exist to improve pasture in upland areas depending on the type of
soil being treated (Newbold, 1985). These include stock control, the addition of artificial
fertiliser and reseeding with more productive grasses (Reynolds and Ormerod, 1993).
However, the application of lime is by far the most widespread management practice used
to overcome the natural acidity of many upland soils.
2.7.2.3. Catchment liming
The application of lime to sensitive upland soils has major consequences for the
exchangeable cation complex causing increases in calcium and decreases in aluminium
(Hornung et al., 1986). Soil pH also increases and with it the pH dependent CEC.
Liming causes changes in soil water chemistry and can alter the susceptibility of upland
catchments to acidification (Boon and Kay, 1990). Following the application of lime to the
experimental catchment at Llyn Brianne there was a rapid increase in pH, alkalinity and
calcium concentrations in surface horizons (Reynolds and Ormerod, 1993). Increases in
ammonium encourages nitrification and increased nitrate concentrations which, if leached,
can lead to acidification (Meiwes, 1995). If NOg' is leached with base cations the soil is more
susceptible. Surface waters are affected if NOg' is leached with H or Al^. These changes
in soil water chemistry are also reflected in improvements in the surface water, with
increases in pH, calcium and magnesium levels and reductions in aluminium concentration
in comparison with streams draining unimproved catchments (Hornung at a!., 1990c).
Consequently, agricultural liming has lately been used to ameliorate the effects of surface
water acidification (Crawshaw and Diamond, 1988). The benefits to surface waters of
catchment liming are dependent on the type of materials used, the rate (Gasser, 1985) and
48
method of application and the placement within the catchment (Hornung et al., 1990c).
Additionally the effectiveness in combatting episodic acidification needs to be considered
(Porcella etal., 1989).
The utility of this approach is still under review. Although pasture improvement has been
shown to reduce surface water acidity and increase base cation concentrations (Adams and
Evans, 1989), evidence suggests that the major hydrologically active areas in the catchment
should be specifically targeted for treatment (Waters et al., 1991). Deleterious effects on
flora (Mackenzie 1989; Larsson, 1995) and fauna (Mackenzie and Shore, 1989) following
liming have also been noted.
2.7.2A. Lake liming
Liming direct to lakes and streams is now widely practised, particularly in Sweden and
Norway, and a number of long term studies have been undertaken to assess its impact on
surface water acidity (Henriksen and Brodin, 1995; Svenson ef a/., 1995). Generally, these
programmes have enhanced water quality and increased species richness and diversity
(Appleberg, 1995). A variety of agents, doses and strategies are employed (Henriksen et
al., 1995). Problems can be encountered with direct applications to streams as the supply
of base cations from slowly dissolving lime may not adequately counteract acid pulses
(Milner and Varello, 1990). Lake liming is more widespread and less expensive than stream
liming (Svenson et al., 1995). The efficacy of this approach is strongly dependent on the
turnover time (Werritty and Maucotel, 1992).
2.7.3 Catchment morphology and hydrology
2.7.3.1 Introduction
49
The influence that geology, soil and land use have on surface water chemistry within a
catchment is transmitted via a hydrological network. This is defined as both the flow of water
within a catchment and the network of pathways along which the water flows. The
neutralising capacity of a catchment will be highly dependent on the interactions that occur
in these pathways (Harriman and Wells, 1985). The morphology of the catchment
determines the response to precipitation inputs. This section examines the role of
catchment morphology and hydrology in determining stream and lake chemistry.
2.7.S.2 Catchment morphology
The three most important morphological attributes affecting sensitivity are catchment area,
drainage density and topography.
Small headwater sub-catchments are generally more susceptible to acidification (Matschullat
et al., 1992) and streamflow pH usually increases downstream with increasing catchment
area. This is due mainly to the increasing proportion of baseflow. As this is derived from
groundwater it is likely to have undergone acid neutralising processes following contact with
the soil, bedrock and reworked deposits. Additionally larger catchments are more likely to
contain more base rich soils and geology, particularly at lower altitudes further downstream.
In upland regions headwater areas are often characterised by peat covered interfluves which
are prone to pipeflow and surface runoff of low pH (Bird at a!., 1990b).
Higher drainage densities and steeper topography enhance the speed and magnitude of
response to rainfall. Both reduce the amount of time before runoff enters the stream network
and, consequently, the time in contact with the potential buffering soil surfaces within the
catchment. Additionally, the flow paths will be conditioned by the topographic shape of the
catchment (Wolock, 1990)
50
2.7.S.2 Hydrological pathways
The nature of the runoff pathways in a catchment determine whether incoming acid
deposition will be neutralised by contact with exchange surfaces within the soil and
underlying geology (Neal et al., 1986; Lawrence at al., 1988; Sullivan at al., 1986). These
pathways have been classified in a number of ways. At a fundamental level shallow,
unbuffered and deep, well buffered pathways are separated (Christopherson at al., 1982).
At a more complex level runoff processes are divided into shallow throughflow (via
macropores), pipeflow, overland flow, deep throughflow at the soil/rock interface and
groundwater (Bird at a!., 1990b). These will vary within catchments according to the
antecedent moisture conditions (Bishop at a!., 1990a), topography (Billet and Cresser,1992),
soil types (Wheater at a!., 1990), vegetation and surficial deposits.
Deep throughflow and groundwater will tend to neutralise acid waters before they reach the
stream network. Shallow throughflow, pipeflow and overland flow are more likely to result
in elevated stream acidity as water following these pathways will have a reduced residence
time. Streamflow responses under these conditions, particularly where saturation overland
flow and pipeflow dominate, tend to be very flashy with high stormflow/baseflow ratios. This
leads to the occurrence of acid pulses during high magnitude rainfall events, or following
snowmelt (Stoddart, 1995) which has major implications for stream ecosystems (Townsend
at a!., 1990). It may only require one episode producing critical conditions to wipe out a fish
population (Henriksen at a!., 1986). At a more complex level, a three phase chemical
response to flow has been identified. Baseflow dominated low flow is characterised by high
alkalinity and pH. As flow increases a mixing of soil drainage waters occurs, reducing
alkalinity and pH, and at higher flows dominated by well mixed contributions chemistry
remains stable (Jenkins at a!., 1990b).
51
The hydrological pathways within a catchment are closely linked with temporal variability of
streamwater chemistry, particularly with regard to episodic acidity (Bishop et al., 1990b).
Temporal variability can be short term or seasonal. The former is associated with high
magnitude precipitation events or snowmelt (Semkin eta!., 1994). As runoff increases rapid
macropore flow increases (Wilson et a!., 1990) and the contributions from different soil
horizons change relative to each other (Mulder eta!., 1990). In the first instance, the contact
time with the cation exchange complex in the soil matrix is reduced and the chemistry of the
stream runoff is largely unchanged from that of the precipitation (Creasey et a!., 1986). In
the second case, flow pathways are concentrated in the organic upper layers of the soil, as
opposed to the mineral soils beneath, and are enriched in and (Neal et a!., 1986).
The contribution of organic acids here may be as important as the lack of buffering (Seip et
a!., 1990). In both situations the buffering capabilities of the catchment are effectively
bypassed and, as a consequence, streamwater pH, alkalinity and base cation concentration
are reduced. However, where antecedent conditions are favourable, Ca^* concentration in
streamwater can initially increase as the products of accumulated weathering are flushed
from the soil (see Muscatt et al., 1990).
At a seasonal scale, chemical variation in surface waters is primarily determined by climatic
and biotic factors (Semkin et al., 1994). These include evaporation and precipitation levels
in the first case and nutrient uptake, mineralization and production of organic acids in the
second.
2.8 An integrated approach : catchment characteristics as predictors of surface water chemistry
This chapter has examined the various factors influencing the acidification of surface waters
on an individual basis. Surface water chemistry is dependent on a wide range of
environmental factors. These do not act independently but form part of a highly integrated
52
system which can exacerbate or ameliorate the acidity of surface waters. Included in this
system is the capacity of the freshwaters themselves to neutralise acidity through alkalinity
or ANC which enables incoming to be absorbed by proton acceptors. Sediments in lakes
can also neutralise acidity in overlying water through the release of calcium by desorption
and the reduction and storage of SO4 (Norton et al., 1990). However this chapter has
demonstrated the overriding importance of catchment processes and attributes. Many
studies have highiighted the link between soil acidity and acid waters (e.g. Reuss and
Johnson, 1986; Rosenqvist, 1990; Sverdrup at a!., 1992) and the controls that geology
exercises on surface water quality (Kramer, 1976; Norton, 1980; Creasey et a!., 1986).
Reference has already been made to attempts at using these relationships to identify where
acid waters may occur using various environmental parameters at regional and national
levels (Edmunds and Kinniburgh, 1986; UKAWRG 1988; Langan and Wilson 1992). At the
catchment scale, other studies have related surface water chemistry to soil (Rees et a!.,
1989) geology (Duarte and Kalff, 1989) and land use (Hunsaker ef a/., 1991) separately, or
using an integrated approach (Lynch and Dise, 1985).
The work discussed above has been undertaken at a variety of spatial scales with varying
data resolutions and the response of surface waters has been characterised using a variety
of measurement techniques. The regional studies employ national datasets relating,
primarily, to soil, geology and land use, to produce maps of surface water sensitivity to
acidification (e.g. Ullyet et a!., 1994; Hornung eta!., 1995b). At this regional scale geology
and soils are the major determinants of surface water chemistry (Likens eta!., 1979; Norton,
1980). However, at increasing spatial resolutions, factors relating to vegetation, hydrology
and land use become important (Hornung eta!., 1990a). Furthermore, at higher resolutions
there is some debate as to whether catchment attributes over an entire drainage basin are
less reliable as predictors of water quality than attributes close to the stream (Lynch and
53
Dise, 1985; Osborne and Wiley 1988; Hunsaker etal., 1991; Billet and Cresser, 1992). The
important issue here concerns the difference in accuracy of predictions based on very high
resolution data at intensely monitored catchments compared to more readily available data
at lower resolution.
2.9 Summary
The relationships between catchment characteristics and surface water chemistry have been
examined. Those characteristics which are most important in determining freshwater
sensitivity are soil, geology, land use and the nature of the hydrological pathways in the
catchment. Previous attempts at using catchment characteristics to predict freshwater
sensitivity have been noted, both at a regional level and for individual catchments. The
objective of this thesis, as outlined in Chapter 1, is to produce a model which will allow
estimates of sensitivity to acidification to be made for individual catchments using nationally
available data. The next chapter reviews how the critical load approach can be used to
quantify sensitivity.
54
CHAPTER 3 : THE CRITICAL LOADS CONCEPT
3.1 Introduction
This chapter outlines the theoretical background, development and current applications of
the critical loads concept. Various methods of calculation are presented and the potential
for future development is discussed. The scope for developing the critical loads approach
at the catchment scale is considered in the light of current applications.
A definition of critical loads is presented in Chapter 1. The critical load for freshwaters is
deemed to have been exceeded if levels of sulphur and nitrogen deposition are greater than
the acid neutralising or buffering capacity of the contributing catchment, initiating changes
in the aquatic ecosystem. Therefore critical loads are an integrated measure of sensitivity
to acidification. Further critical values can be identified as systems acidify and the most
sensitive organisms decline. These can be used to set target loads defined by the
requirement to protect particular species. The dose-response relationship implicit in the
critical loads concept is illustrated by Figure 3.1. Critical loads for sulphur, and total acidity
have been calculated for a range of receptors including freshwaters, soils, vegetation and
materials (CLAG, 1994, Posch et al., 1995). This thesis is concerned with critical loads for
freshwaters.
3.2 Critical loads for freshwaters
There are two methods currently used to set critical loads nationally. The diatom model is
a dose-response model that predicts the first point of biological change in an aquatic
ecosystem from an empiricaliy derived relationship between S deposition (loading) and water
Ca " concentration (sensitivity) (Battarbee at a i, 1996). The Henriksen (or steady-state
55
water chemistry - SSWC) critical load is based on the assumption that the production of
base cations in a catchment is equal to, or greater than, the input of acidic ions, maintaining
acid neutralising capacity above a definable level and producing a positive critical load
(Henriksen et al., 1986). Both models assume steady state conditions {i.e. inputs are
balanced by output) and can therefore be applied using mean water chemistry (Harriman
et a!., 1995b). Both techniques rely on estimations of pre-acidification ionic concentrations
and are derived empirically. Diatom and SSWC critical loads maps for Great Britain were
initially produced for sulphur (S) (CLAG Freshwaters, 1994). The models used to calculate
S critical loads are described here to introduce the concepts underpinning their development.
Critical loads models incorporating nitrogen (N) are presented briefly, but more thorough
reviews are given by Kàmàri et al., (1992b), Henriksen et al., (1992), and Hornung et al.,
(1994).
Figure 3.1: Critical and target loads concept (after Battarbee et al., 1994). The critical load for a site is exceeded at (a): (b) and (c) are critical loads for specific species. A target load (T) can be chosen to protect selected species as acid deposition declines in the future. Full recovery is represented by point (a) on the 'future' curve.
Critical and Target Loads
CO
ÜCD
CD
10Ü’O)Ogm15ÜECD_cO
S decreasingS increasing
FUTUREPAST 1980
56
Two further approaches are currently being developed for national (mapping) application.
The First Order Acidity Balance (FAB) model (Posch, 1995) is a mass-balance approach that
enables prediction of maximum nitrate leaching given certain S and N deposition scenarios.
Dynamic models, previously used on a site specific basis (e.g. Jenkins et a i, 1988), are
being simplified to broaden their applicability (M.Renshaw pers. comm.) (see Section 3.2.4).
3.2.1 Diatom critical load
Battarbee et ai, (1996) present an empirical critical loads model calibrated using diatom-
based pH reconstructions from lake systems. The use of these reconstructions is based on
the sensitivity of these diatom flora to changes in the acidity of surface waters. Diatom
assemblages respond to increasing levels of acidity, tending to shift to more acidophilous
taxa. These changes are recorded in the lake sediments which can therefore be used to
identify the timing and magnitude of recent acidification (Battarbee, 1990). The change in
pH, as evidenced by the sedimentary record, can be considered as the critical load
exceedance point, or the first ecological response of the aquatic system to acidification
(Battarbee et al., 1992a). Based on the assumption that diatoms are among the most
sensitive indicators of acidification, the diatom model provides a baseline critical load value
(Battarbee et a!., 1996). This is site specific as it indicates the point of acidification at a
particular site regardless of other chemical thresholds (Battarbee eta!., 1993). This contrasts
with the SSWC model which, for mapping purposes, is constrained by the pre-selection of
an acid neutralising capacity (ANC) appropriate for a target organism (Harriman and
Christie, 1992).
Although the acid loading at the time of first diatom response is unknown, an empirical
model has been derived using the acidification status of lakes of varying sensitivities and
deposition regimes (Battarbee et a!., 1996). This is based on the assumption that
57
acidification is a function of sensitivity in relation to deposition and that this relationship is
constant between sites (Battarbee et al., 1996).
The diatom model uses ionic calcium concentration as a measure of sensitivity and mean
sulphur deposition for loading. Acidification is defined as a change in the sedimentary diatom
assemblages to a more acid tolerant flora. To calibrate the model for sulphur deposition, 41
sites were used (Battarbee eta!., 1994). The Ca:S ratio was derived for each of these and
logistic regression analysis was undertaken to determine the ratio that optimally separates
acidified from non-acidified sites by calculating the probability of acidification at different
loading ratios. The optimal ratio is where the probability of acidification is 50% and was
found to be 94:1. This is the 'critical ratio’ (CR) (Battarbee et a!., 1996) and it is used to
define critical load in the following way:
1. Pre-industrial Ca "" values (Ca%) are calculated using the Henriksen 'F' factor
(Henriksen eta!., 1986)
2. The critical load for S is calculated using the critical ratio as Ca% / CR
3. This is re-expressed as keq ha^ yr'\
The diatom model as described applies only to critical loads for S. To incorporate the effects
of N deposition and modify the model for total acidity, it has been recalibrated for total
effective acid deposition (Allott et a!., 1995a). The critical ratio for total acidity is 89:1 and
the critical load is calculated as;
CL = [Ca^*]„/89 (3.1)
Pre-industrial calcium levels are calculated as follows (Allott et a!., 1995a);
58
[Ca^lo = [C a n - Fcao ([S O /]/ + [N O ^ - [S O /l* - [NOJ,) (3.2)
where o signifies pre-industrial values
t signifies contemporary values
’ signifies the non-marine component
Fcao is the F factor for pre-industrial calcium.
A number of uncertainties have been identified which concern both the diatom model
specifically, and both empirical models in general.
1. It is not certain which is the most appropriate historical calcium value to use in the critical
load calculation. While Ca^\ is known, Ca% is dependent on the Henriksen F factor. This,
in turn, is based on a number of assumptions relating to historical conditions and catchment
characteristics (Henriksen, 1982; Brakke etal., 1990). The optimum calcium value would be
that at the point of acidification {i.e. >Ca%<Ca%). Attempts are currently being made to
assess the sensitivity of the diatom model to different calcium values (T.AIIott, pers. comm.).
2. The model has been calibrated using lake sites. It is assumed that it can be applied to
stream sites although, given the absence of diatom histories it is not possible to test this
(Battarbee eta!., 1996).
3. A problem with applying the diatom model for N critical loads is that, used in tandem with
deposition data, the calculated values only relate to contemporary N leaching. Consequently
it cannot be used to test future N deposition scenarios (Hornung et a!., 1995b). Although this
point relates more to the calculation of exceedances (see Section 3.3) it is included here for
comparison with other critical loads models.
59
In spite of these uncertainties the diatom model is robust and accurately discriminates
between acidified and non-acidified sites when validated against sites not used in the model
calibration (Battarbee et al., 1996). It remains the most appropriate model for setting
baseline critical loads that can be mapped nationally.
3.2.2 Henriksen (steady state water chemistry) Critical Load
The Henriksen or steady state water chemistry model for S (Henriksen et a!., 1986) is based
on the principle that excess base cation production should equal or be greater than acidic
anion input. It is calculated using
CL = ([BC]o* - [ANC„,,]).Q - [BC];.R (3.3)
where:
CL = critical load
= non-marine component
BCq = excess base cation concentration prior to acidification
ANC,|^;( = critical level appropriate to a target organism
Q = runoff
[BC]j = non-marine base cation concentration in precipitation
R = rainfall
BCq is calculated using
[BC]o* = [BCj; - F ([SO%]; - [SOMo* (3.4)
where:
[SO / ]/ = present day excess {i.e. non-marine) sulphate concentration
60
Equation 3.4 estimates the proportion of extra base cations leached from the non-marine
base cation complex by excess sulphate inputs using the ’F factor. This estimates the part
of the base cation flux resulting from soil acidification and is derived empirically from
historical data (Henriksen, 1995). It is calculated using Equation 3.6.
F = sin [90.([BCV / S)] (3.5)
where S is the base cation concentration at which F=1 and varies between 200-400|ieq/l.
The pre-acidification sulphate level is determined from
[S O \. = 15 + 0.16.[BC], (3.6)
based on data from pristine sites (Henriksen, 1979). Although the F-factor is not necessarily
important for calculating critical loads in pristine environments it is probably more useful in
areas with higher acid deposition levels (Henriksen 1995).
To determine a critical load the SSWC model is set at a fixed ANC value (Henriksen et al.,
1992). Therefore SSWC critical loads and exceedances relate to the deposition values
necessary to maintain water chemistry above the critical ANC threshold. The model can be
applied to set critical loads for individual species. Occurrence and survival of particular biota
are related to a range of chemical determinands (e.g. pH, Ca^ , AP ) (Harriman and Christie,
1992). These determinands are, in turn, correlated with ANC, and, in acid waters, species
occurrence is strongly related to ANC (Harriman et a!., 1995c). Species response to ANC
can be modelled (using logistic regression) to assess the probability of occurrence at
different ANC levels (Lein eta!., 1992; Juggins etal., 1995). An ANC limit of Opeq Y\ used
for mapping in the UK, indicates a 50% probability of damage to brown trout populations.
61
(Harriman et al., 1995b). In Norway an ANC of 20peq 1' is used and above this few fish
populations are damaged (Henriksen et a!., 1992). Thus critical loads calculations using
SSWC are dependent on the ANC,j j, set and the maps produced from these can be
interpreted in terms of species occurrence.
As with the diatom model, the SSWC model has been modified to incorporate N deposition
and, as a consequence, to calculate total acidity critical loads (Hettelingh etal., 1992a). The
same methodology is adopted as for S with the exception that present N leaching (N,each)
is included in critical load exceedance calculations (Henriksen and Posch, 1995). This is
discussed further in Section 3.3.
3.2.3 The First Order Acidity Balance (FAB) Model
The initial critical loads models were developed in response to a requirement for an effects
based approach to S deposition. More recently, critical loads applications have incorporated
the role of N deposition. In addition, emphasis is now moving away from the empirical
models towards approaches which incorporate a temporal component. Whereas the diatom
and SSWC critical loads models for total acidity relate to current NO3 ' leaching, the First
Order Acidity Balance (FAB) model (Kàmàri etal., 1992b; Downing et al., 1993; Henriksen
etal., 1993) enables maximum potential NO3 ' leaching for given S and N deposition values
to be predicted. This allows the model to be used to assess future deposition scenarios and
the consequences, in the future, of current deposition levels.
The FAB model is based on quantifying the charge balance within a catchment. The full
charge balance for a catchment containing a lake is expressed by Equation 3.7.
+ S,., = /N„ + (1 -/)(Ni + N J + ri'J,, + BC, - ANC, (3.7)
62
where N gp is N deposition
S ep is S deposition
is N uptake by vegetation
Nj is N immobilisation in soil
Npg is N denitrification in soil
Nret is in-lake N retention
Sret is in-lake S retention
BCj is leaching of base cations
ANC, is leaching of ANC
f is fraction of forested area
r is lakeicatchment ratio
Derivations and empirical values for these parameters are available, depending on the
nature of the catchment, from a variety of sources (see Posch, 1995 for a review).
The FAB model enables the determination of the 'critical loads function’ which separates
individual S and N deposition values which cause exceedance from those which do not
(Henriksen and Posch, 1995). This concept is discussed in more detail elsewhere (Posch
etal., 1993; Bull etal., 1995) but essentially, the critical loads function comprises an acidity
function and a nutrient N function and can be used to identify combinations of S and N
deposition reductions which will, in theory, provide protection for aquatic, and other,
ecosystems.
However, to calculate critical loads, the FAB model requires substantially more data than the
empirical models. As such, it is currently limited in its application by the number of sites
where such data are available. Attempts are being made, using published values for input
parameters, to apply FAB at a national (UK) scale (C.Curtis, pers. comm.). In addition,
63
freshwater critical loads for total acidity, calculated using the FAB model, are to be used in
forthcoming European mapping exercises (Bull, 1995).
3.2.4 Dynamic modelling
One of the major drawbacks limiting the steady-state critical loads models (e.g. the diatom
and SSWC models) is that they do not incorporate rate of change. This is particularly
disadvantageous when assessing future deposition scenarios in terms of ecosystem
recovery. It is not possible, for example, to consider hysteresis effects using steady-state
approaches. Additionally, in worst case scenarios, steady-state models cannot account for
the cumulative effect of acid deposition with regard to N breakthrough or base cation
depletion. Dynamic models such as the Model of Acidification of Groundwaters in
Catchments (MAGIC) (Cosby et al., 1985b; 1990) incorporate the dynamic response of
surface waters over time and, consequently, can be used to assess the impact of different
deposition scenarios on freshwaters. Typically, dynamic models incorporate cation exchange
processes, anion adsorption and the role of land management to examine catchment
response to acid loading within a time-dependent framework (e.g. Cosby et a!., 1995b;
Jenkins etal, 1990b).
Dynamic models, although initially developed for S deposition are now being calibrated for
N deposition (e.g. MAGIC with Aggregated Nitrogen Dynamics - MAGICwand) (Perrier etal.,
1995). However, the processes of N cycling within a catchment are more complex than
those involving S. MAGICwand requires parameterisation of nitrification, mineralisation,
fixation and denitrification dynamics. Additionally, N deposition comprises both oxidised and
reduced N and the relative contributions of these need to be quantified. Furthermore,
elevated N levels in catchments can result either in acidification or eutrophication of surface
waters. Consequently dynamic models can only be applied where detailed input data are
64
available for individual catchments. Work is under way to simplify the MAGIC model in an
attempt to produce national critical loads maps for sulphur (M.Renshaw, pers. comm.)
3.3. Critical load exceedances
Critical loads exceedance maps for S are produced by superimposing sulphur deposition
data onto the critical loads, the difference being the exceedance, which can be negative or
positive. Maps can also show estimated future exceedances, based on current critical loads,
using a variety of deposition scenarios (Metcalfe and Whyatt, 1994; Allott et al., 1995b;).
These calculation and mapping exercises for S are (fairly) straightforward. Calculation of
exceedances for total acidity {i.e. including N deposition) require more data primarily
because the sulphur models assume no S retention in the catchment.
For the diatom model, total acidity exceedance calculations require the fraction of deposited
N which is leached into the surface waters to be determined. This is achieved by
differencing the SO/VNOg' proportions in the surface water and S/N deposition for the site
(Equation 3.8). This provides a value for the effective acid deposition and is a combination
of the fraction of N deposition leading to acidification (a^) and S deposition (Allott et a!.,
1995a).
aN = S'dep:Nd,p/[SO/'];:[NOg], (3.8)
Equation 3.8 assumes equilibrium between S deposition and S O /’ in the surface water. It
does not apply to freshwaters where there are catchment inputs of N {i.e. agricultural
lowlands). Exceedances are calculated using Equation 3.10.
~ ^ ■ N( dep) (3-9)
65
For the SSWC model exceedance for total acidity is calculated using Equation 3.10.
^ d e p " N | e a c h j g p “ CL^q (3.1 0)
Losses of N through leaching are calculated using
N,each = Npgp - Ng (3.11)
where Ng represents all N sinks in the catchments.
Equation 3.11 shows that N input data are not required for the calculation of total acidity
exceedance as NO3 ' leaching is assumed to equal deposition minus N sinks.
The monitoring and mapping of S and N deposition is described in Chapter 2 which
addresses some of the uncertainties inherent in deposition mapping. These uncertainties
affect the calculation of critical loads exceedances. Although deposition data mapped at
2 0 kn f resolution incorporate altitude and land use components, the deposition at a particular
catchment may be considerably different from the value ascribed to the grid square within
which it is located. Nevertheless the main areas of mapped exceedances for S are those
where acidification has been noted previously, including north Wales, Cumbria and south
west Scotland (Allott et al., 1995b) where a combination of sensitive catchments and high
loadings increase vulnerability of ecosystems. Regional variations in exceedance of total
acidity critical loads are similar to those for 8 deposition. However, calculations of the
increase in exceedance from leached N deposition suggests that in areas such as Cumbria,
the Pennines and Galloway, N deposition alone can cause exceedance (Allott etal., 1995c).
3.4 Mapping freshwater critical loads in the UK
66
To examine the spatial variation of freshwater critical loads nationally, a series of maps were
produced by the Department of Environment Critical Load Advisory Group freshwaters sub
group (CLAG Freshwaters sub-group) (CLAG Freshwaters, 1995). These followed a national
water chemistry sampling programme. The sampling strategy was based on the UK national
grid ( 1 0 x 1 0 km) in regions of medium to high sensitivity with 2 0 x 2 0 km squares being used
in low or non-sensitive areas. In each square one site was chosen for sampling. The
selection criteria were based on geology, soil land-use, altitude and size, the objective being
to sample the site likely to be most sensitive to acidification. Validation of the sampling
strategy suggests that, for lakes, this criteria was generally met (Curtis etal., 1995). These
data, theoretically, provide a critical load at which all surface waters in the square are
protected. Only one sample from each site could be taken so standing water bodies rather
than streams were selected (Kreiser et a!., 1993) due to the inherently greater temporal
variation in streamwater chemistry. Ionic concentrations were determined enabling both a
SSWC and a diatom critical load to be calculated for each grid square. The SSWC critical
loads are for ANC 0, at which there is 50% probability of damage to brown trout populations
(Harriman etal., 1995b).
The provisional critical loads maps produced following this sampling programme use five
classes ranging from less than 0 . 2 to greater than 2 . 0 keq H'"ha'Vear\ each 1 0 km grid
square being ascribed a particular class (CLAG Freshwaters, 1995).
3.5 Problems with mapping resolutions
The critical loads approach outlined above is geared towards targeting emission control
strategies at international, national, and (to some extent) regional levels. Critical loads
maps, as they are currently produced, allow emission control strategies to be targeted at
those areas least able to neutralize acidifying compounds and at those point sources whose
67
emissions of sulphur and nitrogen compounds are deposited in these sensitive areas.
However, maps at a resolution of 10km are of limited use at the catchment scale which of
necessity, is the scale at which drainage basin management decisions (relating, for example,
to forestry and conservation) are made (Newson, 1991).
A key consideration with regard to scaling for current maps is that the critical load for each
square is based on a sample taken from a single point within that square. Although sites
were chosen on the basis of sensitivity so that the critical load for each square was
estimated to be the lowest critical load within that square, this fails to account for variations
of water chemistry (and consequently critical loads) within the square (Langan and Harriman,
1993). In catchment management terms this could be a problem both if the ascribed critical
load for a square is higher or lower than that dictated by the water chemistry within a
particular catchment. If, for example, the critical load for a catchment is 0.2keq H"" ha' yr'
it may be that the addition of a coniferous plantation may cause the critical load of that
catchment to be exceeded. Clearly, in this context, it will be important to know which
catchments within a particular area are most sensitive. These issues are also important for
critical loads applications for conservation purposes. For example, areas where dippers may
be at risk are likely to coincide with areas sensitive to acidification. ’Stock at risk’ inventories
of this kind undertaken by conservation bodies will be most effective if the unit of
measurement is the catchment rather than the 10km grid square. Management decisions
taken using 1 0 km resolution data will inevitably fail to account for the variations that will exist
at higher resolutions (Hettelingh et al., 1992b).
3.6 Catchment scale critical loads: a predictive model
68
To apply critical loads at a catchment scale it is necessary to either;
1 . take water samples from all sites of interest or,
2 . predict or estimate critical loads using secondary data sources.
Where an organisation requires critical loads data at a national level the sampling option is
likely to present logistical difficulties. Previous attempts to predict freshwater critical loads
using secondary data (e.g. Hall etal., 1995; Kernan, 1995) have not used catchment scale
data. The remainder of this thesis describes the development of a statistical model to predict
freshwater critical loads using higher resolution data than used previously. Such an
approach requires a balance between the use of data specific to individual catchments and
data which are more readily available from national datasets. In modelling terms this would
be somewhere between the data requirements of the empirical and dynamic critical loads
models. Chapter 4 describes a range of variables which are selected as potential critical
loads predictors, in the light of the catchment process review undertaken as part of the
previous chapter.
The development of the predictive model is undertaken within the following constraints;
1 . it will be calibrated to predict the diatom critical load as the intention is to predict the
baseline critical load rather than a critical load for individual species.
2. it will be calibrated to predict the diatom critical load for S only. At the time of analysis this
model is well established and robust (Battarbee et al., 1986). Critical load models for total
acidity are still being developed and modified (e.g. Harriman et a!., 1995d).
The potential for developing the predictive model to incorporate current and future
69
CHAPTER 4 : RESEARCH DESIGN AND METHODOLOGY
4.1 Introduction
This chapter describes the research design of the thesis. The methodology is presented
including reference to the sampling framework, analytical chemistry, secondary data sources,
statistical techniques and problems encountered during the research. The analytical
component of the thesis comprises two fundamental components. The first, a preliminary
scoping study (Phase 1 - see Chapter 5) which, using data from a variety of secondary
sources, seeks to assess the feasibility of producing a predictive critical loads model. The
second, building on the findings of the first, is concerned with the development and
validation of this model (Phase 2 - see Chapters 6 and 7). The structure of this chapter will
reflect the stages in this analysis. Each aspect of the methodology is reported within the
context of this two phase approach. Details of the methods used in the preliminary analysis
(Phase 1) and the catchment model calibration (Phase 2) are discussed separately.
4.2 Site selection
Preliminary analysis - Phase 1
A preliminary analysis was undertaken to examine whether national datasets can be used
to predict surface water critical load. This used data made available by the Department of
Environment (DoE) Critical Load Advisory Group (CLAG) Freshwaters sub-group (CLAG
Freshwaters, 1995). These data were used originally to present critical loads and critical
loads exceedance maps of sulphur, for Great Britain. This programme is discussed in more
detail in Chapter 3. Water chemistry for 1573 sites are contained in a database held at The
Environmental Change Research Centre (ECRC), University College London (UCL) and
71
these form the basis for Phase 1 site selection. CLAG sites where data were missing are
omitted from the analysis. This has resulted in a depletion of sites in central and southern
England where analyses for total organic carbon and aluminium were not routinely
undertaken. Additionally, a number of original CLAG sites were superseded by new sites
following a validation programme (Curtis etal., 1995). These sites were also omitted. Sites
in N.Ireland, Shetland, Orkney and the Isle of Man were excluded because of a lack of
appropriate secondary data (see below) at the time of analysis. The remaining 976 sites
comprised the Phase 1 analysis. Figure 4.1 shows the location of these sites throughout
Great Britain.
Model development - Phase 2
Water chemistry data for the Phase 2 analysis was collected during two field seasons, the
first during autumn 1993, the second during following year. The sampling strategy for the
second field season was based on different criteria than that used during the first. The
strategy for the 1994 fieldwork is detailed below. Analytical chemistry for the samples taken
during the 1993 fieldwork was subsequently found to be unreliable and these data are not
used during the phase 1 analysis. The collection of these samples and subsequent problems
with the analysis are described in Appendix 4.1.
Initially a random sampling strategy based on a gradient across critical load values was
adopted (see Appendix 4.1). However, it was subsequently decided that, rather than
sampling randomly according to the critical load value of an individual site, a strategy based
on a catchment sensitivity gradient would be more appropriate. This shifted the focus away
from a sampling strategy based on the variables being predicted towards one based on the
predictor variables. The purpose of this was to ensure that the model is applicable across
a range of catchment types which might not be the case if the sampling strategy was based
72
on water chemistry data. A second consideration relates to the nature and availability of soil
data. In Scotland, soils can be divided, at a broad scale, into soil associations. These are
based on the parent material from which the soil developed (Macauley Institute for Soil
Research, 1984). In England and Wales soils are progressively divided into major soil group,
soil sub-group and soil series (Avery, 1987). In terms of catchment characterisation therefore
the Scottish soil classification system is not compatible with that used in England and Wales.
Soil classification is discussed in greater detail below (Section 4.6.2.6) Furthermore, Scottish
soils are mapped at a much higher resolution thus constituting a superior dataset. It was
decided therefore to calibrate the model using only Scottish sites.
The first criteria was that the calibration dataset should comprise catchments which included
all the dominant Scottish Soil Associations. The derivation of these are discussed in more
detail below. Of 110 soil associations mapped in Scotland, 19 cover 80.6% of the total land
area with the remainder contributing less than 0.65% each (MLURI, 1984). Similarly, a range
of the dominant geological formations were included. A grid square approach was adopted
whereby 1 0 km grid squares are selected within which groups of sites to be sampled for
water chemistry are defined. This approach was used because the GIS techniques
employed (see below) require that sites be clustered in well defined mapping units and not
be widely dispersed as this reduces the number of maps required to characterise
catchments according to soil and geology. Having studied the geology and soil maps 16 grid
squares, comprising 80 sites were selected on the basis of these soil and geological criteria.
Within the squares the sites were selected subjectively to ensure:
1 . catchments were large enough to enable characterisation at a scale appropriate to the
catchment data used;
2 . that each catchment was wholly contained within the 1 0 km grid square;
74
3. that a range of catchment sizes were selected to establish whether there is a strong
relationship between this parameter and water chemistry;
4. that sampling sites were selected immediately adjacent to confluences or recognisable
map features (e.g. bridges) to facilitate catchment delineation.
To fulfil these criteria it was ultimately necessary to select stream sites only. Most lake sites
were in catchments too large or too small to justify inclusion. A complete list of the sites,
with grid references, is supplied in Appendix 4.2. Figure 4.2 shows the location of these
sites throughout Scotland.
4.3 Temporal variation in water chemistry
The development of a predictive model requires the characterisation of catchment attributes
and water chemistry. For the purposes of this analysis, catchment soils, geology and land
cover can be considered as comparatively stable attributes. The composition of the surface
water chemistry is considerably more dynamic. It is noted in Chapter 2 that water chemistry
can vary with flow regime. Higher flow levels resulting from precipitation events are
characterised by relatively low proportions of base flow (Creasey et al., 1986; Neal et al.,
1986). Water spending less time in the catchment will tend to be less well buffered. This has
implications for the calculation of critical loads. Peak flow will be characterised by lower
calcium concentrations and a lower critical load whereas at low flow the critical load is likely
to be overestimated (Kreiser et al., 1995). It is important therefore to ensure, wherever
possible, that the water sample is representative of the general conditions and the timing of
sampling is crucial in this respect.
A study of the Allt A’Mharcaidh catchment in the Western Cairngorms, Scotland, found that
75
little or no change in pH or alkalinity occurred for flows greater than 0.8 m® s'" (Harriman et
al., 1990). Above this, the minimum Ca "" value also shows little variability, a pattern which
applies generally to more sensitive sites (R.Harriman pers comm). Henriksen critical loads
calculated for inflows into Loch Dee, S.W. Scotland are shown to be comparatively similar,
at mean and high flow, the main difference being at low flow (Langan and Harriman, 1993).
Figure 4.3 shows calcium concentrations at minimum, mean and maximum flows calculated
from a number of Acid Waters Monitoring Network (AWMN) sites (Patrick eta!., 1991). This
was derived from chemical data supplied by the Institute of Hydrology (IH). Where there is
more than one Ca^ value for a particular flow level, the mean Ca^ concentration was
calculated. The data presented cover monitoring of water chemistry over the period 1988
and 1994 and relate to the 11 stream sites within the AWMN. Relationships between Ca "'
and flow are not straightforward. In some instances there is very little variation between the
three flow regimes (sites 2,12,14 and 17). At other sites the Ca^* concentration at high flow
is closer to Ca "" at mean flow than at low flow, echoing the critical loads work of Langan and
Harriman (1993). At two sites (19 and 20), peak flow is characterised by high Ca '' levels.
This may be due to factors specific to these sites (e.g. antecedent soil moisture with high
Ca "" levels, see Muscatt eta!., 1990). The relationship between Ca^ and flow is plotted for
individual sites in Appendix 4.3. There are some extreme flow events which are
characterised by high Ca '" levels (e.g. Bencrom River). However, most of the relationships
show that variation in Ca '*’ diminishes above a certain flow regime. A similar study of 20
AWMN sites found that the differences between mean and minimum critical load, for most
of the sites, were relatively small compared with the differences between mean and
maximum critical load (Harriman et a!., 1995b).
Overall these studies indicate that, although there are substantial variations in surface water
chemistry associated with flow conditions, samples taken during periods between mean and
peak flow are more likely to be representative of mean conditions than samples taken at low
77
flow. Consequently, to maximise the number of sites included in the calibration database,
it was decided to take a single spot water chemistry sample from each site. Additionally,
because the data examined here indicate that samples should be taken between mean and
Figure 4.3: Calcium concentrations at minimum, mean and maximum flow for 11 Acid Waters Monitoring Network sites
220
20 O --
180--
160-'
1 140+ '5 8 "2 120+
E 100--
u 80-'
6 0 " ■È- • a -
4 0 - f -" A "
20
O1 □
- I - - - - - - - - - - - - - - - - - - - - - - - - - 1- - - - - - - - - - - - - - - - - - - - - - - - - 1- - - - - - - - - - - - - - - - - - - - - - - - - 1- - - - - - - - - - - - - - - - - - - - - - - - - 1- - - - - - - - - - - - - - - - - - - - - - - - - 1—
12 13 14 17 18 19A W M N site
— 1—22
Low flow a Mean flow High flow
20
Site 2; Allt a’Mharcaidh, Cairngorms Site 3: Allt na Coire nan Con, N.W. Scotland Site 9; Dargall Lane, Galloway Site 12; River Etherow, South Pennines Site 13: Old Lodge, Ashdown Forest
Site 14: Narrator Brook, Dartmoor Site 17: Afon Hafren, Central Wales Site 18: Afon Gwy, Central Wales Site 19: Beaghs Bum, Northern Ireland Site 21: Bencrom river. Northern Ireland Site 22: Coneyglen Bum, Northern Ireland
and peak flow, fieldwork was undertaken during October and November to avoid the
minimum flow periods likely to be encountered in the summer months. Given the
78
geographical scope of the sampling programme, winter months were avoided because of
the reduced daylight available, the difficulties associated with accessing some of the more
remote sites and the fact that a number of sites may be characterised by lower flows as a
result of snow accumulation
4.4 Sampling techniques
Samples for Phase 1 analysis
Water sampling protocols for the CLAG programme are identical to those used in the Royal
Society Surface Water Acidification Programme (SWAP) (Stevenson etal., 1987). These are
also used during the Phase 2 analysis and are described more fully below. Samples were
taken over a period of two and a half years between spring 1990 and autumn 1992 (Kreiser
et al., 1995) with the bulk of the fieldwork undertaken during spring and autumn months in
an attempt to obtain representative water chemistry (see above).
Samples for Phase 2 analysis
Samples were collected in 500ml acid washed narrow mouth polyethylene bottles. During
sampling, the bottles were rinsed in the stream and, once the sample was taken, the lid was
sealed underwater to minimise the amount of air trapped. Samples were taken at a depth
of approximately 40cm and care was taken to avoid the collection of re-suspended
particulate matter. All samples were analysed at the Freshwaters Fisheries Laboratory (FFL),
Pitlochry and were delivered within 4 days of sampling. Although 80 sites were identified
in the sampling strategy access was denied at two as a result of deer stalking activities.
These were not replaced.
79
4.5 Analytical chemistry
Water samples used in Phase 1 analysis
The Phase 1 analysis uses samples taken and analysed as part of the CLAG sampling
programme (Kreiser et al., 1995). The analytical methods used by CLAG, together with
detection limits, are described in Patrick et al., (1991). These are similar to the methods
used during the Phase 2 analysis described below. All analyses were undertaken at FFL.
The following key chemistry determinands were extracted from the CLAG database for
inclusion in the Phase 1 analysis (units of measurement and abbreviations subsequently
used in this chapter are included); pH, alkalinity, (Aik), (peg l‘ ); conductivity, (Cond), (pScm'
); sodium, (Na^), (peg I ' ); potassium, (K""), (peg 1' ); magnesium, (Mg^^), (peg 1' ); calcium,
(Ca^l, (peg 1' ); chloride, (Cl), (peg 1' ); nitrate, (NO3 ) (peg 1' ) sulphate, (S O /l, (peg 1' );
total organic carbon, (TOC), (mg M); non labile aluminium, (AL-NL), (pg 1' ); labile aluminium
(AL-L), (pg M).
Two alkalinity values were determined at FFL using dual point titration. These were
phenolphthalein alkalinity (titrated to the endpoint of pH 8.3) and total alkalinity (titrated to
endpoint approximately pH 4.5) (Madëra et al., 1982). The latter value is used in the
analysis as this Is more appropriate for sensitive, low-alkalinity waters. The diatom and
Henriksen critical load value (DCL and HCL respectively) for each site were also calculated
(refer to Chapter 3 for methods).
Water samples for Phase 2 analysis
All analytical chemistry for Phase 2 analysis was undertaken at FFL. Alkalinity and pH were
determined using a Radiometer TTT85 titration system with remote reference electrode and
80
40cm KCI head. Alkalinity values were produced using dual end point titration, titrated to end
point pH 4.5. Conductivity was analysed using a Philips PW9256 digital conductivity meter.
Na"", K"", Ca^ , IVIg ", SO/', Cl' and NO ' were determined by ion chromatography (Dionex).
Three calibration standards were used together with an internal (quality control) standard
on which precision and accuracy are based. Soluble monomeric aluminium and non labile
monomeric aluminium AI were determined colorimetrically using Catechol Violet. Soluble
labile monomeric AI is derived by difference between these two determinands. Precision
and accuracy data for FFL methods are displayed in Table 4.1. Accuracy estimates are
provided by the mean value of population X{. The closer the mean is to the Nominal
(standard) value the greater the analytical accuracy. Precision can be gleaned from the
standard deviation and the lower this is the greater the precision (Golterman et al., 1978).
Detailed chemistry for each site sampled during Programme 1 is provided in Appendix 4.4.
Table 4.1: Precision, accuracv and detection limits of analvtical methods. Freshwater Fisheries Laboratory (FFL).
Parameter n Mean Nominal*Value
Standard deviation*
|ieq 1 ' pg 1.,
Detection limit"
pcq 1* pg 1*
pH 189 4.69 4.70 0.019
Aik 189 109.0 110.0 2.30 115.0
NO,- 73 31.5 32.0 0.77 11.0(as N) 0.76 IO(as N)
SO," 73 41.1 42.0 1.03 50.0 0.73 35
Cl 73 136.0 141.0 2.36 84.0 0.64 23
Na* 73 132.3 130.5 1.90 44.0 0.20 5
K* 73 38.7 39.4 0.90 35.0 0.20 8
Ca"* 73 152.4 150 3.87 77.0 0.20 5
Mg"* 73 127.4 123.8 3.29 40.0 0.20 3
NH,* 73 29.1 29.5 1.65 23.0(as N) 0.25 4.5(as N)
TOC 250 1.92 2.0 120.0 2(X)
A L-TM 250 170.6 171.0 3.5 5
1 Based on internal standards set in mid range of calibration curve.2 Set at twice the baseline noise giving <1(XX) area counts
81
4.6 Secondary data sources
For both Phase 1 and Phase 2 analyses a variety of secondary data sources are employed
to provide the ’explanatory’ catchment variables for modelling. Some of these data sources
are used for both sets of analyses. The fundamental difference is that, for the most part,
Phase 1 data relates to the 1 km square in which the sampled site is situated whereas in
Phase 2, data are more specific to the catchment within which the sampled site is located.
For Phase 2 analysis each catchment attribute variable relates to a definable catchment
area. However, in Phase 1, catchment specific data are not used. A catchment is defined
as the contributing area upstream of the sampling site. Secondary data for CLAG sampling
sites are not available in this form and are based on 1km resolution digital maps. This
means that the secondary variables do not relate directly to the contributing area but to the
1km grid square that contains the sample location. However, this was not felt to be
problematic at this exploratory stage given that the selection criteria for the CLAG water
sampling programme included the requirement that the most sensitive site within each
square be sampled, and this tended to preclude sites with large catchment areas.
Additionally, the secondary data do not, for example, relate directly to soils, geology or land
use but are used as surrogates for these variables in the absence of more readily available
data. Thus ordinal variables such as soil critical load and site sensitivity, because of the way
they are derived, are used as substitutes for soil type and geology. Similarly, land
classification and land cover data were used to assess the influence of land use. These are
all at 1 km resolution and relate to the 1 km square in which individual CLAG sampling sites
are situated. Every site, therefore, could now be described, using a class value according
to each of these criteria. The derivation of each of these variables is outlined below.
82
4.6.1 Phase 1 secondary data
To enable a preliminary exploration of the relationship between catchment properties and
water chemistry a number of categorical {i.e. based on classification) data variables were
obtained, in digital form, from the Institute of Terrestrial Ecology (ITE) at Monks Wood.
These are soil critical load, land classification, land cover and site sensitivity, each dataset
comprising a series of nominal or ordinal classes. These are introduced below. Additionally,
a number of variables held on the CLAG database were extracted.
4.6.1.1 Catchment properties from the CLAG database
In addition to water chemistry the CLAG database also provided data for each site relating
to key environmental variables. These comprised the following;
■ Total non-marine S deposition (Sdep) (1989-92) (keq H"" ha' year^)
■ Total atmospheric N deposition (Ndep) (1989-92) (keq ha^ year'')
■ Rainfall (1989-92) (Rain) (mm yr' )
■ Site altitude (Alts) (m)
Figures for sulphur (8 ) and nitrogen (N) deposition are based on data mapped at 20x20km
resolution. The derivation of these is reviewed in greater detail elsewhere (UKRGAR, 1990).
Values for each site (or 10x10km square) are based on the value for the 20km square in
which the site falls. Clearly there are problems inherent in relating atmospheric deposition
at this resolution to specific catchments and these are recognised by the CLAG Freshwater
sub group (Kreiser at a!., 1993; Hall at a!., 1995b). Additionally, distance from sea was
determined by inputting the grid references for each site into the ARCMNFO GIS system
(see below).
83
4.6.1.2 Soil Critical Loads
A provisional map of critical loads for acidity of soils for Great Britain was produced by
CLAG (Hornung, 1993). This is based on 1km squares from the UK 0 .8 National Grid and
1:250,000 soil maps. The latter were produced for England and Wales by the Soil Survey
and Land Research Centre (SSLRC) (Soil Survey of England and Wales, 1983) and for
Scotland by the Macauley Institute for Soil Research (1981). A critical load for total
deposited acidity was allocated to each square based on the critical load of the dominant
soil map unit. These in turn were based on classes defined at a workshop held at
Skokloster, where soil materials were divided into five classes based on the dominant
primary weatherabie minerals of the parent material (Nilsson and Grennfelt, 1988). The
amount of base cations produced by weathering of these minerais thus determines the
critical load. The classes were amended according to a series of criteria inciuding
precipitation, vegetation and soil texture, drainage and depth, and further modified to
incorporate the role of secondary minerals using a classification produced by Sverdrup and
Warfvinge (1988). These modifications are discussed in greater detail by Hornung et al.,
(1994). The soil critical loads bands and associated characteristics are shown in Table 4.2.
This includes the number of Phase 1 sites in each class.
However, organic soils, particularly ombrotrophic peat soils, cannot be classified on the basis
of parent material mineralogy because the base cation supply in the surface layers stems
from atmospheric input rather than from bedrock weathering (Smith eta!., 1993; Wilson at
a!., 1995). These were allocated to one of the Skokloster classes using an approach
deveioped at the University of Aberdeen (Cresser at a!., 1993). The critical load is based on
the H"" load (as HgSOJ which would lead to a pH reduction of >0.2 in the surface layers of
the soil. More recently, peat profiles have been examined to assess the chemical
mechanisms controiling the response of organic soils to H deposition (Wilson at a!., 1995).
84
Rather than using the ranges suggested at Skokloster, ITE use a single critical load, the
upper limit of each band, to facilitate mapping of exceedances.
The CLAG classification has also been modified for land use using a database developed
by ITE (Bunce and Heal, 1984). This was used to identify 1km squares where the dominant
land use was arable or improved pasture. The critical load of each square in such areas was
increased by one class, the assumption being that lime would be added to the soil to
maintain pH at appropriate levels for arable crops and grasslands. However, for this
preliminary analysis the land use modification was removed as this component is used as
an independent predictor variable. Unmodified data are now being used for critical loads
mapping purposes as they are more appropriate when considering impacts to natural
systems (J.Hall, pers. comm.).
soils (after Nilsson and Grennfelt. 1988 and Sverdrup and Warfvinge, 1988 - modified bvHornunq etal., 1994)
ClassMinerals controlling weathering
Parent Critical weathering material (kmol H* km' yr ‘)
Critical load (keq H* ha 'yr ')
Equiv. amount of S (kg y f')
No. of Phase 1 Sites
5 QuartzK-feldspar
GraniteQuartzite
<20 .1 <3 91
4 Muscovite Plagioclase Biotite (<5%)
GraniteGneiss
20-50 .5 3-8 462
3 BiotiteAmphibole (<5%)
GranodioriteGreywackeSchistGabbro
50-100 1.0 8-16 119
2 PyroxeneEpidoteOlivine (<5%0)
GabbroBasalt
100-200 2.0 16-32 109
1 Carbonates LimestoneMarl
>200 4.0 >32 173
E 954*
This total excludes outliers removed during initial analysis (see Chapter 5)
85
4.6.1.3 Land classification data
The land classification database was developed by ITE Merlewood in 1977 and 32 classes
were subsequently defined and characterised (Bunce et al., 1981). The classification was
initially based on the identification of 281 attributes (related to information on climate,
topography, human artifacts and geology). Indicator species analysis (Hill etal., 1975) was
applied to produce 32 classes. These classes were updated using automated data capture
methods and the classification now covers all the 1 km squares in Great Britain. A description
of the classification system is provided by Bunce et a!., (1982) and is summarised in
Appendix 4.5a. The 32 classes have subsequently been aggregated using TWINSPAN
analysis (Hill, 1979). At the coarsest level four groups have been defined and these are
labelled arable, pastoral, marginal upland and upland (Bunce and Howard 1992). This and
two intermediate aggregations are described in Appendix 4.5b.
The data received from ITE {i.e. the 32 group classification) was manipulated in PARADOX
to reproduce the aggregations described above. This was done to facilitate the future
application of regression and ordination techniques by reducing the number of classes. The
coarsest division (into 4 classes) is used here. Table 4.3 shows the number of sites falling
into each of the land classification classes.
Table 4.3: Number of sites in each land classification class
Aggregate class Description No.of Phase 1 sites
1 Arable 1302 Pastoral 1493 Marginal upland 2814 Upland 393
Z 954*
’ This total excludes outliers removed during initial analysis (see Chapter 5)
86
4.6.1.4 Land cover data
A land cover dataset derived from satellite imagery is also available at ITE. This was
provided at 1 km^ resolution although it is currently available at 25m^. These data are derived
using Landsat TM images. Details of the classification system are provided by Fuller et al.,
(1994). A 25 class dataset covering Great Britain has been aggregated to a 17 class level
using cover per 1km grid square (Fuller and Groom, 1993a). Grid references for all the
CLAG water chemistry sites were matched by ITE (Monks Wood) with the 17 class dataset
providing a land cover class for each site based on the dominant class within the 1km
square containing the site. The 17 class and 25 class datasets are shown in Appendix 4.6.
As with the land classification data it was felt that the number of classes were too great for
valid application of the appropriate statistical techniques. Consequently the data were initially
aggregated into the 9 classes shown in Table 4.4. This table also indicates the number of
Phase 1 sites in each land cover class.
These data were subsequently aggregated into 6 classes (Table 4.5) to reduce the number
of variables and facilitate the multivariate analysis to follow. This also enables a comparison
to be made between different levels of aggregation in terms of their ability to explain water
chemistry variation (see Chapter 5).
Table 4.4: Aggregated land cover classification (9 classes)
87
Class Description No. of Phase 1 sites
1 Water & built/bare ground (inc beach) 212 Agricultural grass 2183 Arable 884 Deciduous woodland 405 Coniferous woodland 476 Lowland semi-natural grass/moor a) grass 357 Lowland semi-natural grass/moor b) dwarf shrub 1508 Upland semi-natural grass/moor a) grass 3239 Upland semi-natural grass/moor b) dwarf shrub 32
E 954'
' This total excludes outliers removed during initial analysis (see Chapter 5)
Table 4.5: Aggregated land cover classification (6 classes)
Class Description Aggregation from Table 4.4 No. of sites
1 Water & Built/bare ground (inc beach) 1 212 Agricultural grass and arable 2 & 3 3063 Deciduous woodland 4 404 Coniferous woodland 5 475 Lowland semi-natural grass/moor 6 & 7 1856 Upland semi-natural grass/moor 8 & 9 355
E 954*
* This total excludes outliers removed during initial analysis (see Chapter 5)
4.6.1.5 Site sensitivity
ITE (Monks Wood) also provided data relating to site sensitivity. These data were originally
used to produce a map of Great Britain illustrating the sensitivity of surface waters to
acidification (Hornung etal., 1995a). Five sensitivity classes were identified and these are
described in Table 4.6.
The classes were derived from a combination of soil, geology and land use information
88
(Hornung etal., 1995a). For soil, data were obtained from digital soil maps of England and
Wales (Soil Survey of England and Wales, 1983) and Scotland (Macauley Institute for Soil
Research, 1981). These data are generally available at Ikmf resolution (derived from
1:250,000 soil maps) with each square classified according to the dominant soil type within
that square. A buffering capacity/sensitivity classification was subsequently derived from the
mean base saturation or pH of the B horizon. This is discussed more fully in Section 4.6.2.6.
Each soil series was allocated to one of three sensitivity classes (Hornung at al., 1995a).
The geological data used to produce the sensitivity classes were based on a map produced
by Edmunds and Kinniburgh (1986). The map units on the 1:650 000 geological map of
Great Britain were allocated to one of four buffering capacity classes and subsequently used
to predict groundwater sensitivity to acidification. The classification is based on the
mineralogy and geochemistry of the dominant rock type. The map units define stratigraphie
units and do not incorporate any lithological variations which may have a significant bearing
on buffering capacity. Where the scale allowed the data used by Hornung at al., (1995a)
were modified to include these lithological anomalies.
To incorporate the influence of land use on the buffering capacity of soils the ITE Land
Classification Database (Bunce and Heal, 1984) was used to identify 1km squares where
arable or intensive grassland production was dominant. These were assumed to be subject
to liming. Where soils with high or medium sensitivity coincided with these types of land use
the square was moved to the low sensitivity class.
The land use modified soil map and the geological map were combined using a GIS overlay
technique. Twelve possible combinations of geological and soil sensitivities were generated
which were subsequently aggregated to form the five classes shown in Table 4.6. As with
the other categorical variables, the site sensitivity value for each (Phase 1) site relates to
89
the 1 km grid square containing the site, not the site itself.
Hornuna et al. . 1995a)
Class Soil/Geology sensitivity Likelihood of acidification No. of Phase 1 sites
1 soil low/geol. non sensitivegeol. low sensitivity geol. med. sensitivity geol. high sensitivity
Acidic waters will not occur 318
2 soil med./geol. non sensitivegeol. low sensitivity geol. med. sensitivity
soil high/geol. non-sensitive
Acid waters very unlikely 58
3 soil medVgeol. high sensitivity soil high/geol. low sensitivity
Acid waters unlikely 91
4 soil high/geol. med sensitivity Acid waters likely at very high flows
61
5 soil high/geol. high sensitivity Acid waters will occur at all flows
426
Z 954*
This total excludes outliers removed during initial analysis (see Chapter 5)
4.6.2 Phase 2 secondary data
4.6.2.1 Introduction
The catchments used for model calibration in Phase 2 were characterised according to a
number of criteria. These included soil, geology and land cover. Information regarding these
were derived from a variety of sources. Most of these involved the use of a Geographical
Information System (GIS), ARCMNFO (a vector based GIS from the Environmental Systems
Research Institute Inc., USA). A GIS was employed because of the spatial nature of the
analyses. GIS techniques allow different datasets relating to the same area to be matched
90
and analysed spatially and have been used previously to determine relationships between
water chemistry and catchment characteristics (e.g. Kalkhoff, 1993). In this instance
ARCMNFO enabled information from maps and satellite imagery to be converted to digital
form, thus facilitating the spatial aspects of catchment characterisation. This section deals
with the application of GIS techniques in this context. Additionally, issues relating to the
classification and aggregation of catchment data, particularly soil data, are discussed.
4.6.2.2 Catchment delineation
The catchment boundaries of each site were derived using contour information from
1:25,000 0.8 maps. These were then digitised using a Summagraphics Microgrid II series
digitising tablet and stored as INFO data files, or coverages. The information contained
within these files includes a series of coordinates defining a vector, which describes the
catchment boundary, and a number of reference (TIC) points which can be related to known
’real-world’ coordinates (in this instance the UK national grid) from the digitised map using
the TRANSFORM function in ARCMNFO. The PROJECT command then converts the spatial
units from digitising units to metres. By examining the polygon attribute file (PAT) for each
boundary coverage the area (in m ) for each catchment can be determined.
4.6.2.S Land use
The land use data employed during Phase 2 analysis are the same as those used in the
Phase 1. National data for land classification (Bunce and Heal, 1984) and land cover (Fuller
and Groom, 1993a) were made available by ITE (Merlewood) and ITE (Monks Wood),
respectively. The former are stored at 1 km^ resolution, the latter at 1 km and 25m^. Data are
stored as pixels {i.e. in grid form). The INFO files containing the catchment boundaries were
superimposed onto each of the ITE digital datasets in turn using a programme written in Arc
91
Macro Language (AML). This used the transformed TIC points for each coverage to locate
each catchment relative to the national databases. The AML programme also calculated the
area in each catchment covered by pixels representing each land class or land cover type.
These were then exported into a spreadsheet package (QUATTRO PRO). Subsequently the
percentage of each land classification/cover class type present in each catchment was
determined. The three datasets were aggregated as in Sections 4.6.1.3 and 4.6.1.4
producing the following 5 datasets;
■ Land cover (6 classes) at 25mf resolution
■ Land cover (9 classes) at 25nf resolution
■ Land cover (6 classes) at 1 km resolution
■ Land cover (9 classes) at 1 km resolution
■ Land classification (4 classes) at 1 km resolution
A list of the percentage of each land cover type (25m^ resolution, 6 class aggregation)
present in each catchment is provided in Appendix 4.7. This dataset was ultimately used in
the Phase 2 analysis (see Appendix 6.3a).
4.6.2.4 Solid geology
The geology for each of the 10km squares selected for sampling was digitised using
ARCMNFO and stored as a data coverage. Where possible 1:50,000 or 1:63,360 series
maps published by the British Geological Society, Nottingham, were used but in some
instances coverage at this resolution was not available and 1:625,000 maps were used. The
coverage for each grid square contains a series of polygons, each representing a geological
mapping unit. The coverages were transformed to UK national grid coordinates. Using the
legend accompanying the maps every polygon was labelled according to the map unit it
92
represented. This was subsequently modified to fit the classification system used in the
1:625,000 scale legend. Using the ARCMNFO INTERSECT function the catchment
boundaries were then overlaid onto the geological coverages enabling the area covered by
each map unit in each catchment to be determined by examining the PAT file for the
intersected coverage.
The legend for the 1:625,000 map for Northern Britain lists 99 map units. These needed to
be aggregated prior to statistical analysis. Furthermore the system used to aggregate the
rock types was required to be meaningful in terms of the purpose of the statistical model {i.e.
in this instance to relate geological attributes to surface water chemistry). A straightforward
hierarchical classification (i.e rock types divided into sedimentary, metamorphic and igneous
classes) is likely to have little bearing on the relationship between surface water chemistry
and catchment geology. Consequently the map units were aggregated according to the
system used by Kinniburgh and Edmunds (1984) who allocated each of the map units from
the 1:625,000 solid geology map of the UK to one of four categories based on
buffering capacity or impact on groundwaters. This was used to produce a map indicating
the regional susceptibility of UK groundwaters to acidic deposition. Table 4.7 describes the
nature of the buffering categories and the generic rock types ascribed to each. To calibrate
the predictive model each geological map unit was allocated to one of these buffering
capacities in consultation with M. Hodson at MLURI (Table 4.8). Thus each catchment is
characterised according to the percentage of each of the four buffering categories within its
boundaries. The percentages are listed for each site in Appendix 4.8.
93
Table 4.7: Categories adopted for classification of the solid geology map (1:625,000) of the UK according to sensitivity to acidification (after Kinniburgh and Edmunds. 1984)
Category Buffer capacity Rock type
Most areas susceptible to acidification, little or no buffer capacity, except where significant glacial drift.
Granite and acid igneous rock, most metasediments, grits, quartz sandstones and decalcified sandstones, some Quaternary sands/drift.
Many areas could be susceptible to acidification. Some buffer capacity due to traces of carbonate and mineral veining.
Intermediate igneous rocks, metasediments free of carbonates, impure sandstones and shales, coal measures.
Little general likelihood of acid susceptibility except very locally.
Basic and ultrabasic igneous rocks, calcareous sandstones, most drift and beach deposits, mudstones and marlstones.
No likelihood of susceptibility. Infinite buffering capacity.
Limestones, chalk, dolomitic limestones and sediments.
Table 4.8: Classification of individual map units of the solid geology map (1:625,000) of the UK according to sensitivity to acidification (after Kinniburgh and Edmunds, 1994)
MapUnil Type Epoch
SEDIMENTARY FORMATIONS
97-y Kimmcridgc Clay Upper Jurassic96 Comhrash Middle94-5 Great & inferior oolite93 Upper Lias Lower92 Middle Lias91 Lower Lias
90 Triassic mudstones NR Sandstone Permian &S9 Undifferentiated Permian & Triassic sandstones TriassicX7 Permian mudstonesS6 Magncsian limestoneS5 Permian basal breccias, sandstones & mudstonesS2-3 Westphalian ( "coal measures") Silesian CarboniferousXI Namurian ("millstone grit series")SO Toumasian & Vise an ("Carboniferous Limestone Di nanti an
scries")79 Basal conglomerate
78 Upper Old Red Sandstone Devonian77 Middle Old Red Sandstone75 Lower Old Red Sandstone
74 Ludlow Silurian73 Wenltxtk72 Llandovery
70-1 Ashgill & Caradoc Ordovician68 Llanvim & Arenig67 Durness Limestone
63 Serpuliie Grit & Fucoid beds Cambrian62 Pipe rock & basal quartzite
61 Sandstone and grit Torridonian
(continued overleaf)
94
(Table 4.8 cont.)
Unit Type Epoch Period Era Cla.ss
IGNEOUS ROCKS
59 Undifferentiated tuff Tertiary Extrusive 258 Rhyolite, trachyte & allied types 157 Basalt spilitc 556 Basalt Permian 555 Undifferentiated tuffs, mainly basaltic Carboniferous 554 Rhyiilite. trachyte & allied types 155 Basalt & spilite 5
52 Tuff (including ignimbrite) Devonian & 151 Rhyiilite, u-achyte & allied types OR Sandstone 150 Andesitic & basaltic lavas & tuffs, undifferentiated 249 Ba.salt & spilite 5
48 Tuff, undifferentiated, mainly andesitic Silurian & 246 Rhyolilic lava Ordovician 145 Andesitic tuff 244 Andesitic lava & tuff, undifferentiated 245 Basaltic tuff 542 Basalt, spilite, hyaloclastic and related tuffs 5
58 Agglomerate in rock Intrusive 257 Rhyolite, trachyte, felsite, evlans & allied types 156 Porphyrite, lamprophyre & allied types 255 Basalt, dolerite, camponite & allied types 554 Granite, syenite, granophyre & allied types 155 Diorite & allied intermediate types 252 Gabbro & allied types 551 Ultrabasic rock 5
Arca-s of intense granite veining 1
METAMORPHIC ROCKS
28 Foliated granite, syenite and allied types Metamorphosed igneous 127 Epidiorite, homblende-schist & allied types rocks in Moine & 226 Serpentine Dalradian 5
25 Limestone (Upper Dalradian) Dalradian 424 Limestone 425 Graphitic schist & shale 222 Black shale with chert (Upper Dalradian)21 Slate, phyllite & mica-schist (Upper Dalradian) 220 Slate, phyllite & mica-schist 219 Quartz, mica-schist, grit, slate & phyllite (UD) 118 Quartzosc mica-schist 117 Quartzite grit, interstrati lied quartzose-mica-schist 116 Boulder bed & conglomerate 515 Epidioritc-chlorite-schist. commonly homblendic-Grcen beds (UD)14 Epidiorite-chlorite-schist, commonly hornblcndic-Green beds15 Undifferentiated schist & gneiss of Shetland & Tyrone 1
12 Granitic gneiss Moine 111 Miea-schist, semi-pelitic schist & mica-schist10 Quartz-feldspar-granulite 19 Quantité 1X Undifferentiated 1
Granite migmatitc complex Lewisian complex 17 Gneissose granite, granite & pegmatite 16 Intermediate & basic rock5 Ultrabasic rock4 Anorthosite5 Marble2 Metasediments1 Undifferentiated gneiss 1
95
4.6.2.5 Drift deposits
Drift geology was digitised, where data was available, for each of the calibration 10 km
squares from 1:50,000 and 1:63,360 maps. Six drift types were identified from map legends
and these comprised boulder clay, alluvium, peat, moraine, sand and gravel and raised
beach deposits. Catchment boundaries were overlain onto the drift coverage and the
percentage of each drift type per catchment was determined. Alluvium and raised beach
deposits occurred rarely as the former tends to occupy the flood plains of major river
systems and the latter occurs primarily in low lying coastal areas. The map coverage for
superficial deposits is much poorer than that for solid geology and there were no readily
available drift data for numerous squares. Appendix 4.9 lists the percentage of each drift
type for those catchments which had map coverage.
4.6.2.6 Soil
Characterising the catchments used for Phase 2 calibration according to soil type presented
a number of difficulties. Soils are mapped at the national scale in Scotland at 1:250,000
resolution. The system used for the 1:250,000 map comprises a hierarchical classification
whereby soils are classed, initially, into a number of soil associations (Macauley Institute for
Soil Research, 1984). These are based on the parent materials underlying the soil. There
are 110 soil associations with coverage ranging from 16.22% of the land area of Scotland
(e.g. Arkaig) to less than 0.01% (e.g. Rackwick). These associations are further divided into
soil map units. The Arkaig soil association, for example, derived from schists, gneisses,
granulites and quartzites primarily of the Moine series, comprises 19 soil map units. These
are classed according to a combination of the component soils and landform (e.g. soil map
unit 26 of the Arkaig series comprises peaty podzols, peat and peaty gleys on hummocky
valley and slope moraines). The dominant soils separate one map unit from another together
96
with landform criteria such as slope and rockiness. The classification is thus based on
morphological features rather than chemical characteristics (Macauley Soil Research
Institute, 1984).
At a higher resolution soil maps based on field surveys are available for the agriculturally
productive areas of Scotland, primarily at a 1:63,360 scale. However, the Highlands and
Southern Uplands have only been provisionally mapped at 1:50,000 scale based on
reconnaissance surveys and the interpretation of air photographs. Together these maps
provide comprehensive national coverage. However, the classification systems adopted in
the legends for the two types of map are significantly different. Whereas the provisional
1:50,000 soil maps are based on the same soil association/map unit hierarchy as the
1:250,000 scale maps, the 1:63,000 maps for the surveyed areas are based on a soil
association/soil series hierarchy. Here, each association comprises a number of series, each
of which equates to a generic soil type, whereas most map units are soil complexes
characterised by two or more dominant soils. For example, the Tarves association can be
divided into a number of series including Thistlyhill, an imperfectly drained brown forest soil,
Tillypronie, a freely drained iron podzol and Pettymuck, a very poorly drained peaty gley.
Coverage at this resolution is divided fairly evenly between these two scales which, given
the geographical spread of the calibration sites, presents difficulties in terms of standardising
soil as a variable. Digital soil maps for each of the 16 10km grid square in the Phase 2
dataset were produced by digitising the appropriate hard copy soil maps at either the
1:50,000 or 1:63,360 scale. Each digitised polygon represented either a map unit or a soil
series. The problem therefore is to reconcile these two classifications and standardise the
approach.
Each polygon digitised at 1:50,000 scale was labelled with a soil map unit and those at the
97
1:63,600 scale with a soil series. It was then possible using a conversion link developed at
the Macauley Land Use Research Institute (MLURI) (S.Langan, pers. comm.) to relate the
soil series classification to the map unit classification. The index comprises a list of the soil
series which constitute each soil map unit. The series in each unit are ranked according to
dominance. It was possible therefore to link each soil series identified in the mapping
exercise to a soil map unit. Where a particular soil series was not the dominant type in any
one map unit it was labelled according to a map unit where it is the sub-dominant series.
This enabled a consistent approach to polygon labelling and the INFO files relating to each
coverage were amended accordingly. Once this had been achieved the catchment
boundaries were overlaid onto the soil coverages. The area covered by each map unit was
then determined.
There are 580 soil map units defined by the Soil Survey of Scotland. Clearly this is far too
many for relationships between soil and surface water chemistry to be established. A
number of different aggregations of this classification can be made. Each map unit can be
ascribed a soil association label thus reducing the number of classes fivefold. This in itself
is still too many. However, Langan and Wilson (1991) have developed a classification
whereby the sensitivity categories applied to the solid geology of Scotland are applied to soil
associations based on the parent material used to define them. This work was also
undertaken at the soil series level (Hornung et al., 1995a). Using mean pH or mean
percentage base saturation values derived from the Macauley Land Use Research Institute
(MLURI) soils database, each soil series was allocated to a sensitivity class based on the
soils acid neutralizing capacity (ANC). The classes were defined, in accordance with an
observed empirical reaction of soil response (Reuss and Johnson, 1986), and are shown in
Table 4.9. Thus it was possible, using the conversion index to produce a sensitivity class
for each catchment based on soil buffering capacity, and to determine the percentage cover
of each class in each catchment.
98
A second classification approach can be adopted (at this level) using the soil critical load
classes as an aggregation system. A soil critical load class is allocated to each soil map unit
Table 4.9: Soil series sensitivity classes (after Hornunq et al., 1995a)
Sensitivity pH Base saturation
High sensitivity (H) - low ANC <4.5 <20%Medium sensitivity (M) - medium ANC >4.5 < 5.5 >20% <60%Low sensitivity (L) - high ANC >5.5 >60%
based on the five Skokloster classes discussed in Section 4.6.1.2 (see Table 4.2). Using the
same approach as above, each soil map unit can be classed according to a sensitivity and
a soil critical load classification.
The sensitivity classification is based on concentration and base saturation data from
approximately 2000 soil profiles. Each series has values for each of these variables based
on varying numbers of observations. Using mean values it is possible to ascribe a H*
concentration value (H'’) or percentage base saturation (%BS) to each series. Conversion
to soil map units allows each polygon within each catchment to be quantified in these terms.
Using a weighted averaging approach based on the area occupied by each map unit it is
possible to ascribe or %BS values on a catchment basis producing a single figure for
each catchment. Thus the total area occupied by all occurrences of an individual soil map
unit are summed and multiplied by the value for H"" or %BS for the map unit. This is
repeated for each different soil map unit within the catchment. These products are then
summed and divided by the total area of the catchment. The approach is described, for H ,
by Equation 4.1. This is also adapted for %BS.
99
^ --------- (4.1)
SS = 1
where [H^c is a value for soil water concentration for a specific catchment
Si-...Sn are the soil map units in the catchment
Ag is the area of soil map unit s
H% is the hydrogen ion concentration value for soil map unit s
is used rather than pH units to simplify calculations.
A similar weighting averaging approach was applied to the lower limit of each critical load
class giving a quantitative soil critical load for each catchment. Catchment averaged
attributes are commonly used in dynamic models such as PROFILE (Sverdrup and
Warfvinge, 1988) and MAGIC (Cosby et al., 1995).
Further quantification is possible at the soil association level. Weathering rates have been
determined for each of the major soil associations in Scotland (Langan at al., 1995). Base
cation release was calculated from detailed soil mineralogy in tandem with soil chemistry
and physical attributes using the PROFILE model, a steady-state mass balance model used
to calculate weathering rates and critical loads for soils and soil water (Sverdrup and
Warfvinge, 1988). Three soil profiles were analysed from each of the major parent materials
from which Scottish soils are derived and the base cation release rate calculated for each.
100
The weathering rate at 50cm and the range throughout the profile are iisted in Table 4.10
together with the Skokloster critical loads for each association.
• ------- :— • . —-, -, T ----v - .-----— _ — . 1- - ------ -------------------------- ' ---------- --------- ^ -----—- — ,
A s so c ia tio n S k o k lo s te r C L P R O F IL E @ 5 0 c m
(k e q H* h a ' y r ‘)P R O F IL E ra n g e
(k e q H* h a ' y r ' )
Arkaig 0.2 - 0.5 0.3 0.2 - 0.3Balrownie 0.5 - 1.0 1.9 0.9 - 3.2Corby 0.2 - 0.5 0.3 0.2 - 0.3Countesswells 0.2 - 0.5 0.3 0.1 - 0.5Darleith 1.0-2 .0 0.7 0.4 - 0.8Dumhill <0.2 0.1 0.0 - 0.1Ettrick 0.5 - 0.1 0.8 0.4 - 0.8Foudland 0.2 - 0.5 0.3 0.2 - 0.8Hobkirk 0.5 - 1.0 0.4 0.1 - 0.5Insch 1.0 - 2.0 0.5 0.2 - 0.6Lochinver 0.2 - 0.5 0.2 0.1 - 0.4Rowanhill 0.5 - 1.0 0.5 0.2 - 0.6Sourhope 0.5 - 1.0 0.6 0.3 - 1.1Strichen 0.2 - 0.5 0.4 0.3 - 0.7Tarves 0.5 - 1.0 0.8 0.7 - 0.9Thurso 0.5 - 1.0 0.7 0.5 - 1.0Torridon 0.2 - 0.5 0.2 0.1 - 0.2
Using the weighted averaging approach employed above it is possible to determine a single
weathering rate for each of the calibration catchments.
A number of ways of characterising catchments according to soii have been identified
(Figure 4.4) In Chapter 6 these are analysed individually in an attempt to estabiish a single
soil variable which best explains variation in water chemistry. Data relating to the soil in each
catchment using these ciassification and quantification techniques is provided in Appendix
4.10.
4.6.2.7 Other attributes
Catchments were also characterised according to a number of other attributes, similar to
101
those used in the Phase 1 analysis. Sulphur and Nitrogen deposition and rainfall were
extracted from the CLAG critical loads database. Distance from sea was derived using GIS
Figure 4.4: Flow diagram Illustrating the soil variables available for characterising catchments
^ SOIL CRITICAL | I LOAD JWEATHERING ! ^
RATE !
SOIL CRITICAL LOAD, LOWER LIMIT
SENSITIVITY I
MLURICONVERSION
SOIL SERIES
SOILASSOCIATION
SOIL MAP UNIT
INDEX
Ti
J :pH SOIL CRITICAL
LOAD JSENSITIVITY 1 % BASE
SATURATION
LEGEND
Classification System
Aggregated nominal classification system
Quantitative variable
102
as above. Altitude for the site and the highest point in the catchment was noted from the
1:25,000 O.S maps. These are listed for each site in Appendix 4.11.
4.7 Statistical analysis
The underlying premise of the Phase 2 analysis is that the catchment attributes described
above can explain a significant proportion of the observed variation in water chemistry at the
sample sites. To predict the latter from derived catchment data, the most appropriate
approach is to use regression techniques. Here, a response variable is some function of one
or more explanatory or predictive variables. These can also be termed dependent and
independent variables. Simple linear regression relates a single dependent variable, Y, with
a single independent variable, X. Multiple regression is used when there is a single response
variable but two or more X variables (Manly, 1992).
The ionic concentrations and other chemical determinands in this context form the response
dataset. The catchment attributes are considered as explanatory or predictor variables. The
initial analyses are undertaken on a dataset comprising a large number of response
variables and a large number of explanatory variables. Where there is more than one
response variable the most appropriate techniques for data exploration are those which can
be described under the umbrella term of ordination.
4.7.1 Ordination
4.7.1.1 Introduction
103
Ordination is a term used to describe multivariate statistical techniques which arrange sites
or samples along axes based on the attributes of those sites (ter Braak 1987a). Although
these techniques are most commonly used within an ecological context with species
composition as response variables, in this study the response data comprises the ionic
composition of water samples. These techniques have been previously applied in this
context (e.g. Boyle et al., 1989; Marchetto at a!., 1995) while other uses outside ecology
have focused on, for example, geological data (Le Maitre, 1968; Miesch, 1980), soil
variables (Webster, 1977; Odeh at a!., 1991; Davis, 1986) and Scotch malt whiskies
(Lapointe and Legendre, 1994) as responses.
Ordinating combinations of sites and their chemistry and catchment data allows complex
relationships to be summarised statistically and graphically through the use of ordination
diagrams. Each site, chemical determinand and catchment attribute has a score, or loading,
for each axis, or component. These scores are used to position sites and variables relative
to one another in an ordination plot.
At the most fundamental level an ordination diagram arranges sites in two dimensional
space so that the proximity of two sites as represented in the diagram reflects, in this case,
the dissimilarity in the chemical composition of those sites. The interpretation of ordination
diagrams is covered in greater detail in Chapter 5. Additionally scores can be used to
quantify the relationship between variables and the ordination axes. Generally, the higher
the score (both positive and negative) the stronger the relationship between the variable and
the axis (Kent and Coker, 1992).
More detailed interpretations are possible depending on the type of ordination technique
employed. At the most fundamental level these can be divided between indirect and direct
gradient analysis and are summarised by ter Braak and Prentice (1988). Indirect gradient
104
analysis is used to examine variation within the chemistry data. Direct gradient analysis is
subsequently used to assess how much of the variation in the chemistry can be explained
by the catchment data.
4.7.1.2 Indirect gradient analysis
Within the context of both the Phase 1 and Phase 2 analysis, indirect gradient analysis
arranges sites solely on the basis of their chemical composition. This is done without
considering explicit independent variables which may affect the water chemistry at each site.
The ordination thus reflects the underlying chemical structure of the data. The assumption
here is that the ordination axes act as surrogates for latent explanatory variables acting upon
the chemistry data. The ordination can then be interpreted in the light of knowledge of the
other environmental factors at the sampling sites. In this way indirect ordination can be
likened to regression analysis, the difference being that the explanatory variable is replaced
by a theoretical latent variable represented by the first ordination axis (ter Braak and
Prentice, 1988). Along this axis the dispersion of the response variable is maximised.
Subsequent axes also maximise dispersion with the provision that they must be independent
of previous axes. The amount of variation in chemical composition explained by each
ordination axis is quantified in the analysis by the use of eigenvalues, which are equal to the
maximised dispersion of the response variable along that axis (ter Braak and Prentice,
1988).
Indirect gradient analysis comprises a number of different ordination techniques, the
applicability of each being dependent on the nature of the underlying response model. It is
possible to differentiate between a linear response model (where the response variables
increase or decrease linearly or monotonically in relation to changes in the latent explanatory
variables) and a unimodal response model (whereby responses are limited to specific
105
ranges of values of the latent variables) (ter Braak and Prentice, 1988). The appropriate
techniques are principal components analysis (RCA) and correspondence analysis (CA),
respectively. The latter is more applicable in an ecological context where species are likely
to respond unimodally to underlying environmental gradients.
Principal components can be equated with the correlation coefficients or eigenvectors of a
variance-covariance or a correlation matrix (Davis, 1986). PCA thus allows the identification
of the correlation structure or intercorrelations between the response variables, in this
instance the water chemistry determinands. PCA can also indicate potential areas of
collinearity. This enables a reduction in the number of variables being analysed (Johnston
1991). Correlation coefficients between the response variables and the ordination axes
which represent the latent explanatory variables are also produced. Each ordination axis is
described by an eigenvalue which indicates how much of the variation in the chemistry data
can be explained by a particular axis. This is one of the strengths of PCA. Additionally, PCA
of chemistry data can allow inferences to be made about the nature of the catchment in the
absence of direct data relating to catchment attributes. A wide variety of environmental
variables influence the surface water chemistry. These may be difficult to quantify and the
chemical response to particular variables may be uncertain. Therefore ionic concentrations
may provide more information about the catchment environment than a specified group of
independent catchment variables. PCA, in tandem with redundancy analysis (see below),
one of the direct ordination techniques, can show whether any important environmental
variables have been omitted from the analysis (ter Braak, 1987a).
4.7.1.3 Direct gradient analysis
Direct gradient analysis, or constrained ordination, requires the input of some explanatory
data. The ordination diagram in this case shows patterns of variation in the chemistry data
106
as well as the primary relationships between water chemistry and each catchment variable.
The axes, previously reflecting the influence of an unknown explanatory variable (or
variables), are now constrained to be linear combinations of the explicit explanatory
variables. This combines ordination and regression in one operation with each chemical
determinand acting as a response variable. Ordination scores for water chemistry
determinands can be predicted on the basis of the values for the explanatory catchment
attributes. As with indirect ordination a number of techniques are available depending on the
nature of the underlying response model. In linear cases redundancy analysis (RDA) (Van
den Wollenberg, 1977; Israels, 1992) is used while canonical correspondence analysis
(CCA) is more appropriate for a unimodal response model (ter Braak and Prentice, 1988).
These are the constrained equivalents of PCA and CA respectively (ter Braak, 1987a). The
use of redundancy analysis here is based on the premise that water chemistry
concentrations will have a linear, or at least monotonie, response surface with respect to the
environmental variables.
RDA can be viewed as a multivariate form of regression analysis. The response data are
thus modelled as a function of the ordination axes which are in turn a function of the
explanatory data (ter Braak, 1994). It performs a multiple regression of the site scores on
the explanatory variables and uses the fitted values of the regression as the new site scores.
These can then be plotted on an RDA biplot. The interpretation of these is discussed in
Chapter 5. The use of ordinations constrained by explanatory variables facilitates the
identification of a limited number of linear combinations of these variables that fit the
chemistry data best (ter Braak and Prentice, 1988). Within the context of the development
of a catchment critical loads model, this would allow identification of the most important
explanatory variables for inclusion in the model. Regression analysis can subsequently be
undertaken on these as this allows a greater accuracy of prediction and calibration than the
ordination techniques discussed.
107
There are a number of tools associated with the use of RDA. These include forward
selection of explanatory variables and significance testing by Monte Carlo permutation tests
(ter Braak, 1990). The former is similar to stepwise regression in that it identifies the
individual explanatory variables that explain most variation in the response data. The latter
allows the significance of individual variables and canonical axes to be assessed. Both these
techniques are covered in greater detail in Chapter 5.
4.7.1.4 Direct gradient analysis as a data reduction tool
The exploratory analysis undertaken during Phase 1 is primarily designed to establish
relationships between the water chemistry variables, and between water chemistry and
surrogate catchment attributes. During the model calibration stage the role of prediction is
emphasised. However, PCA and RDA are also used to examine the effect that using more
sophisticated catchment specific data has on the relationships identified in the Phase 1
analysis. Using this approach it is be possible to identify those variables which best explain
these relationships. In this context RDA will be used as a data reduction tool. This
reductionist approach follows two paths. Firstly, those variables that explain most of the
variation in base critical load are selected for direct input into a multiple regression analysis.
Second a number of composite variables can be created by combining the effects of several
related variables in a similar way as the soil sensitivity and soil critical load classifications
were derived. Soils, geology and land use might be combined, for example, to produce a
single buffering variable. Altitude, rainfall and deposition might be combined in a similar
fashion. The relative explanatory powers of these composite variables (independently and
is association with other composite variables) can be assessed using a technique known as
variance partitioning (Borcard et al., 1992).
108
4.7.2 Variance partitioning
This is a technique proposed by Borcard et ai, (1992), based on RDA or CCA, which allows
the partitioning of variance in a response dataset into different groups of explanatory
variables. The basis of this approach is that each individual independent variable can be
allocated to a ’primary’ class of variable. For example, 0kland and Eilertsen (1994)
examined variation in species data by defining i) environmentally driven variation, ii) spatially
driven variation iii) shared environment-spatial variation iv) unexplained variation. Similarly,
Borcard and Legendre (1994) created four composite explanatory variables, i) environment
at a local scale, ii) a spatiai environment component, iii) a purely spatial component and ii)
an undetermined component, to explain the abundance of orabatid mites {Acari Oribatei)
along the shore of a small lake. Using RDA with covariables {i.e.. those removed from the
analysis) it is possible to decompose the variation in water chemistry generally, or diatom
critical load specifically, into a number of primary components. In this way variance
partitioning will allow the explanatory power of these composite variables to be quantified
individually as well as calculating the covariation term {i.e.. explanation shared by
combinations of components) and the proportion of the variance remaining unexplained.
4.7.3 Multiple regression
Having examined the overall relationships between water chemistry and catchment data
attention is focused on the main thrust of the thesis, namely whether critical load can be
predicted by one or more catchment attributes.
Simple linear regression examines the relationship between a single dependent variable, Y,
and an independent variable, X, where the latter is thought to determine the former to some
extent. When X and Y are plotted against each other linear regression produces a line
109
through the series of datapoints in the scatter which minimises the sum of squares of errors,
or deviations, between the observed values of Y and those predicted by the line. This line
can be described by the equation:
V ' = a + p X + e (4.2)
where Y is the response variable, a is the intercept of the line with the y axis, (3 is the slope
of the line, Xis the explanatory variable and e is the error term. Where the response variable
is to be predicted by values of two or more explanatory variables multiple regression is used.
This is described by the equation:
+ 2^ 2 + ..... + + E (4.3)
where X , ....X„ are the explanatory variables. This equation finds the coefficient of all
the X values that minimises the error sum of squares (Manly, 1992). The partial regression
coefficients here will only be identical to coefficients from separate simple regressions when
the predictor variables are uncorrelated. The prior use of RDA, as well as facilitating a
reductionist approach, allows the identification of collinearities between environmental
variables. This eliminates the need for an ’extra sum of squares’ approach (Manly, 1992)
where variables X, to X„ and successive regressions are fitted, relating Y to X , Y to X and
Xg and so on until Y is related to all X variables. Variation in Y accounted for by X„ on top
of that accounted for by X, to X ., is given by the extra sum of squares accounted for by
adding X„ to the model.
Residual plots are use to assess whether the assumptions of the regression model are
fulfilled (Manly, 1992; Johnston, 1991). These are discussed further in Chapter 7. There
should be no pattern when residuals are plotted against Y estimates or X values. Tests for
110
randomness, constant variance and normality can be applied to the residuals to ensure
these criteria are met. Outliers can also be identified using residual plots and these can be
examined to assess whether the statistical relationships generated by the regression model
are being disproportionately influenced by individual observations.
4.8 Discussion
A number of issues arising from the research design and methodology need to be
addressed, particularly relating the adequacy of the data used to characterise the calibration
catchments. These are introduced here and discussed further in Chapter 8.
4.8.1 Land use data
Land cover and land classification data is only available nationally in raster (grid) format.
Land cover is available at 1km and 25m resolution. Ground based and aerial surveys have
been undertaken to validate the satellite derived land cover database. These show 75% to
95% accuracy depending on the spatial scale (Fuller and Groom, 1993b). At the 25m scale
certain cover types (e.g. suburban and arable) appear more likely to be classed erroneously
than others (0.Curtis, pers. comm; J.Hall, pers. comm.). However, despite these
uncertainties, the ITE land cover dataset represents the most accurate picture of land cover
presently available in Britain at this resolution.
Land classification data are only available at 1km resolution. Superimposing small
catchments onto such grid data could lead to a significant loss of information. This will be
particularly problematical at boundaries between different classes.
4.8.2 Geology data
111
It has to be accepted that geological maps at any resolution are only a guide as to the
conditions existing on the ground. This is particularly the case with geological boundaries.
The scale of geology maps is such that small outcrops of a rock with a high buffering
capacity may not be mapped. These may have a disproportionate ameliorating effect on
surface water acidity.
A further problem relates to the sensitivity classification proposed by Kinniburgh and
Edmunds (1986). This is based on the stratigraphy, or age, of the rock and not on the
lithology. As such, it allows for broad lithological differentiation but some facies changes may
not be mapped, a difficulty noted by the authors. Limestones and clays, for example, may
be present in poorly buffering formations.
Additionally, a classification of rocks on a scale from acidic to basic, although indicative of
the chemistry of the rock, might not necessarily reflect the chemistry of the run off water
(M.Clark, pers. comm). The movement of water through bedrock is important in this context.
Where fissure flow dominates over matrix flow the geochemistry of the fissure surface will
control buffering. Preferential flow along these pathways can remove the buffering cement
that has accrued (Kinniburgh and Edmunds, 1984).
4.8.3 Soil data
Mapped soil data provide the same difficulties of resolution as those encountered with
geology. The boundaries between soii map units are even less solid in reality and are
impossible to represent accurately with two-dimensional maps. The quantification of soil data
also presents difficulties. The pH and base saturation data for each soil map unit are based
on an insufficient number of samples at best and a single sample at worst. Clearly soils
within a catchment will not display such homogeneity. Additionally, Langan et al., (1995)
112
caution against the use of using mean weathering rate to characterise soil associations.
4.8.4 Deposition data
The uncertainties associated with using monitoring atmospheric input of N and S, for
individual catchments is discussed in detail in Chapters 2 and 3. These uncertainties also
apply to rainfall values although these are based on a much larger monitoring network.
Although a series of problems have been recognised it is likely that these will always be
encountered when attempting to use nationally available data at a local scale. However, the
development of a nationally applicable statistical model requires that the best avaiiable data
is used. In the absence of comprehensive and freely available data at high resolution, the
data used here offer the most pragmatic means of developing a predictive catchment critical
loads model.
113
CHAPTER 5 : PHASE 1 - PRELIMINARY STUDY OF MODEL FEASIBILITY
5.1 Introduction
This chapter presents a preliminary analysis of a water chemistry dataset covering the whole
of Great Britain together with a variety of secondary catchment (explanatory) variables
relating to the sites from which water samples were taken. These variables are described
in more detail in Chapter 4. At this stage most of the explanatory variables are surrogates
for catchment specific data and their use is an attempt to gauge the degree to which broad
scale mapping of freshwater critical loads might be achieved using readily available data
relating to catchment sensitivity. Similar approaches have been adopted using nationally
mapped data to estimate the likelihood of surface water acidity at a regional scale (Langan
and Wilson, 1992; Hornung etal., 1995a) while Hall etal., (1995a), examined the accuracy
with which national soil, geology and land use databases can predict surface water critical
load class. This analysis relates a suite of chemical determinands to catchment parameters
prior to focusing on the relationship of the latter to the diatom or baseline critical load. The
aim is to examine the relationship between these readily available catchment data and water
chemistry, and hence assess whether a closer examination of catchment/surface water
interactions, using more detailed data, might allow the development of a more accurate
predictive model. The results will also enable comparison between approaches using
different spatial scales.
The statistical techniques used in this section are introduced in Chapter 4. Initially, chemistry
(response) determinands and catchment (predictor/explanatory) attributes are analysed using
data for all sites extracted from the CLAG database (the ’full dataset’). These data do not
constitute a random sample of the population of UK freshwaters as a whole (Kreiser et a i,
1993), Nevertheless, given the geographical scope of the critical loads programme it is likely
114
that the data encompass the full spectrum of chemical characteristics to be found across the
UK. Subsequent analyses focuses on sites that are likely to be more susceptible to
acidification (the 'sensitive sub-set’) and a comparison is made between the two sets of
results. The analytical methodology is as follows:
■ Exploratory analysis of the water chemistry data for 954 CLAG sites,
■ Exploratory analysis of the surrogate catchment parameters,
■ Examination of the relationships between response and explanatory variables,
■ Examination of the relationships between diatom critical load and catchment
characteristics,
■ Examination of the relationships between diatom critical load and catchment
characteristics for a subset of the data comprising more sensitive sites selected on the basis
of current calcium concentration.
5.2 Analysis of the full dataset (954 CLAG sites)
5.2.1 Exploratory data analysis of response variables (water chemistry)
This section explores the nature of the response dataset. The interrelationships between the
water chemistry determinands for the 954 CLAG sites are quantified and the underlying
structure of the data is examined.
Prior to analysis, the response variables were transformed to facilitate the application of
parametric statistical techniques (Jager and Looman, 1987). Water chemistry data are often
characterised by a log-normal distribution (Ott, 1990). Consequently these were transformed
(except pH) to log o (x+c) where x is the value of the determinand and c is a constant added
to ensure that x+c is zero or positive. Following transformation the Z scores for each
115
observation (for each determinand) were calculated. Within the context of a population, Z
scores have zero mean and unit variance (Norcliffe, 1982). From this, the position of an
observation along a normal distribution can be expressed in terms of the number of standard
deviations from the mean. An arbitrary Z score value of 3.0 was used to determine potential
outliers. This identified sites where the value for a particular determinand is greater than
three standard deviations from the mean. Sites with higher scores were investigated to
assess whether they should be removed from the analysis. This process was repeated for
all determinands. A total of 22 outlier sites were identified across a range of determinands.
Subsequent examination of site description reports led to 19 of these being omitted from the
analysis leaving a total of 954 sites. Reasons for removal included quarrying and
interference from recreational and industrial sources. These influences were not deemed to
be related to the inherent sensitivity of the catchment and as such do not appear to be
’genuine members' (Barnett and Lewis, 1994) of the population.
5.2.1.1 Summary statistics
Summary statistics were calculated for untransformed water chemistry together with the
derived critical load values. These are shown in Table 5.1 which includes the unit of
measurement and an abbreviation which is used in all subsequent tables and diagrams.
Values for the mean, standard deviation (SD), minima and maxima are given. Transformed
data are presented in Appendix 5.1.
Table 5.1 illustrates the extended lengths of the chemical gradients across which the sites
are spread even after outliers have been eliminated. Sulphate and alkalinity have ranges of
6041 and 6415 peq 1-1 respectively (mean 303.28 and 550.42 peq M) while pH varies from
3.82 to 9.21 with a mean value of 6.53. The very large calcium gradient (8088 peq l'\ mean
714.44 peq 1' ) is reflected by the wide range of critical load values, particularly the diatom
116
Table 5.1: Summary statistics for untransformed chemistry/response variables (n = 954)
Mean SD Min. Max.
pH 6.53 1.10 3.82 9.21Alkalinity (|ieq 1‘) (Aik) 550.42 988.44 -210.00 6205.00Conductivity (jiScm’’) (Cond) 147.09 159.46 11.00 1191.00Sodium (fieq 1'*) (Na*) 443.94 418.39 43.00 4635.00Potassium (|xeq 1‘) (K*) 35.79 55.12 B.D 479.00Magnesium (jieq 1') (Mg^*) 240.47 351.10 9.00 4116.00Calcium (|ieq I'*) (Ca *) 714.44 1203.07 12.00 8110.00Chloride (^eq 1'‘) (Œ ) 556.03 539.97 38.00 5681.00Nitrate (p.eq 1'*) (NO,") 43.79 132.83 B.D 1489.00Sulphate (peq 1"‘) (SO ) 303.28 589.65 14.00 6055.00Total organic carbon (mg 1"‘) (TOC) 4.87 4.11 0.10 39.10Non labile aluminium (fig 1"‘) (Al-NL) 17.24 25.70 B.D 224.00Labile aluminium (|ig 1"‘) (Al-L) 19.30 57.26 B.D 594.00Henriksen critical load (keq H+ha"' yr"')(HCL)
4.10 5.03 0.00 33.82
Diatom critical load (keq H+ha' y r') (DCL)
7.49 13.06 0.00 85.29
B.D = below detection limit, SD == standard deviation, Min - minimum. Max = Maximum.
critical load (from 0 to 85.29 keq H'^ha y r T h e SD for each determinand is concomitantly
large. The broad range of values for these determinands reflects the wide geographical
coverage of the CLAG sampling programme. Because the data were taken from a national
database the chemistry are derived from environments as diverse as NW Scotland and East
Anglia as evidenced, for instance by the range of nitrate values (1489peq 1' , mean 43.79p.eq
r ) reflecting the polarity of chemical characteristics between remote upland areas and
lowland agricultural areas.
5.2.1.2 Correlation Structure
Table 5.2 is a matrix of Pearson's product-moment correlations showing the relationship
between the transformed water chemistry determinands together with the derived critical
loads values. The significance level of each correlation coefficient was established by
determining the p-value. A p-value greater than 0.01 indicates that a correlation is not
117
00
Cond 0.501
Aik 0.836 0.728
Na* 0.252 0.830 0.416
K+ 0.404 0.830 0.616 0.707
Mg:+ 0.575 0.902 0.745 0.721 0.750
Ca * 0.779 0.830 0.893 0.503 0.697 0.819
Cl 0.242 0.845 0.419 0.962 0.725 0.722 0.522
NO, 0.287 0.468 0.383 0.259 0.442 0.521 0.490 0.316
SO/ 0.408 0.861 0.602 0.641 0.750 0.803 0.783 0.673 0.567
TOC 0.039 0.407 0.251 0.341 0.427 0.331 0.351 0.353 -0.021 0.317
AI-NL -0.561 -0.181 -0.413 -0.060 -0.098 -0.217 -0.320 -0.046 -0.095 -0.116 0.390
Al-L -0.531 -0.194 -0.424 -0.179 -0.164 -0.223 -0.309 -0.183 0.024 -0.060 -0.031 0.394
HCL 0.803 0.661 0.881 0.306 0.527 0.718 0.890 0.301 0.453 0.568 0.214 -0.398 -0.342
DCL 0.775 0.808 0.929 0.466 0.676 0.777 0.968 0.487 0.457 0.725 0.342 -0.349 -0.346 0.922
pH Cond Aik Na+ K+ Mg:+ Ca*" Cl NO, SO/ TOC Al-NL Al-L HCLCorrelation coethcients with /j-values <U.Ui are shaded, i hese are signiticant at the level.
Table 5.2: Matrix of Pearson product-moment correlations for 14 transformed water chemistry determinands from 954 CLAG sites
significant at the 1% level. Those relationships that are statistically significant are shaded.
This level was chosen (rather than the 5% level) because of the large number of
observations in the dataset. The matrix shows the expected high positive inter-correlations
between the ionic concentrations (Na , Cl', Mg "", K"", Ca^l, conductivity and alkalinity. As
well as calcium, the diatom critical load (DCL) is highly positively correlated with alkalinity
and conductivity. The coefficients for the steady state water chemistry critical load (HCL)
with the other determinands are lower than those for DCL, an artefact of the different
methods of calculation {i.e. DCL is calculated using parameters derived from water chemistry
while the HCL calculation includes runoff, rainfall and ANC components (see Chapter 3).
Both aluminium species exhibit weak inverse relationships with most of the chemistry
determinands. Nitrate and TOC are not strongly correlated with any of the other
determinands.
A closer examination of the relationships between these variables is possible using the
unconstrained ordination technique, PCA, described in Chapter 4. This approach allows an
examination of the underlying structure of the data.
5.2.1.3 Principal Components Analysis
Principal Components Analysis (PCA) was undertaken on the transformed water chemistry
determinands from the full dataset (Appendix 5.1). This was implemented using the
FORTRAN program, CANOCO, Version 3.10 (ter Braak, 1990). The chemistry data were
centred and standardised to zero mean and unit variance, ter Braak (1987a) distinguishes
between variable centred or 'ordinary' PCA and standardised PCA. The former involves
weighting the value of each variable by its variance. Those with high variances thus
dominate the PCA results while those with low variances have much less influence (Odeh
119
et al., 1991). In standardised PCA the value of each variable is further divided by its
standard deviation. Consequently, variables with low variances unduly influence the PCA
outcome. The implications apply equally to RDA. The merits and drawbacks associated with
standardisation have been discussed widely (Noy Meir etal., 1975; Davis, 1986; ter Braak,
1987a; Odeh et al 1991; Le Maitre 1982) the consensus view being that data measured in
different units must be standardised. Given that the response data includes ionic
concentrations, pH and organic fraction, standardized PCA is used here.
The PCA results are presented in Table 5.3. The eigenvalues in a PCA represent the
proportion of the total sum of squares in the response data extracted by each ordination axis
and can be expressed as the percentage of the total variance in the chemistry data
accounted for by each ordination axis. This equates to the total sum of squares of the
chemistry variables on the latent explanatory variable and as such the eigenvalue is equai
to the goodness of fit.
In Table 5.3 the percentage variance explained by the ordination axes is shown cumulatively
and is calculated by dividing the eigenvalues by the sum of all unconstrained eigenvaiues
{I.e. the total variance) (ter Braak, 1990). In PCA, as implemented in CANOCO, the latter
is always unity. The computed determinands, HCL and DCL, are included in the analysis
as passive variables. This means that although they appear in the output in the same way
as the other determinands, they do not exert any influence on the derivation of the ordination
axes as their scores are caiculated afterwards by means of transition formulae (ter Braak
1987a).
Approximately 54% of the variance associated with the water chemistry for the 954 sites can
be explained by the latent variable(s) represented by the first PCA axis. The second axis
accounts for 16.5% while axes 3 and 4 account for 8 . 8 and 7.4% respectively. The first axis
120
thus dominates the structure of the chemical data. Examination of the variable loadings
Table 5.3: Results of PCA on transformed water chemistry determinands (954 sites).
1PCA Axes
2 3 4
Eigenvalue .541 .165 .088 .074Cum % varianceVariable loadings (correlations)
54.1 70.6 79.4 86.8
pH .6611 -.6714 -.0850 -.1867Aik .8286 -.3830 -.0642 -.2599Cond .9609 .1652 .0062 .0739Na+ .7770 .3575 -.1647 .4446K+ .8529 .2192 -.0054 .0055
.9280 .0442 .0779 -.0012Ca + .9056 -.2033 .0373 -.2832CI .7919 .3711 -.1297 .4214NQ,• .5314 .0101 .6851 -.1228SO/- .8603 .1869 .2449 -.0505TOC .3843 .5078 -.4972 -.4961Al-NL -.2770 .7706 -.0907 -.3443Al-L -.3247 .5560 .5429 -.1397
HCL (passive) .7568 -.3900 .0614 -.3343DCL (passive) .8791 -.2498 -.0022 -.2894
(which equate to correlations with the PCA axes) shows that this first component is most
highly correlated with Cond, Mg^*, Ca "" and DCL (loadings of .9609, .9280, .9056 and .8791
respectively) although, with the exception of the aluminium species, all the chemical
variables have positive loadings and the axis is almost unipolar. The major gradient
represents ionic concentration and this is reflected by the correlation structure identified in
Table 5.2. The implication here is that this ionic strength gradient is being driven by a latent
environmental variable that covaries in the same direction. The second component has a
high negative correlation with Al-NL and high loadings are also exhibited by Al-L, TOC (none
of which had strong correlations with the other determinands in Table 5.2) and pH (which
has a high negative loading). PCA Axis 3 displays reasonably high positive correlations with
NOg' and Al-L while Axis 4, with high loadings for TOC, Na"" and Cl' suggests a sea-salt
gradient. However, the variance explained by PCA axes 3 and 4 is comparatively low.
121
Figure 5.1, a plot of ordination loadings (the co-ordinates of the arrowheads) for PCA axes
1 and 2 for the 15 chemistry determinands, illustrates the patterns quantified in Table 5.3.
The vectors for each water chemistry determinand point in the direction of maximum
Figure 5.1: PCA correlation biplot of 15 water chemistry determinands for 954 CLAG sites (plotted using CALIBRATE - Juggins and ter Braak. 1993)
\odII
< N
c/2
<
l-N L
0.6 —
TOC
0.4 —
Na
S 040.2 —
Cond
-Mg-----0.0 — N 03
- 0.2 —
DCL
-0.4 — HCL Aik
- 0.6 —
0.75 0.90 1.050.45 0.600.00 0.15 0.30-0.30 -0.15-0.45
Key
AikCondAl-NLAl-L
Alkalinity Conductivity Non labile aluminium Labile aluminium
Axis 1 = 0 .541
TOC ; Total Organic CarbonDCL : Diatom Critical LoadHCL ; Henriksen Critical Load
The key is applicable to subsequent ordination diagrams
variation, the length being proportional to this variation. Therefore, arrowheads furthest from
122
the origin are the most important indicators of site differences while those nearest are less
important (ter Braak & Prentice, 1988). Variables with vectors characterised by acute angles
are inferred to be positively correlated, the correlation increasing proportionally with the
length of the arrow which equates to the correlation coefficient. Obtuse angles signify
negative correlations and the inverse relationships between pH and the aluminium species
from the axis 2 variable loadings are illustrated in the PCA bi-plot. Because the data are
standardized, correlations are displayed rather than covariances. By projecting the
arrowhead of a vector onto, for example, PCA axis 1, the magnitude of the correlation
between the variable represented by the vector and the axis can be estimated, (ter Braak,
1995). The further away from the origin that the projected line bisects the axis, the greater
the correlation. The PCA bi-plot enables the underlying structure of the chemistry data to
be examined, showing which of the first two PCA axes each determinand is most associated
with. PCA bi-plots can also be produced for axes 3 and 4.
Alkalinity, conductivity, the anions and the base cations all display a similar direction and
magnitude of correlation with the first axis except NOg' (PCA axis 1 loading of .5314). The
vector for calcium indicates a maximum variation along the first principal component axis
with DCL varying concomitantly. A PCA biplot incorporating the sites (Figure 5.2) shows how
each of these is arranged relative to the first two PCA axes. Those sites with the highest
positive scores on axis 1 are characterised by high Ca '" concentrations whereas those with
negative scores have lower Ca "" values. Table 5.4 shows the untransformed Ca "" and DCL
values for seven sites selected subjectively from across the first PCA axis. These sites have
been flagged on Figure 5.2 and read from left to right across the plot. The PCA Axis 1
scores for each site are included to further illustrate the sequence. There is a sequential
increase in Ca '" concentration from left to right across the plot except for the relative
positions of CZNS25 and CZNF75. The latter has very high Na and Cl' values (1377 and
1699peq 1' respectively) and this has pulled the site further along the first axis. The strong
123
positive relationships between individual determinands and the first ordination axis is further
illustrated by a scatterplot of DCL against PCA axis 1 (Figure 5.3).
A number of other sites have been labelled on Figure 5.2. These are towards the extremes
of the first and second PCA axes and are included to illustrate how the structure of the
chemistry data determines the position of the sites in the ordination diagram. Table 5.5
shows values for some of the key determinands for these sites. The chemistry of CZSD61
Figure 5.2: PCA biplot showing the position of sites relative to the first two PCA axes (vectors have been multiplied bv three for clarity)
04.c /3
<<u
4.5
CZSD61
3.0 — CZNR15
CZSE12
NaC Z U
ondCZNH57 '
0.0 —
> '
CZNN99CZSS92
• CZN057
-3.0
3.0- 1.0 0.0 1.0 2.0- 2.0-3.0
PCA Axis 1 = 0.541
is dominated by Al-L (91|ieq C) and is positioned at the furthest extreme of its vector.
CZSE12 has a similar AL-I level but is characterised by high concentrations of base cations
124
{eg. - 3035p.eq 1' ) and SO /' (3684|ieq i' ) and is thus located at the positive extreme
of the PCA plot. The site with the highest axis 1 score {i.e. furthest to the right) (CZNZ32)
has a similar chemical composition but with much higher Ca^ (6200peq 1' ) and alkalinity
(5896peq 1' ). CZSS92 has lower base cation concentrations but a high pH (7.86). Both
CZNN99 and CZNJOO have fairly dilute chemistry but the latter bisects the Al-L vector
further away from the origin due to its greater ionic concentration (38|ieq I'"'). The positions
of CZNR15 and CZN057 are primarily determined by their TOC values (9.5 and O.Smg \'\
respectively) which place them at opposite ends of the TOC vector.
The structure of the chemistry data is dominated by a gradient representing ionic strength.
This is highly correlated with DOL suggesting that the latent explanatory variable(s) implicit
in Axis 1 is (are) important in determining surface water sensitivity to acidification. The next
section examines the nature of the surrogate catchment data and subsequently, the explicit
relationships between these and the water chemistry variables are examined.
Table 5.4: Ca^ and POL values for seven sites along the first PCA axis
CZNJOO CZNH37 CZNC40 CZNS25 CZNF75 CZNS66 CZNZ32
Ca^ ( leq 1') 12 30 73 337 300 1901 6200
DCL (keq H* ha * yr) 0.11 0.24 0.65 3.37 2.68 19.9 64.43
PCA Axis 1 score -2.2088 -1.2130 0.1195 0.1951 0.9732 1.5393 2.9391
125
Figure 5.3: Scatterplot of DCL against PCA axis 1 site scores
X<<UCL
3.0
2.0 —
1.0 —
iSf0.0 —
- 1.0 —
- 2.0 —
-3.0
0.6 0.8 1.0 1.2 1.6 ,8 2.00.2 0.4 1.4 10.0
Transformed DCL
Table 5.5: Key chemical values for selected outlying sites (t=ueg \'\ *=uScm'\ BD=Below detection limits)
Sites pH Alkt Cond* Na^t Mg*t Ca^t SO^^t A l-L t
CZSD61 3.82 0 122 239 60 39 233 91
CZNR15 4.82 0 717 4635 800 336 541 10
CZSE12 7.77 1586 848 3013 1591 3035 3684 89
CZNZ32 8.07 5896 1191 3500 2948 6200 2630 BD
CZSS92 7.86 774 167 331 330 967 266 BD
CZN057 7.37 263 59 186 82 220 135 BD
CZNN99 5.84 18 11 43 9 33 27 18
CZNJOO 7.74 546 71 197 153 350 19 38
126
5.2.2 Exploratory data analysis - explanatory variables
This section presents exploratory analysis on variables which will subsequently be used to
explain, in a statistical sense, variation in the chemistry data. Summary statistics and
correlation analysis are undertaken on the continuous data (e.g. altitude, deposition and
rainfall). Bar charts are used to summarise the distribution of sites among the nominal and
ordinal variables. The derivation of these explanatory variables in described in Chapter 4.
The aim here is to summarise the explanatory data, prior to relating it directly to the
response data.
5.2.2.1 Summary statistics for continuous variables
The continuous catchment attribute data were transformed to reduce skewness and fulfil the
assumption of normality required by the statistical techniques used. Logio transformations
were used with the appropriate constant except in the case of altitude where a square root
transformation was more useful. Table 5.6 shows the mean, standard deviation, minimum
and maximum values for the untransformed explanatory data. Summary statistics for the
transformed data are presented in Appendix 5.2.
Table 5.6: Summary statistics for untransformed catchment/explanatorv variables (n-954)
Mean S.D Min. Max
Total S dep (89-92) 0.87 0.43 0.24 2.64(keq H-kha ' y f ') (Sdep)Total N dep (89-92) 1.15 0.45 0.34 2.75(keq H*ha ' y f *) (Ndep)Site altitude (m) (Alts) 259.41 186.83 0.00 1000.00Rainfall (1989-92) mm y f ‘ (Rain) 1414.93 690.07 468.00 3749.00Distance from sea (km) (Dist) 20.86 17.71 0.15 78.04
The nature of the CLAG sampling strategy means that all national deposition and
127
precipitation regimes are represented comprehensively in the dataset. Broad gradients in
altitude and distance from sea are also covered.
5 2.2.2 Correlation
Table 5.7 is a correlation matrix showing the relationship between the transformed
continuous independent variables. There is a high correlation between Sdep and Ndep
(r=0.893) but these are poorly correlated with rainfall (r=0.176 and 0.270, respectively).
Rainfall and altitude are positively correlated (r=0.433). Caution needs to be exercised when
interpreting these correlations as the deposition and rainfall data are not site specific and,
in the former instance, are extrapolations from a sparse monitoring network (UKRGAR,
1990).
Table 5.7: Matrix of Pearson product-moment correlations between transformed catchment attributes for 954 CLAG sites
Sdep 0.176
Ndep 0.302 0.893
Rain 0.433 0.169 0.270
Dist 0.406 0.319 0.399 -0.281
Alts Sdep Ndep Rain
All correlations have p-values < 0.01 and are significant at the 99% level
5.2.2.S Distribution of nominal/ordinal variables
The number of sites in each class of the nominal/ordinal classifications is shown by a series
of bar charts (Figures 5.4). The meaning of each class is detailed by the accompanying key.
The soil critical load classification is dominated by the SCL2 class which is based on granite
and gneiss parent materials. With the land classification system, the distribution of sites is
128
Figure 5.4: Bar charts showing the distribution of sites for the nominal/ordinal explanatory variables
a) SOIL CRITICAL LOAD b) LAND CLASSIFICATION
0 50 I (XI 150 21X1 250 3(X) 350 4000 50 100 150 200 250 3(X) 350 41X1 450 5(X1No. o f sites
c) LAND COVER (9 CLASS) d) LAND COVER (6 CLASS)
KX) 150 21X1 250 3(X) 350 4(X)
No. o f sites
e) SITE SENSITIVITY
0 50 to o 150 20 0 250 3(X) 350 4(X) 450No. o f sites
129
Key
SOIL CRITICAL LOAD (SCL)
Critical load Class (keq H* ha ' yr'*)
4.02.0 I.O 0.5 0.1
LAND CLASSmCATION
Class Description
ArablePastoralMarginal Upland Upland
LAND COVER (9 CLASS) (LVa)
Class Description
1 Water and built/bare ground2 Agricultural grass3 Arable4 Deciduous woodland5 Coniferous woodland6 Lowland semi-natural grass/moor a) grass7 Lowland semi-natural grass/moor b) dwarf shrub8 Upland semi-natural grass/moor a) grass9 Upland semi-natural grass/moor b) shrub
LAND COVER (6 CLASSES) (LVb)
Class Description
1 Water and built/bare ground2 Agricultural grass and arable3 Deciduous woodland4 Coniferous woodland5 Lowland semi-natural grass/moor6 Upland semi-natural grass/moor
SITE SENSITIVITY (SS)
Class Description
1 Acid waters will not occur2 Acid waters very unlikely3 Acid waters unlikely4 Acid waters likely at very high flows5 Acid waters will occur at all flows
heavily weighted towards the upland (LC4) and marginal upland (LC3) classes. Similarly,
the modal classes for the land cover classifications both represent upland moor (LVa8 for
the 9 class hierarchy and LVa6 for the 6 class hierarchy) although there are almost as many
sites in the agricultural grass/arable classes. The site sensitivity classification is
characterised by a bimodal distribution with most sites occurring at either end of the
sensitivity scale. The modal class is the most sensitive one (SS5). The distribution of sites
across these classifications exhibits a consistent trend. The majority of sites are associated
with classes which represent sensitive conditions (e.g. sensitive soils and upland areas).
This is not surprising given the nature of the CLAG sampling strategy which a), targeted the
most sensitive areas in each square, and b), sampled at a lower spatial resolution in regions
130
of low or no sensitivity to acidification (Kreiser et al., 1993).
5.2.3 Direct gradient analysis - chemistry and catchment data
This section examines the relationships between the water chemistry for each site and the
surrogate catchment attributes relating to that site. The direct gradient analysis techniques
discussed in the preceding chapter are employed to examine the degree to which variation
in chemical composition can be explained by external independent data. The constrained
ordination method, RDA, is used to quantify these relationships.
The Phase 1 explanatory dataset includes five different classifications. The soil sensitivity
and soil critical load classifications are characterised by overlapping attributes (see Chapter
4). The same applies to the land cover and land classification variables. Additionally, there
are two hierarchical aggregations of the land cover classification. Prior to analysis of the full
dataset a number of preliminary analyses were undertaken to select the classifications with
which to proceed further. The objective is to reduce the number of explanatory variables to
limit the amount of spurious explanation.
5.2.3.1 Preliminary Redundancy Analysis (RDA) of the nominal/ordinal variables
To assess the relative importance of the different classification variables a number of RDA’s
were undertaken so that the explanatory power of each classification can be assessed
independently. The results of these are summarised in Table 5.8. RDA requires that
nominal variables be converted into dummy variables. Thus each of the classes in the five
classifications used became an individual variable with a value of either 0 or 1 for each site.
This approach is akin to multiple discriminant analysis whereby each class is represented
131
by its mean (ter Braak, 1987b).
Table 5.8: RDA of chemistry and selected classifications
Eigenvalues
Variables Axis 1 Axis 2 Axis 3 Axis 4 constrained eigenvalues
Site sensitivity (SS) .282 .006 .001 .001 .290Soil critical load (SCL) .285 .011 .001 .001 .298
Land classification (LG) .290 .008 .002 .000 .300Land cover - 9 classes (LVa) .297 .008 .005 .002 .313Land cover - 6 classes (LVb) .256 .006 .002 .000 .265
Classifications were selected for inclusion in the full analysis on the basis of the eigenvalue
and the number of categories in each class. Thus LVb with 6 classes is selected ahead of
LVa (9 classes) to reduce the number of explanatory variables with only a negligible loss of
explanatory power. Land classification data is omitted from subsequent analysis because
it includes soil and geology components as well as land use and thus replicates aspects of
the soil critical load and sensitivity classifications. Additionally the land cover data are
available nationally at 25nf resolution although this factor will only become important during
the catchment specific approach described in Chapter 6 . Soil critical load (SCL) is to be
included but site sensitivity is not. Both classifications include a soil and geology component
and are therefore likely to exhibit collinearity. Additionally, the site sensitivity classes were
amended to include the influence of land use whereas the SCL classification used here is
unmodified for land use and will not be collinear with the land cover class. During the next
analytical phase SCL is treated as a continuous variable because of its ordinal nature. An
RDA using SCL classes as dummy variables has previously been undertaken (Kernan,
1995) and the results are similar to those presented below.
S.2.3.2 RDA of chemistry and catchment variables
RDA is an analytical progression from PCA in that it arranges sites according to the
132
influence of specific, rather than latent, environmental variables. A multiple regression of site
scores on the predictor variables is performed and the new site scores are simply the fitted
values of this regression. The axes are constrained to be linear combinations of the
explanatory variables and the sum of canonical eigenvalues indicate how well these
variables explain the total variation in the chemistry data. RDA was implemented using
CANOCO Version 3.10. The water chemistry data were centred and standardised, and DCL
and HOL were rendered passive. CANOCO also standardises the explanatory variables prior
to running the ordination. Table 5.9 shows a summary of the RDA. The percentage of the
variance of the chemistry explained is presented cumulatively for each RDA axis. This is
supplemented by the cumulative percentages for the fitted data. These equate to the fraction
of the fitted relationship explained by given axes and is calculated by dividing the sum of
those axes by the sum of all canonical eigenvalues. Additionally the biplot scores for each
of the water chemistry determinands are shown together with the percentage of each
response variable explained by the model. Biplot scores for each of the explanatory
variables are also included. The variance inflation factor (VIF) (Montgomery and Peck, 1982)
for each of these is presented. If the VIF for a variable is greater than 20 this indicates that
the variable is almost perfectly correlated with the other variables and, as such, does not
contribute any extra fit to the regression model. In the terminology of RDA its canonical
coefficient is unstable and it should not be interpreted (ter Braak, 1986). The VI Fs for Sdep
and Ndep are considerably higher than those for the other variables and this indicates that
these are likely to be collinear. LVbG has no VIF as it is the K-Xh dummy variable of a
dummy variable with K classes and will therefore be multicollinear with the other classes.
The eigenvalues for the first two RDA axes are .437 and .045 respectively. Therefore
approximately 47.5% of the variance in the water chemistry data can be explained by the
first two ordination axes. This is considerably different from the cumulative variance
explained by the first two axes from the PCA model (70.6% - see Table 5.3). The implication
133
Table 5.9: Results of RDA on chemistry and environmental variables (954 sites)
Axes 1 2 3 4 Total variance
Eigenvalues .427 .043 .024 .010 1.00
Cumulative percentage variance of;response data 42.7 47.0 49.4 50.5fitted relationship 83.8 92.4 97.2 99.2
Sum of all canonical eigenvalues .509Response variable scores (adjusted for variance) % explained
pH .4911 -.1977 -.2144 -.0192 32.92Aik .6733 -.2145 -.1603 -.0321 52.78Cond .8656 .0762 .0476 -.0245 76.01Na+ .7210 .4184 -.0310 .0227 69.71K+ .7887 -.0081 -.0173 -.0276 62.72Mg2+ .8132 -.0045 .0191 .0424 66.78Ca^ .7956 -.2210 .0033 -.0596 68.69CI .7370 .3438 .0028 .0857 69.75NQ,- .4652 -.2653 .1204 .2313 36.17SO/ .8201 -.1330 .2543 -.0119 76.20TOC .4368 .0223 .0135 -.2420 25.88AI-NL -.1280 .0874 .2163 -.0387 8.30AI-L -.1617 -.0642 .3401 -.0645 15.28
HCL (passive) .5720 -.2579 -.1136 .0209 42.12DCL (passive) .7476 -.2218 -.0641 -.0396 61.83
Biplot scores of explanatory variables V I F
Alts -.8509 -.4020 .1250 .0296 2.7Sdep -.0590 -.3380 .6614 .2558 5.6Ndep -.2303 -.3708 .5255 .0320 6.6Rain -.7748 .2654 -.1794 .3143 2.6Dist -.1149 -.8868 .1784 -.3297 2.5SCL -.7874 .1799 .2547 -.1052 2.1LVb I -.0504 -.0430 -.1000 .3166 1.1LVb2 .6857 -.3589 -.2516 .2945 2.4LVb3 .0938 -.0559 .0884 .1333 1.2LVb4 .0172 .0091 .2331 -.0209 1.2LVbS -.1674 .0919 .4236 .0448 1.5LVb6 .5565 .3035 -.2142 -.3898 0.0
here is that the explanatory variables do not fully reproduce the latent axes extracted by the
PCA and that the water chemistry data are influenced significantly by external variables not
included in the RDA (ter Braak 1987a). This is not surprising given the nature of the
surrogate catchment variables. There are no data relating specifically to soil, geology or
hydrology and none of the data (except altitude and distance from sea) relate specifically
to the catchment from which the water sample was taken.
134
A comparison of the first axis eigenvalues for the PCA and RDA (.541 and .427 respectively)
indicates that the constrained ordination reproduces a substantial proportion of the dominant
gradient exhibited by the unconstrained PCA. This is not surprising given that the RDA axis
1 accounts for 83.3% of the fitted relationship. The explanatory power of the model is thus
dominated by the first RDA axis.
Detailed examination of Table 5.9 yields useful information about the different relationships
between the chemistry and catchment variables. The axis biplot score for a variable is its
eigenvector coefficient and shows its position along that axis. In PCA and RDA these are
correlations with the ordination axes. A score of 1.0 along a given axis will mean that the
variable is perfectly correlated with that axis. A number of chemical determinands have high
scores along the first axis (Cond, Na\ K , Ca "", Cl', S O /' and DCL) and these reflect the
dominant chemical gradient that was also observed in the PCA plot.
Two types of score can be distinguished for the explanatory variables. A biplot score for
quantitative data places the arrowhead of the vector for a variable relative to the canonical
axes. This value equates to a correlation with the canonical axis. For nominal data the
position of each class is based on its centroid score which locates a class at the centre of
all those sites characterised by that class (although the biplot score still provides a
correlation and this is used in Table 5.9). The highest explanatory scores along axis 1 are
exhibited by altitude, rainfall, SCL LVb6 and LVb2. This suggests that these variables are
driving the first RDA axis and are thus most important in explaining variation in the chemical
determinands with the highest scores. The percentage of the variance of each of these
response variables determined by the model is also included in Table 5.9 and the figures
confirm this supposition. Those with the highest percentages explained are also those with
the highest RDA axis 1 scores.
135
With the exception of the aluminium species, the chemistry determinands are positively
correlated with RDA axis 1. This also applies to LVb2 while altitude, rainfall, SCL and LVb6
are negatively correlated. These relationships are illustrated by Figure 5.5, a correlation
biplot of the redundancy analysis. Correlations here are estimated by orthogonal projection
of the arrowhead, or centroid, onto the axis, or bisecting a line between the arrowhead or
centroid, and the origin of another variable. As conductivity, DCL and the concentrations
of the controlling ionic determinands decrease, altitude, rainfall and SCL increase. This
indicates that chemistry varies along altltude/ralnfall and SCL gradients. Similarly those sites
in class LVb6 , (upland moor and grassland) are likely to be characterised by low ionic
concentrations, conductivity and DCL (and are therefore more sensitive). Conversely, LVb2
sites (arable and managed grassland) will be at the less sensitive end of this gradient.
Although none of the other axes are characterised by high variable response scores a sea-
salt gradient (high loadings for Na'' and Cl ) is apparent along RDA axis 2. This appears to
be driven by distance from sea (Axis 2 bi-plot score = .8 8 6 8 ). Labile aluminium in surface
waters is statistically linked with high S and N deposition by virtue of their concomitantly
high scores for axis 3.
To test the statistical significance of the ordination results a Monte Carlo permutation test
was employed (Hall and Titterington, 1989). This technique involves randomly permuting the
sites in the environmental dataset and then randomly matching these with the chemistry
data. The eigenvalue and sum of all eigenvalues (trace) is then calculated for the
randomized dataset. This process is repeated many times (in this instance 99). If the
chemistry is responding to the environmental data the observed test statistic should be
larger than most (/e 99%) of those produced randomly. Where the observed value is in the
top 1 % highest values it can be concluded that water chemistry is significantly related to the
surrogate environmental variables. Tests of significance were run both for the first canonical
ordination axis and for the overall dataset. In both cases the outcome was a p-value of 0.01
136
Figure 5.5: RDA correlation biplot of chemistry and surrogate catchment data (954 sites) showing water chemistry (solid vectors), continuous catchment parameters (dashed vectors) and dummy variables (filled circles).
cnSdII
(Nc/3
<<§
0.6
NaLVb60.4 —
Rain ^LVbS
0.2 —
A I-N L Cond
M g ' "
S 04
0.0 —
A l-L
- 0.2 — CoDCLLVb3L V b l* N 03 HCL
|S&pNdep
Alts
- 0.6 —
Dist
0.6 0.8 1.00.2 0.4- 0.2 0.0- 0.6 -0.4- 1.0 -0.8
R D A Axis 1 = 0.427
Key
LVbl Water and built/bare ground SCL Soil Critical LoadLVb2 Agricultural grass and arable Dist Distance from seaLVb3 Deciduous woodland Sdep Sulphur depositionLVb4 Coniferous woodland Ndep Nitrogen depositionLVb5 Lowland semi-natural grass/moor Rain RainfallLVb6 Upland semi-natural grass/moor
which means that the relationships are significant at the 99% level. This level was chosen
in view of the size of the dataset, relationships tending to be of greater significance in larger
137
datasets. The second, third and fourth RDA axes were also found to be significant at this
level.
The use of RDA as a dimension reduction tool is exemplified in the next section which deals
with the forward selection of explanatory variables.
S.2.3.3 Forward selection of explanatory variables
Forward selection in CANOCO Version 3.1 (ter Braak, 1990) allows the identification of a
minimum set of catchment variables that explain the chemistry data almost as well as the
full set. This has advantages in terms of simplicity and the logistics of data collection. The
implications of eliminating variables from a multivariate statistical model depends on the use
being made of the model. Where prediction and estimation of responses are the primary
objectives regression is fairly tolerant towards the elimination of variables (Rawlings, 1988).
At each forward selection step CANOCO selects the variable that adds most to the total
explained variance of the chemistry data. Additionally, it is possible to test the statistical
significance of the variable selected at each step using a Monte Carlo permutation test.
Bonferroni adjustment can be used to prevent too many variables being judged significant
(ter Braak, 1990, Manly, 1992). This requires that the significance level at each stage of the
forward selection is increased by dividing the initial significance by the number of the
selection. Following the first selection (p <0.01) the second selection requires a p- value of
0.01/2, the third 0.01/3, and so on, until no further explanatory variables are significant. For
this reason forward selection is thus used as a test of significance of specific variables.
Table 5.10a shows the results of the forward selection using the RDA undertaken on the
critical loads data. This shows the cumulative variance explained by the addition of each
138
variable Identified by the forward selection procedure. Also included is the Bonferroni
adjusted significance level and the number of permutations required to determine it. Once
the selection procedure has been stopped an RDA is run using only those variables selected
together with any that are multicollinear with them. The eigenvalues for each RDA axis and
the cumulative percentage variance are included in Table 5.10b.
Table 5.10a : Forward selection of environmental variables
V a r ia b le a d d ed c u m u la tiv e v a r ia n c e o f
se lected v a r ia b le s
n u m b e r o f
p e rm u ta tio n s
B o n fe rro n i
re q u ire d
s ig n ifican c e
s ig n ific a n c e
a ch ieve d
Alts .32 99 .01 .01R ain .42 999 .005 .001LVb2 .45 999 .0033 .001Sdep .47 999 .0025 .001D ist .48 999 .002 .001SCL .50 999 .0016 .001N dep .50 9999 .0014 .0001LVbl .50 9999 .00125 .0001LVb6 .50 9999 .00111 .0001
V a r ia n c e e x p la in e d b y a ll
v a r ia b le s .51
N o o ther variables sign ificant
Table 5.10b: RDA summary using variables identified in the forward selection procedure
Axes 1 2 3 4 T o ta l v a r ia n c e
E ig e n v a lu e s .426 .043 .024 .009 1.00
C u m u la t iv e % v a r ia n c e o f
species d a ta 42.6 46.9 49.3 50.2
S u m o f a l l c a n o n ic a l e igenva lues .509
The sum of all canonical eigenvalues using forward selection (.506) compares favourably
with that for the whole dataset (.509). However, only LVb classes 3, 4 and 5 are not
statistically significant. All the continuous variables are selected. Given the size of the
dataset this is not surprising.
139
The structure of the relationships between and within the chemistry and catchment datasets
has been examined in some detail. DCL is seen to vary along the same gradient as the
dominant axes in both the unconstrained and constrained analyses. The next section
examines the relationships between DCL alone and the catchment attributes.
S.2.3.4 RDA using DCL as a single response variable
An RDA was undertaken with the full suite of explanatory variables but with DCL as the sole
response variable, the other chemistry variables being made passive in an attempt to identify
the catchment attributes that best explain variation in baseline critical load. This approach
is equivalent to the use of multiple regression of a number of independent variables onto a
single dependent variable. With a single response variable there will only be one constrained
axis. Table 5.11 provides a summary of the RDA with biplot scores along the first axis for
the water chemistry and the catchment attributes. The eigenvalue for the RDA axis is .618
which is equivalent to the proportion of the variance in DCL explained by the catchment data
in the RDA of the full dataset (see Table 5.9) but greater than for the chemical composition
as a whole. The VIF values for the explanatory variables are the same as those for the full
response dataset. A Monte Carlo permutation test on the first (and only) canonical axis
showed that it was significant at the 99% level.
A fon/vard selection was undertaken with Monte Carlo permutation testing to determine the
explanatory variables selected. Bonferroni adjusted significance levels were used. Table 5.12
shows the results of this analysis.
As DCL was highly correlated with the first axis in the first RDA it is not surprising that the
most influential catchment attributes are the same as those identified for the full suite of
chemical determinands. However, the forward selection identifies SCL as the dominant
140
influence here and there are fewer significant selections. LVb2 (arable and agricultural
grassland) is the only significant dummy variable.
Table 5.11: Results of an RDA on DCL and catchment attributes (954 sites)
Eigenvalue for 1st axis .618
Percentage variance of species data 61.8
Sum of all canonical eigenvalues .618
Response variable scores - Axis 1 % explained catchment attribute(adjusted for variance) biplot scores- all passive except DCL
pH .5418 32.92 Alts -.7025Alkalinity .7186 52.78 Sdep -.0138Conductivity .7999 76.01 Ndep -.1373Na+ .5655 69.71 Rain -.8089K+ .7551 62.72 Dist .1435Mg2+ .7680 66.78 SCL -.8289Ca-* .8242 68.69 LVbl -0191CI .5850 69.95 LVb2 .7516NQ,- .4946 36.17 LVb3 .0986SO/ .8010 76.20 LVb4 .0053TOC .4160 25.88 LVb5 -.2310AI-NL -.1665 8.30 LVb6 -.5743AI-L -.1591 15.28HCL .6283 42.12DCL .7863 61.83
The relationships between DCL and the catchment variables are further illustrated by box-
plots (Figures 5.6a and 5.6b) in the case of the nominal data (For clarity SCL is presented
as a nominal variable) and scatterplots for the qualitative variables (Figure 5.7).
Table 5.12: Forward selection with DCL as a sole response variable
BonferroniVariable added cumulative variance of
selected variablesnumber of
permutationsrequiredsignificance
significanceachieved
SCL .42 99 .01 .01Rainfall .55 999 .005 .001LVb2 .58 999 .0033 .001Alts .49 999 .0025 .001Dist .61 999 .002 .001
Variance explained by allvariables .62
No other variables significant
The box and whisker plots provide a visual representation of the shape of the data
distribution related to soil critical load (Figure 5.6a) and land cover (Figure 5.6b). The
141
median, range and fourths (or quartiles) of each class are summarised relative to one
another (Thompson, 1992). For soil critical load there is no overlapping of the median (the
line between the upper and lower boundaries of each box). The distribution of sites related
to soil critical load suggests that even at this crude level of analysis it is possible to relate
the water chemistry of a site to the soil critical load of the 1 km square containing the site.
With regard to the land cover classification, LVbl (bare surfaces) ranges across the whole
critical load gradient but Figure 5.6b does confirm that sites characterised by higher critical
loads tend to be arable or managed grassland sites (LVb2) while low critical loads coincide
with upland areas (LvbS). The scatterplots (Figure 5.7) confirm the inverse relationships
between critical load and both altitude and rainfall and the poor correlations with N
deposition, S deposition and distance from sea.
Figure 5.6: Box plot showing DCL values classed according to:
a) soil critical load
■s
CO§
dÛ
2.0
1.8 -
1.6 —
1 .4 -
1.0 -
0 .4 -
0 .2 -
0.0
Soil critical load class
142
b land cover
■s
a
2.0
1 .8 -
1 .6 -
1 .4 -
1 .2 -
1.0 -
0 .8 -
0 .6 -
0 .4 -
0.2
0.0
1 2 3 4 5 6
Land cover classFigure 5.7 Scatterplots showing DCL against continuous catchment variables
40
1 0
0
0.6
Alts Ndep Dist0.6“
f . • •2.0"
0.5- ' •• • 1.6-
0.4-
0.3-
0.2-
0.1 -*•
1.2-
0.8-
0.4-
0.0-1
1 #• •--- n----1 1 1
0.0 0.4 0.8 1.2 1.6 2.0
0.4-
0 .2 -
0.0
Sdep RainJ.O *
5 : - ' . ' 3.4-«b W » :
-----r 1 1 1
3.2-
3.0-
2.8-
2.6-1 1 r 1 1
0.0 0.4 0.8 1.2 1.6 2.0 0.0 0.4 0.8 1.2 1.6 2.0
DCL
143
5.3 Analysis of a reduced dataset of more sensitive sites (Ca <200^eq I )
The analysis carried out on the 954 CLAG sites (954 dataset) reported above shows the
extent to which variation in DCL can be explained by catchment attributes across a wide
chemical gradient. Prediction of sensitivity to acidification has been undertaken using low
resolution data (Hornung et al., 1995a). Additionally, surrogate catchment classifications
have been used to identify differences in sensitivity (Hall etal., 1995a; Kernan, 1995). The
extent to which catchment characteristics can predict surface water chemistry at the more
sensitive end of the spectrum is less well documented. In this section, the analyses of the
full Phase 1 dataset is repeated using a more sensitive subset of sites. The criterion used
to define these sites is a calcium concentration of less than 200 peq/l. These sites are more
likely to be susceptible to acidification at current deposition levels and very few CLAG sites
(<5%) with Ca "" concentrations higher than this threshold have been acidified (C.Curtis,
pers. comm.) This section specifically examines whether the relationships identified across
the broader chemical gradient are repeated for a shorter sensitivity gradient. Some
exploratory analyses of the sensitive sub-set has been undertaken but these are not central
to the objectives of this chapter and are presented in Appendix 5.3.
Following the selection of Ca "^<200 peq/l as the sensitivity threshold, the number of sites
in the analysis was reduced to 469 (469 dataset). The data were transformed using the
same criteria as the full dataset. Z scores were calculated to identify potential outliers. These
mainly consisted of sites with very low alkalinity. There was no reason (e.g. measurement
error) for these to be removed from the analysis.
5.3.1 Redundancy analysis of water chemistry and catchment variables
An RDA was undertaken on the sensitive subset using the same approach as that for the
144
954 dataset. Table 5.13 summarises the results.
Table 5.13: Results of RDA on chemistry and environmental variables (469 sites)
Axes 1 2 3 4 Total variance
Eigenvalues .256 .100 .026 .022 1.00
Cumulative percentage variance ofspecies data 25.6 35.6 38.2 40.4species*environment correlations 62.1 86.4 92.7 98.0Sum of all canonical eigenvalues .412
Response variable scores (adjusted for variance) % explained
pH -.0896 -.4011 -.0753 .2096 22.16Aik -.0221 -.4246 -.1215 .2051 24.66Cond .7955 -.0249 .0342 -.0555 63.94Na* .7606 -.3033 -.0961 -.0979 69.00K* .6010 -.0189 -.0298 .0128 40.13Mg2* .7122 -.0709 .0396 .1119 52.97Ca^ .3878 .0279 -.2493 .2101 26.98CI .7579 -.2861 .1442 -.0782 69.15N Q / .0626 .4312 .1777 .2694 30.09SO / .6301 .5090 .0196 -.1136 67.65TOC .2786 .1232 -.4218 -.1188 28.76AI-NL .1356 .2035 -.1013 -.1562 10.76Al-L .1219 .5243 .0082 -.0142 29.35
HCL (passive) -.0708 -.1834 -.1492 .2217 15.24DCL (passive) .1302 -.1809 -.2619 .2033 10.76
Biplot scores of explanatory variables V IF
Alts -.8818 .2854 .1291 -.0679 2.17Sdep -.2247 .7016 .2086 .1206 10.09Ndep -.2576 .6580 .0498 .0638 10.27Rain -.4142 -.4534 .3692 -.0932 1.79Dist -.5864 .5470 -.4517 .2556 2.60SCL -.2989 .0922 -.0047 -.5485 1.25LVbl -.1574 -.0332 .5500 .0419 1.06LVb2 .2040 -.0009 .0708 .8534 1.32LVb3 .0636 .1142 .0119 .0952 1.11LVb4 .0414 .1262 .0120 .0944 1.12LVb5 .1017 .3152 .0878 -.0970 1.28LVb6 -.1848 -.3466 -.3072 -.4858 0.00
The eigenvalues for axes 1 and 2 (0.256 and 0.1 respectively) total 0.356 and explain a total
of 35.6% of the variance in the chemistry data. This compares with 47% for the full dataset.
The highest response scores are exhibited by those determinands with the highest scores
on the PCA (Cond, Na"", Mg "" and Cl'; see Appendix 5.3). The variance explained for each
145
of these determinands individually has remained stable (in comparison with the 954 dataset)
but has fallen from 69% to 27% for Ca and from 62% to 11% for DCL. The highest scores
exhibited by the explanatory variables are those for Alts and LVbl along axis 1. Sdep and
Ndep have relatively high VIF's indicating that they are collinear. The second axis is more
important in the 469 dataset (explaining 10% of the response variance and 24.3% of the
fitted relationship) than in the full dataset (4.3% and 9.1%, respectively). This second axis
is most strongly correlated with Al-L (.5243), SO /' (.5090), NO ' (.4312), pH (-.4011), and
alkalinity (-.4246). The explanatory variables most correlated with RDA axis 2 are Sdep
(.7016), LVb3 (.7369), Ndep (.6580) and LVb4 (.6139). RDA Axes 3 and 4 explain,
statistically, 2.6 and 2.2% of chemical variation, respectively. On Axis 3, distance from sea
and LVbl (bare ground) have the highest variable loadings. The chemistry determinand
most associated with this axis is TOC which exhibits a positive relationship with distance
from sea and varies negatively with LVbl. This reflects the lower levels of organic matter
which occur in catchments with high proportions of bare rock. Additionally, along the west
coast of Scotland peat depth decreases with distance from sea in response to the increasing
elevation.
An RDA biplot (Figure 5.8) illustrates the structure of these relationships. Axis 1 appears to
represent an inverse relationship between altitude and ionic strength. This is similar to the
first axis from the full dataset with the exception of the Ca "" vector which is obviously much
reduced. LVbl (bare ground) is positively correlated with altitude whereas sites classed as
arable/managed grassland (LVb2) are characterised by higher conductivity, Mg "", Na and
Cr levels. The influence of LVb2 on the chemical composition of less sensitive sites is
clearly important. The second axis illustrates the inverse relationship that pH and alkalinity
exhibit with NO3 , S O /' and Al-L. This acidity gradient is clearly related to Sdep and Ndep.
These variables were seen to have little explanatory power in the RDA applied to the 954
dataset because deposition of N and S does not determine sensitivity. However the
146
relationship between these input parameters and water chemistry is much stronger at
sensitive sites. With a reduced catchment buffering capacity pH, alkalinity, labile aluminium
and nitrate and sulphate concentrations are influenced by atmospheric deposition.
Interpretation of some of the nominal classes should be tempered by the fact that some of
these are very poorly represented in the 469 dataset. Less than 19% of sites are in land
cover classes 1 to 4. Under these circumstances sites with high values for a particular
determinand may be over influential.
Figure 5.8: RDA biplot of chemistrv and environmental data (469 sites)
0.75Sdep
NdepLVb4
LVb5 'Dist
S 0 4
0.50 —
Nd3
8 Alts
0.25 —O Al-NJ
CNC/D<<
TOC
SCL CaLVb2-----0.00 —
Cond
DCLHC|LVbl-0.25 —
LVb6 No
Rain Aik
-0.50
1.00-0.50 -0.25 0.00 0.25 0.50 0.75-0.75- 1.00
R D A Axis 1 = 0.256
The first RDA axis was found to be significant at the 1% level using a Monte Carlo
permutation test. This was repeated for axes two to four and all were found to be significant
at this level.
147
A forward selection was carried out to see which variables were significant in explaining the
variation in chemistry. Bonferroni adjusted significance levels were used.
5.3.2 Forward selection of environmental variables
The sum of all canonical eigenvalues using forward selection is .394, almost identical to that
obtained using all the explanatory variables. The land cover classification is represented by
LVb1 and LVb2. Most of the variance is explained by Alts, Rain, Sdep and Dist. SCL, Land
cover (LVb) classes 3-6 and Ndep were not significant. Significance in this instance relates
primarily to the first canonical axis as most of the variation in chemistry occurs along this
gradient. The results of the forward selection are displayed in Tables 5.14a and 5.14b.
Table 5.14a: Fonward selection of environmental variables
Variable added cumulative variance of number of Bonferroni required Significanceselected variables permutations significance achieved
Alts .21 99 .01 .01Rainfall .26 999 .005 .001Sdep .33 999 .0033 .001Dist .37 999 .002 .001LVb2 .39 999 .0016 .001LVbl .40 9999 .00125 .0001
Variance explained by allvariables .41
No other variables significant
Table 5.14b: RDA summary usina variables from fon/vard selection
Axes 1 2 3 4 Total variance
Eigenvalues .255 .099 .022 .018 1.00
Cumulative % variance ofspecies data 25.5 35.4 37.6 39.4
Sum of all canonical eigenvalues .394
148
5.3.3 RDA using DCL as a single response variable
The analysis using DCL as a sole response variable was repeated for the sensitive sites.
With such a reduced Ca^ gradient, DCL variation is restricted. Table 5.15 shows that the
eigenvalue for the single canonical axis is 0.164 indicating that 16.4% of the variation of
DCL can be explained by the catchment data. Forward selection identifies only Alts, Dist and
Sdep as significant (Table 5.16). The eigenvalue for this axis is 0.097 (Table 5.16b) which
is significant at the 99% level. The weak relationships exhibited by DCL and the surrogate
catchment variables are illustrated by Figures 5.9a and 5.9b (box and whisker plots of DCL
against the SCL and LVb classifications, respectively) and Figure 5.10 (scatterplots of DCL
against the continuous Sdep, Ndep, Dist, Rain and Alts). Neither of the nominal
classifications show clear differentiation between DCL value and class. Similarly, the
scatterplots suggest very poor relationships between DCL and the continuous explanatory
variables.
Table 5.15: Results of an RDA on DCL and catchment attributes (469 sites)
Eigenvalue for 1st axis .164
Percentage variance of species data 16.4
Sum of all canonical eigenvalues .164
Chemistry determinand scores • Axis 1 % explained catchment attribute(adjusted for variance) biplot scores
pH .2976 22.16 Alts -.5577Aik .3685 24.95 Sdep -.4232Cond .2177 63.94 Ndep -.3274Na+ .2707 69.00 Rain -.1703K* .2076 40.13 Dist .0191Mg"+ .2881 52.97 SCL -.3747Ca + .3900 26.98 LVbl -.3742CI .2461 69.15 LVb2 .4357NQ,- -.1568 30.17 LVb3 .0430S O / .0521 30.09 LVb4 .1253TOC .2417 67.65 LVb5 -.2648AI-NL -.0558 28.76 LVb6 .0476Al-L -.2096 10.76
HCL (Passive) -.2805 29.35DCL (Passive) .4050 15.24
149
Table 5.16: Forward selection with DCL as a sole response variable
Variable added cumulative variance of selected variables
number of permutations
Bonferroni required significance
significanceachieved
Alts .05 99 .01 .01Dist .08 999 .005 .001Sdep .10 999 .0033 .001
Variance explained by allvariables .16
No other variables significant
Figure 5.9: Box and whisker plots of distribution of site DCL according to classificationvariables - sensitive sites
a) Soil critical load classification (SCL)
■s
Ii
0.6
0 .5 -
0 .4 -
0.3 —
0 .2 -
0.1 -
0.01 2 3 4
Soil critical load class
150
b) Land cover classification (LVb)
■2
c/5
g
do
0.6
0 .5 -
0 .3 -
0 .2 -
0.1 -
0.01 2 3 4 5 6
Land cover class
Figure 5.10: Scatterplot showing DCL against continuous environmental variables - sensit; sites
40
30-
2 0 -
10 -
0
Sdep3.6
3.3^
3.0-
Rain!*>* «V. »
2.70.0
T0.2
T
0.4
Alts Ndep Dist
#u.t>
0.5-
0.4-
0.3-
0 .2 -
0.1 -*
z.u
1.6-
1.2-
0.8-
0.4-
0.0-1
11:1 I
0.0 0.2 0.4 0.6
0.6
DCL
151
5.4 Discussion
The purpose in undertaking this preliminary analysis was first, to assess the feasibility of
developing a predictive critical loads model based on a paramaterisation of catchment
attributes and second, to identify the variables that would be most valuable in this context.
With regard to the first objective, the patterns emerging from the ordinations undertaken on
the full dataset showed that the sites themselves exhibit a reasonably well defined chemical
composition. A strong calcium gradient was identified along the first ordination axis and this
is encouraging for the development of a critical loads model which will be based on this
determinand. Further, the addition of the secondary environmental variables during RDA
showed a significant level of explanation. Forward selection and selective RDA then
provided some idea of which of these were likely to be of most importance in the
development of the catchment model. Approximately 51 % of the variation in water chemistry
is explained by catchment attributes, primarily by altitude and rainfall (Table 5.10a).
However, the discussion in Chapter 2 emphasised the importance of soil geology and land
use in determining surface water chemistry. Modelling studies and ion budget calculations
suggest that deposition inputs (for which altitude and rainfall can be viewed as surrogates)
and catchment processes are equally important in controlling the acid-base chemistry of
surface waters (Cosby etal., 1990; Nilsson, 1993). However, predicting critical loads is the
same as predicting sensitivity to acidification and, within the context of a steady state
empirical model, sensitivity is more likely to be determined by within catchment attributes
rather than acid deposition, rainfall or altitude (although, over time the cumulative effects of
acid deposition can increase sensitivity - see Chapter 6 ). Separate analysis here shows that
SCL (soil critical load), rainfall, LVb2 (agricultural grass and arable land cover) altitude and
distance from sea accounts for 62% of the variation in diatom critical load (Table 5.12). The
influence of soil critical load is particularly prominent when explanation relates to DCL only.
152
The fact that altitude and rainfall remain highly influential when DCL is the sole response
variable may be a consequence of the lack of data relating specifically to the characteristics
of the catchment within which each site is contained. Altitude may thus be acting as
surrogate for variables such as soil type and geology where upland areas are more likely
to be characterised by thin acid soils and base poor, slowly weathering parent material.
These analyses have been carried out on a large, very noisy dataset. There are likely to be
strong interrelationships between soil, geology and altitude. Consideration should also be
given to the fact that SCL is only a crude surrogate for catchment geology and soil type and
that all the nominal variables do not relate to the catchment from which the water sample
was taken, but to a 1km grid square. Despite the disparity in eigenvalues for the first two
axes between the PCA and RDA the outcome of this initial exploration of the data has
proved encouraging and has provided a general overview of relationships between surface
water chemistry and the surrogate catchment variables. Each of the environmental variables
selected has some influence, directly or indirectly on the water chemistry of the sample sites.
With the ultimate objective being the development of a predictive model there are a number
of intermediate steps to be taken. Essentially these will involve the identification and
quantification of those environmental and response variables which best explain the
relationship between catchment and water chemistry. This initial analysis has provided some
pointers towards these.
However, the strong relationships identified in the full dataset are weaker if the same RDA
techniques are applied to a sensitive subset. Although an ionic gradient is identified using
the 469 dataset this excludes Ca "" (and therefore DCL) with conductivity, Na\ Mg^ . Cl and
s o / ' predominant. Other determinands still vary widely when Ca is less than 200peq 1-1.
Certain catchment parameters are still strongly correlated with the first axis (Alts, LVbl and
LVb2) but this relates to sea-salt rather than overall ionic composition. An acidity gradient
153
is observed along the second canonical axis and this seems to be driven primarily by
atmospheric deposition. It is possible that for more sensitive sites estimations of alkalinity
or pH may be made more successfully.
In spatial terms, the use of catchments as the unit of measurement for predictor variables
rather than square kilometres will also be extremely important. The grid base approach used
here does not consider the processes operating within the catchment. These ultimately
determine the sensitivity to acidification of the surface waters. Soil, geology, land cover and
hydrology may vary substantially throughout the catchment. The 1km grid square in which
the sample site is situated may or may not be representative of the catchment attributes as
whole.
There is clearly potential, as evidenced by the Phase 1 analysis, for predicting whether or
not a site will be sensitive across a wide chemical gradient. However the catchment
variables used in this analysis are not sufficient to explain DCL at the sensitive end of the
spectrum to the degree that the development of a predictive model will require. This
conclusion is similar to that reached by Hall et a i, (1995a) and confirms the findings of
Hornung et al., (1995a) with regard to the efficacy of applying national data at a local
resolution. The use of more sophisticated catchment specific data in the next chapter seeks
to establish whether a higher resolution approach may be more useful in this context.
154
CHAPTER 6 : PHASE 2 - MODEL DEVELOPMENT AND CALIBRATION
6.1 Introduction
The Phase 1 analyses undertaken in Chapter 5 indicate that variation in surface water
chemistry can be explained, statistically, by a readily available dataset of catchment
characteristics. Across a broad chemical gradient, these data, a combination of observations
relating to the catchment itself, the geographical area and mapped attributes relating to the
grid square in which the catchment is situated, explained 50% of the variation in water
chemistry and approximately 60% of the variation in diatom critical load (DCL). To develop
a statistical model which will allow the prediction of surface water critical loads, a higher
level of statistical explanation is required. This requires that the catchment data are specific
to the catchment and at a much higher resolution. The Phase 1 catchment data is grid
based and at an inappropriate resolution for studies of small catchments. As such it does
not specifically relate to the catchment from which the water sample was taken.
By examining a reduced number of sites it is possible to characterise (and quantify)
catchment attributes in much greater detail. This chapter examines the relationships between
a high resolution catchment specific dataset (incorporating land cover, soil, geology and a
number of other variables) and a calibration water chemistry dataset comprising 78 sites
selected according to the criteria outlined in Chapter 4 (Section 4.2). The derivation of the
former is discussed in Chapter 4 (Section 4.5). These data will then be used to develop the
predictive model. These are the 'Phase 2 ’ analyses. The water chemistry determinands
comprise the response or dependent variables while the catchment attributes are the
explanatory (predictor) or independent variables The analyses involve quantifying the
relationship between the response and explanatory variables, identifying which of the latter
best explain variation in the former. Emphasis is then placed on the exploratory analyses
155
of the relationships between the catchment variables and diatom critical load (DCL) as a
single response variable.
The analytical format employed is similar to that used in Chapter 5. The analysis proceeds
as follows:
i) Exploratory analvsis of the water chemistry (response) and catchment (explanatorv)
variables independently: this includes summary statistics and principal components analysis
(PCA) of the water chemistry and the catchment attribute data for 78 calibration sites. PCA
allows the structure of the response and explanatory datasets to be examined independently
of each other. Collinearity within each can thus be highlighted and gradients of variation can
be identified.
ii) Exploratory analvsis of the relationships between the response and explanatory variables
: Redundancy analysis (RDA) is used to quantify the variation in water chemistry explained
by the catchment data. The role of individual chemical and catchment variables is assessed
and forward selection (see Chapter 4) is used to identify catchment attributes which explain
a significant amount of variation in water chemistry, within the context of the RDA statistical
model. The aim is to examine the extent to which the catchment characteristics explain the
major chemical gradients. Following this holistic approach, further RDAs are performed to
assess the relationships between catchment variables and diatom critical load (DCL). The
objective here is to identify the most important catchment attributes in terms of explaining
critical load variation. Partial RDA (ter Braak and Prentice, 1988) is used to characterise the
variation into a series of components (comprising explanatory variable types), covariance
terms between components and unexplained variation.
iii) Analvsis of a sensitive subset of the chemistry data : this echoes the Phase 1 analysis
156
where a subset of the full dataset (sites where Ca "" <2 0 0 |i.eq 1' ) was examined to assess
whether relationships identified in the full Phase 2 dataset (which includes a broad gradient
of water chemistry - see Section 6.2) hold for more sensitive sites. Given that the dataset
used in Phase 2 comprises 78 sites, a threshold value of Ca '' <400peq 1' is used to derive
a sensitive subset (comprising 46 sites). The statistical analyses described in ii), above, are
repeated on this reduced dataset.
Most of the statistical techniques used here were also employed in Chapter 5 and are
discussed in Chapter 4. The rationale and justification for the particular approaches have
already been presented (Chapters 4 and 5).
6.2 Exploratory analysis - chemistry and catchment datasets
6.2.1 Water chemistry data
Exploratory analyses was undertaken on the water chemistry for the 78 sites which
constitute the Phase 2 dataset. This comprises 14 chemical determinands together with
derived critical load values for the Henriksen and Diatom models (see Chapter 3). The
analytical methodology is discussed in Chapter 4. Summary statistics are presented to
illustrate the range and distribution of each determinand. Principal components analysis
(PCA) allows the internal structure of the chemistry dataset to be examined and key
gradients of variation to be identified.
The chemistry variables were log transformed prior to analysis (with the exception of pH
which is measured on a log scale) to facilitate the application of parametric statistical
methods (Ott, 1990). No outliers were indicated following the calculation of Z scores for each
determinand. It is clear that the smaller sample size for the Phase 2 analyses has eliminated
157
the chemical extremes encountered in the larger Phase 1 dataset.
6.2.1.1 Summary statistics
Summary statistics for water chemistry, calculated using SAS, are presented in Table 6.1
2 (calibration) dataset (n=78)
Mean SD Min. Max.
pH 7.06 0.74 4.87 8.21Alkalinity ( leq 1') (Aik) 510.06 586.40 -12.00 3264.00Conductivity (|iScm'‘) (Cond) 109.53 91.74 20.00 476.00Sodium (neq 1‘‘) (Na*) 349.78 207.34 98.00 1240.00Potassium (|ieq 1') (K*) 25.00 25.56 6.00 138.00Magnesium (|ieq 1') (Mg^*) 284.38 282.90 39.00 1462.00Calcium (peq 1‘) (Ca^) 649.72 754.37 60.00 3847.00Chloride (jieq 1') (Cl ) 327.31 195.17 73.00 1269.00Nitrate (^eq 1') (NO /) 72.41 142.06 B.D 612.00Sulphate ( leq 1') (S O / ) 181.94 261.46 28.00 1722.00Total monomeric aluminium 32.76 48.95 B.D 350.00(pg 1') (AL-TM)Non labile aluminium (pg 1') (AL-NL) 23.85 37.90 B.D 258.00Labile aluminium (pg 1‘) (AL-L) 8.91 13.47 B.D 92.00Absorbance (250nm) (Abs-250) 0.31 0.30 0.04 1.48Henriksen critical load 11.42 11.13 1.79 56.11(keq H+ha ' yr *)(HCL) Diatom critical load 7.12 8.14 0.58 40.67(keq H+ha' y f ‘) (DCL)
B.D = below detection limit, SD = standard deviation, Min. = minimum, Max. = maximum
These reflect the chemical gradients across which the 78 model calibration sites vary. Ca^ ,
for example, varies between 60 and 1462peq 1' with a mean value of 649.72|ieq \'\ NOg'
has a range of 612|ieq 1' (the minimum is below detection limit) with a mean of 72.41 jieq
\'\ Mean pH is circumneutral at 7.06 (minimum, 4.87; maximum 8.21) These examples,
together with the summary statistics for the other determinands show that the calibration
sites are characterised by fairly diverse water chemistry. However, a close examination of
158
the minima for each determinands show that this gradient does not encompass very low
ionic concentrations. This is particularly the case for base cation concentration minima (Ca +,
60peq r \ Mg^ , 39peq 1' ). Additionally the minimum values for alkalinity (-12|Lieq 1' ) and pH
(4.87) are relatively high. It is likely therefore, that the chemical gradients shown in Table
6.1 do not encompass sites which are highly sensitive to acidification (Ca^'"<50|ieq \'\
Battarbee et al., 1988). The minima for the Diatom (DCL) and Henriksen (HCL) critical
loads (0.58 and 1.79 keq H ha' yr respectively) bear this out.
Comparison with the national critical loads mapping dataset used during Phase 1 analysis
further illustrates the lack of highly sensitive sites in the Phase 2 dataset. Summary statistics
for the Phase 1 dataset (Table 5.1) relate to water chemistry for 954 sites which have a
much wider geographical spread (Figure 4.1) than the Phase 2 sites (Figure 4.2). A
comparison between Tables 5.1 and 6.1 reveals that the chemical gradients observed during
the Phase 1 analyses are considerably longer than those exhibited by the much smaller
Phase 2 dataset. The sulphate range is reduced from 6041 to 1694peq \' with the mean
value falling from 303.28 to 181.94)ieq \'\ Mean alkalinity remains fairly similar (550peq 1'
in the Phase 1 compared to 510peq 1' for Phase 2) although the range is much narrower
(from 6415 to 3274|j,eq 1' ). In the Phase 2 sites, pH varies from 4.87 to 8.21 compared with
a range between 3.82 to 9.21 for Phase 1 sites. However the mean pH has risen from 6.53
to 7.06. The calcium gradient for Phase 1 (8088peq/l'\ mean 714.44peq I'"') is substantially
smaller than that for Phase 2 sites (3787peq 1' ) although the mean Ca "" is quite similar at
650peq l '\ As discussed, the minimum Ca '" is 60peq \' here compared with 12peq 1' in the
Phase 1 data. As with the Phase 1 analysis the critical load distribution reflects that of Ca "".
For the Phase 2 data, the minimum DCL is 0.58keq H^ha^ yr^ in contrast to the zero values
encountered during Phase 1 analysis. The mean is slightly lower for the Phase 2 dataset
(7.12keq H ha'"' yr^ against 7.49 H^ha' yr' for phase 1). It seems the limited size and
geographical scope of the Phase 2 dataset has produced shorter gradients (with higher
159
minima and lower maxima) for most of the chemical determinands. The exceptions are non-
labile aluminium (Phase 1 maximum 224pg \'\ Phase 2 maximum 258pg \ '\ ), HCL (Phase
1 maximum 33.82keq H^ha' y r \ Phase 2 maximum 56.11 keq H'"ha yr^ ) together with
those determinands with minima below detection limits in both datasets. Mean values tend
to be lower for Phase 2 data, the exceptions being pH (see above), Mg (284|xeq 1' for
Phase 2, 240|ieq M for Phase 1), NOg' (72 and 44peq 1' respectively), AI-NL (23 and 17|ig
r"" respectively) and HCL (11.42 and 4.10 H'^ha'' yr' respectively). The Phase 2 analyses
do not include total organic carbon, but do include absorbance at 250nm (Abs-250) which
is an effective surrogate for TOC. This ranges from 0.04nm to 1.48nm with a mean of
0.31 nm. An Abs-250 gradient between 0.002 and 2.17nm (mean 0.24nm) has been
identified for a 643 site Scottish dataset (Harriman and Pugh, 1994) again showing a much
wider gradient of variation than that exhibited by the Phase 2 dataset.
The comparison between the preliminary Phase 1 dataset and the Phase 2 calibration
dataset presented here shows that the most sensitive sites, characterised by very low ionic
strengths and critical loads below 0.5keq H^ha' yr are not represented in the latter. The
calibration dataset is therefore weighted towards less sensitive sites. Figure 6.1 compares
the proportions of sites in each of the CLAG critical load classes for the Phase 1 and Phase
2 datasets together with the full CLAG database. Although the CLAG sampling strategy was
targeted towards the most sensitive sites within each 10 km grid square (Kreiser et al, 1993)
the overall dataset is still heavily weighted towards non-sensitive sites (Curtis et al, 1995)
because of the large proportion of sites underlain by non-sensitive soil and geology in
lowland areas. Nevertheless 21% of CLAG UK mapping sites fall into the two most sensitive
DCL categories (0-0.2 and 0.2-0.5 keq H ha^ yr^) while none of the Phase 2 sites have
DCL values in these classes.
The lack of highly sensitive sites in the Phase 2 dataset is likely to be a consequence of the
160
sampling strategy employed. This sought to encompass a gradient across the more common
soil types in Scotland as well as covering the main geological and land cover groups. Thus
although a number of the sites were situated in sensitive areas, the attribute-led sampling
strategy has not led to the inclusion of calibration sites which fall into the two most sensitive
CLAG critical load classes. This is despite the fact that of 18 soils targeted to ensure
maximum coverage of Scottish soil types, 9 are defined as highly sensitive based on
Figure 6.1 : Percentage of Diatom Critical Load classes for the Phase 1, Phase 2 and CLAG mapping datasets (Curtis etal, 1995)
0 .0 - < = 0.2 0.2-<=0.5 0.5-<=1.0 Critical load classes
1.0 - < = 2.0 ? zo
| l / / l Phase 2 dataset I I Phase 1 dalsel H CLAG mapping s itâ"
percentage base saturation and soil pH criteria as discussed in Chapter 4 (Langan and
Wilson, 1991). It may be that the sensitivity gradient identified with respect to soil
associations does not translate to surface waters. Equally, it is possible that other catchment
criteria (e.g. geology and landcover) do not encompass the more sensitive types in the
161
calibration dataset as well.
Clearly it will not be possible to calibrate the predictive model to encompass sites outside
the observed DCL range. This does not present problems at the upper end of the scale
where critical loads are much higher than deposition levels which have (from the CLAG
database) UK maxima of 2.64 keq H"^ha'\r'^ (S deposition) and 2.75 keq H^ha'\r'^ (N
deposition). However, in more sensitive areas, relatively low deposition levels can cause
critical load exceedance (Allott etal., 1995a). With regard to the Phase 2 model calibration,
the paucity of sensitive sites is likely to be of greater significance as these sites are under
the greatest threat of acidification. Minimum UK deposition levels for S are approximately
0.2 keq H'^ha'Vr'^ (CLAG database, UCL) which would cause S critical load exceedance at
almost 6 % of the CLAG mapping sites. The lack of very sensitive sites in the calibration
dataset means that the predictive model will not be applicable to these types of sites. The
implications of this for the model calibration are discussed in greater depth in Chapter 8 .
6.2.1.2 Principal components analysis (PCA) - water chemistry data
Principal components analysis (PCA) is used to examine the relationships between the water
chemistry determinands and the underlying structure of the chemistry data. The use of PCA
as a statistical technique is described in detail in Chapter 4. Its use as a tool to examine
trends and interrelationships within a water chemistry dataset is described with regard to the
Phase 1 preliminary analysis in Chapter 5. Here, an identical approach is employed in terms
of the software used, data transformation and standardisation and the zero weighting given
to the two critical load variables. Concepts underpinning the interpretation of the results of
PCA analysis are also detailed in the previous chapter and will not be repeated here.
The PCA results are presented in Table 6.2. A total of 93.3% of the variance in water
162
chemistry over the 78 model calibration sites can be explained by the first four PCA axes
(the cumulative eigenvalues for Axes 1-4 are 0.933). Approximately 58% of the variation is
explained by Axis 1 while 24% is explained by the second axis. The variable loadings show
that most determinands are highly correlated with Axis 1 (Cond, 0.936; Ca '", 0.954 Mg ,
0.898 and DCL, 0.944). This is indicative of a 'total ionic strength’ gradient along the primary
axis. However, the second axis also appears to be of considerable importance. High
loadings are exhibited by Na"’ (0.613), Cl (0.629), Al-L (0.665) and Abs-250 (0.826). This
suggests a gradient associated with sea-salts and organic carbon. Low lying sites closer the
sea, particularly in-the West of Scotland will tend to have thicker peat coverage than
Table 6.2: Results of PCA on transformed water chemistry determinands (n = 78).
1 2P C A Axes
3 4
Eigenvalue .578 .236 .067 .051Cum % variance 57.8 81.4 88.2 93.3
Variable loadings (correlations)pH .7337 -.4191 .4916 .0097Aik .8582 -.2227 .4031 .0361Conductivity .9363 .3047 .0917 -.0295Na+ .6795 .6132 .0397 -.3473K+ .7875 .3969 -.3286 .1721Mg2+ .8981 .2325 .1636 -.0471Ca^ .9543 .0259 .1647 .1260CI .6727 .6285 -.0233 -.3270NQ,- .8800 .0007 -.3373 .1645S O / .8030 .3077 -.3057 .3248Al-TM -.6794 .6586 .1975 .1461AI-NL -.7481 .5892 .0529 -.0299Al-L -.3175 .6651 .3392 .5047Abs-250 -.3879 .8262 .0650 -.1638
HCL (passive) .7970 -.0693 .4353 -.0904DCL (passive) .9442 -.0243 .1909 .0661
catchments in the upland areas further inland. Axis 1 also exhibits high variable loadings for
NOg' (0.880) and AI-NL (-0.748). The highest loadings on axis 3 are for pH (0.492), alkalinity
(0.403) and Henriksen critical load (0.455). This axis explains approximately 7% of the
variation in water chemistry and may represent a secondary acidity gradient although the
163
variable loadings are relatively low compared to those for Axes 1 and 2. Axis 4, with an
eigenvalue of 0.51 (5% of the variation) is dominated by Al-L with a loading of 0.505.
Although most of the variation in water chemistry occurs along the first PCA axis, the
variable loadings show that a number of gradients exist which, given the mathematical basis
of PCA, are not correlated with one another. As discussed in Chapter 4, each ordination axis
in a PCA represents a theoretical explanatory variable. With the chemistry for the Phase 2
sites varying across multiple gradients it is likely that the chemistry is being driven by
several, non-correlated explanatory variables. A matrix of Pearson product-moment
correlations for the water chemistry determinands (Appendix 6.1) quantifies the relationships
between each variables, identifying those correlations which are significant at the 95% level.
The relationships discussed above and quantified in Table 6.2 are illustrated by a correlation
biplot (Figure 6.2.) The length of the vectors for DCL and Ca^ and the acuteness of their
angles with the first PCA Axis are illustrate that these determinands are most closely
associated with variation along this dominant gradient, a supposition confirmed by the
magnitude of their variable loadings in Table 6.2. The vectors for HCL and NOg' also
describe acute angles with Axis 1 but are shorter and therefore, less important (Ter Braak,
1995). Conductivity, K’", SO/", Mg^", Alkalinity and pH are all exhibit their maximum variation
along this gradient. The contrasting water chemistry for sites mkOI (Ca^", 3597peq l '\ Mg^",
1270^ieq l '\ alkalinity 2116peq l '\ NOg', 238peq 1' ) and mk67 (Ca^", 78peq l'\ Mg^", 39peq
r \ alkalinity 29|ieq l '\ NOg, Ipeq 1' ), at opposite ends of Axis 1, illustrates the nature of the
dominant (ionic strength) gradient.
164
Figure 6.2: PCA correlation biplot of Phase 2 water chemistry (plotted using CALIBRATE -Juggins and ter Braak. 1993) - vectors have been multiplied by three to aid clarity.
3.0
A b s -2 5 0mk36
2.0 —A l-I
A l-n lmkOl
S 0 41.0 — mkO C< ndmk42Mg
mk03
mk65CNOo c0.0 —
DCLCOmk70HCL
Aik- 1.0 —
mklOmk68
- 2.0 —
mk22
-3.0
-3.0 - 2.0
KeyAikCondAI-NLAl-L
: Alkalinity : Conductivity : Non labile aluminium : Labile aluminium
- 1.0 0.0 1.0 2.0
PCA Axis 1 = 0..578
Al-tm : Total monomeric aluminiumAbs-250 : Absorbance at 250nmHCL : Henriksen critical loadDCL : Diatom critical load
3.0
The key is applicable to subsequent ordination diagrams
The high correlations with PCA Axis 2 exhibited by Na , Cl', Abs-250 and Al-L in Table 6.2
are illustrated by the positioning of their vectors. Sites mk40 (Abs-250, 0.821; AI-NL, 258pg
\'\ Al-L, 92jig r ) and mk22 (Abs-250, 0.037; all aluminium species below detection limit)
have similar Ca " levels (319 and 370 peg 1' respectively) but are at opposite ends of Axis
2 as a result of the polarity of their aluminium and organic carbon concentrations. It is
possible that the two axes reflect flow conditions at the time of sampling. The high
aluminium concentrations and Abs-250 values are consistent with flow pathways through
organic soil horizons. These tend to dominate during major storm events (Creasey et al.,
165
1986). The dominant determinands along Axis 1 include those which originate primarily from
mineral weathering (e.g. Ca '', Mg^ and alkalinity) which tend to have strong inverse
relationships with discharge (Reid et al., 1981). However it is recognised that, as a
consequence of the geological diversity in Scotland, there is considerable spatial variability
in the acid/base status of Scottish freshwaters (Craig, 1983). The chemistry data structure
described here is similar to that encountered during Phase 1 and for a 643 site Scottish
dataset (Harriman and Pugh, 1994) although it is possible that PCA Axis 1 is related to
discharge, the length of the gradient suggests that there is more variation than might be
expected from temporal variation alone. Additionally, no extremes of flow conditions were
observed during the sampling programme.
The strong relationship between the dominant chemical gradient in the Phase 2 dataset and
DCL is illustrated by Figure 6.3. This plots each site according to its log transformed DCL
and its score along the first PCA axis. The interpretation here is that, because there is a
strong linear relationship (Axis 1 score = 0.9942), the latent environmental variable driving
variation along Axis 1 may also be responsible for the DCL gradient. This is considered in
more detail below (Section 6.3). Figure 6.3 indicates two outlying sites, mk40 and mk76. The
former has a high Axis 1 score relative to DCL while the opposite is true of the latter. The
high aluminium and Abs-250 values which give mk40 such a high Axis 2 score relative to
Axis 1 have already been noted. Site mk76 has a relatively high Ca^ concentration
(1380|ieq 1' ) but Mg "" is lower (144|ieq 1' ) than at other high Ca^ sites while AI-NL is
relatively high (22pg 1' ). This may explain the anomalous position of the site.
The structure of the chemistry data following PCA described here is very similar to that for
the Phase 1 analysis (Figure 5.1) although the level of explanation offered by the second
PCA Axis is nearly 10% higher for the Phase 2 analysis. This indicates that the Phase 1
dataset is dominated to a greater extent by a single chemical gradient {i.e. ionic strength),
166
a result of the greater range of sites included in the Phase 1 data.
Figure 6.3: Scatterplot of DCL against PCA Axis 1 site scores.3 .0
2.0 —
X<<UOh
1.0 —
0.0 —
- 1.0 —
mk53
- 2.0
mkSO6 >
mk40
O O o
.............
CD° mk70mk42
mk28O
° ° omkl6 oO
_ O mk56 O .. .O ..........
Omk76
O
o omk24
0.15
OoOmk61
0.30 0.45 0.60 0.75 0.90 1.05 1.20 1.35
mkOl OO
mkO;
1.50 1.65
Log transformed diatom critical load
6.2.2 Catchment data
This section comprises exploratory analysis of the catchment attribute data for the Phase
2 sites. In terms of the modelling procedure, these represent the explanatory variables. The
derivation of the catchment data is discussed in Chapter 4. A number of the attributes
discussed in Chapter 4 have been omitted from the analyses. Variables were excluded
where the attributes they represented were replicated by similar datasets or data at different
resolutions. This is discussed in more detail in Section 6.3.
167
The purpose of this exploratory analysis is to illustrate the structure of the explanatory
catchment variables, in particular the collinearity between different data types and different
data resolutions. This will also assist in the reduction of explanatory variables during model
calibration.
6.2.2.1 Summary statistics
The catchment attribute data were transformed to reduce skewness and fulfil the assumption
of normality required by the parametric statistical techniques used. Log^o transformations
were used with the appropriate constant {i.e. to avoid negative logarithms), except in the
case of distance from sea (Dist), maximum catchment altitude (Altm) and soil water
concentration (H""), where a square root transformation was employed as this produced
distributions which were closer to normality. Table 6.3 provides the mean, standard
deviation, minima and maxima for the untransformed catchment data.
Catchment size ranges from 0.5km^ to 20km^ with the mean value (4.07km^) towards the
lower end of that scale. These are relatively small catchments, as prescribed in the sampling
strategy. Distance from sea varies between coastal sites to those over 70km inland. Site and
maximum catchment altitude varies from 5m and 30m to 470m to 1130m respectively with
means of 132m and 515m. Maximum altitude is used in all subsequent analysis as this
relates to the catchment as a whole and not simply the site which, by definition is the lowest
point of the catchment and, in larger catchments, may not usefully reflect the relief and
extreme conditions extant in the catchment. 8 deposition varies between 0.38 and 1.58keq
H"" ha' yr' while N deposition ranges between 0.53 and 1.58keq ha' y r'\ A broad range
of rainfall regimes is covered with a minimum of 709mm yr and a maximum of 3319mm
y r \
168
Table 6.3: Summary statistics for untransformed catchment/predictor variables (n=78)
Mean SD Min. Max
Catchment area (kra ) (Area) 4.07 3.02 0.47 19.67Stream length (km) (SJength) 9.06 10.26 1.27 84.80Distance from sea (km) (Dist) 24.51 20.28 0.02 76.05Site altitude (m) (Alts) 131.55 118.81 5.00 470.00Maximum altitude (m) (Altm) 514.67 278.29 30.00 1130.00Total S dep (89-92) 0.71 0.24 0.38 1.58(keq H*ha' y r') (Sdep)Total N dep (89-92) 0.96 0.31 0.53 1.58(keq H^ha ' yr ') (Ndep)Rainfall (1989-92) m m yf' (Rain) 1750.06 950.52 709.00 3319.00Geology Class 1 (%) (G l) 54.62 39.36 0.00 100.00Geology Class 2 (%) (G2) 24.35 33.75 0.00 100.00Geology Class 3 (%) (G3) 15.65 33.09 0.00 100.00Geology Class 4 (%) (G4) 5.35 16.76 0.00 85.83
Soil variables
High Soil Sensitivity (%) (SH) 68.03 37.19 0.00 100.00Medium Soil Sensitivity (%) (SM) 19.35 28.11 0.00 96.87Low Soil Sensitivity (%) (SL) 10.84 21.50 0.00 96.84Bare ground (%) (Bare) 1.78 6.31 0.00 45.87Soil Critical Load Class 1 (%) (SCLl) 3.27 7.79 0.00 42.06Soil Critical Load Class 2 (%) (SCL2) 0.43 2.54 0.00 20.58Soil Critical Load Class 3 (%) (SCL3) 34.98 40.16 0.00 100.00Soil Critical Load Class 4 (%) (SCL4) 53.52 41.34 0.00 100.00Soil Critical Load Class 5 (%) (SCL5) 6.02 19.76 0.00 100.00W A' concentration (H*neq 1') (H*) 13.93 6.60 2.93 29.30WA* Base saturation (%) (%BS) 24.19 16.48 4.76 76.17WA* Soil Critical Loadkeq ha' y r') SCL 5.63 3.55 0.00 15.96
Land Cover Class 1 (%) LCl 5.55 8.01 0.00 38.00Land Cover Class 2 (%) LC2 17.07 26.11 0.00 89.70Land Cover Class 3 (%) LC3 2.91 4.74 0.00 25.39Land Cover Class 4 (%) LC4 5.81 11.63 0.00 74.23Land Cover Class 5 (%) LC5 41.58 20.70 0.00 100.00Land Cover Class 6 (%) LC6 27.07 20.25 0.00 100.00
* Calculated using weighted averages (see Chapter 4)
SD = standard deviation, Min. = minimum. Max. = maximum.
The four geology classes identified in Chapter 4 are all represented in the calibration
dataset. These are represented as the percentage of the catchment surface area underlain
by geological types in each class (see Chapter 4). All classes vary between 0 and 100%
coverage with the exception of Class 4 (limestones, chalk etc.) which has a maximum
coverage of 8 6 %. Class 1 (granite and acid igneous rock) has the highest mean value (55%)
169
while Class 4 exhibits the lowest mean (5%)The soil data used in these analyses are based
on attributes mapped at the soil map unit scale (see Chapter 4). The data include two
classification systems, soil sensitivity to acidification (Langan and Wilson, 1991) and soil
critical load (Hornung etal., 1994). Each soil map unit is ascribed to a sensitivity class and
a soil critical load class. The catchment attributes include the percentage cover of each class
from both classifications. Percent coverage of all three sensitivity classes ranges from 0 to
almost 100%. The high sensitivity class (SH) has a mean coverage of 6 8 % while that for the
low sensitivity class (SL) is 11 %. Similarly, soil critical load classes 1 and 2 (SCLl and
SCL2), which are at the less sensitive end of the classification, are restricted in the dataset
with mean coverages of 3 and <1% respectively. The medium to high sensitivity classes
(SOLS and SCL4) have much greater representation in the dataset with mean coverages
of 35% and 53% respectively while SCL5, the most sensitive class averages 61% coverage
across all sites. This confirms that sensitive soil types are well represented in the dataset
although at the very sensitive end of the scale, soils with granite and quartzite parent
materials (SCL5) are represented in only 15 catchments. Bare ground {i.e. with no soil
cover) averages 2% coverage with a minimum and a maximum of 0 and 46%, respectively.
All but 2 sites (mk17 and mk19) have <10% bare ground.
The composite soil critical load variable, SCL (see Chapter 4 for an explanation of the
derivation of this variable) provides a single soil critical load for each catchment. In the
dataset this ranges from 0 to 16kg 8 y r \ with a mean value of 6 kg 8 yr^ which falls into
class 4 of the 8 CL classification. A single base saturation value is also determined for each
catchment based on weighted averaging (see Chapter 4). Base saturation has a range
between 5 and 76% with a mean of 24% which falls into the lower end of the medium class
of the sensitivity classification (Langan and Wilson, 1991). The soil water H"" concentration,
also derived using weighted averaging, varies between 3 and 29 H^peq 1' with a mean of
14 H'^peq \'\
170
The dominant land cover classes are LC5, (lowland semi-natural grass/moor), LC6 (upland
semi-natural grass moor) and LC2 (agricultural grass arable) with mean coverages of 42%,
27% and 17% respectively. The use of the Land Cover dataset (Fuller and Groom, 1993a)
at this resolution is discussed more fully in Appendix 6.2.
The ranges for N deposition, S deposition and rainfall in the Phase 2 dataset are less than
those from the Phase 1 dataset. This is to be expected given the more restricted
geographical scope of the former. Mean values for N (0.96keq H^ha'\r'^) and S deposition
(0.71 keq H"^ha'\r'^) are higher than those found during the Phase 1 analysis (1.15 and
0.87keq H^ha'\r'^ respectively). This reflects maps for S and N (Allott etal., 1995b; Fowler
at a!., 1994) which show, particularly for N, that deposition is greatest in the industrial
centres of Central and Northwest England with the highest concentrations in Scotland
occurring in the south west and the central lowlands. Most of the Phase 2 sites are in
Northern Scotland (see Figure 4.2). Mean rainfall for the 78 calibration sites is 1750mm yr^
compared with 1412mm yr^ for the preliminary dataset. The latter value is obviously
influenced by the low rainfall regimes in eastern and southern England.
6.2.2.2 Correlation
A Pearson product-moment correlation matrix (Table 6.4) quantifies the relationship between
the transformed catchment variables. The significance level of each correlation co-efficient
was established by determining the p-value. A p-value greater than 0.05 indicates that a
correlation coefficient is not significant at the 5% level. Correlation coefficients which are
statistically significant at this level are shaded. The matrix shows that this a large and
complex dataset with much collinearity and as such is given detailed consideration here.
The degree to which granite and acid igneous rocks (01, areas with little or no geological
171
buffering capacity) are present in catchments is negatively correlated with the presence of
basic and ultrabasic rocks (G3, r - -0.52), SCL3 (soils with parent materials on granodiorite,
greywacke, schist or gabbro, r= -0.69) as well as the composite soil critical loads variable
(SCL, r= -0.52). Positive correlations are found with highly sensitive soils (SH, r= 0.42) and
SCL4 (soil parent materials comprising granites and gneisses, r = 0.64). Conversely,
catchments with high proportions of limestones and chalk (G4) which provide almost limitless
buffering capacity (Kinniburgh and Edmunds, 1984) are positively correlated with SCL1, soils
with parent materials comprising limestone and marl (r = 0.42). These correlations are
probably due, in part to the geology classes being based on similar data. Weighted
averaged percentage base saturation (%BS) is positively correlated with variables which are
indicative of low sensitivity to acidification including SL (r= 0.7), SCL3 (r = 0.55) and the
derived soil critical load variable (SCL, r = 0.66). High negative correlation coefficients
characterise the relationships between %BS and variables where increasing values are
indicative of more sensitive catchments. These include SH (r= -0.79) and SCL4 (r= -0.51).
H is obviously inversely related to %BS (r= -0.83) and the former tend to exhibit the same
magnitude of relationships with the variables detailed above as the latter but in the opposite
direction. The relationships between geology, soil sensitivity and soil critical load reveal the
degree of collinearity exhibited by the catchment dataset. The fact that high negative
correlations coefficients are exhibited between variables of diametrically opposed sensitivity
suggests that the data are characterised by a sensitivity gradient along which sites will tend
to vary.
The strongest relationships exhibited between the land cover variables and the soil variables
are those involving LC2 (arable) and LC6 (upland semi-natural grass/moor). LC2 is highly
correlated with Gl (r= -0.52), SH (r= -0.67), SL (r= 0.68), H+ (r= -0.63), BS (r= 0.658)
and SCL (r= 0.536). Correlation coefficients between these variables and SCL are of
172
'siCO
S_Wn OWI
l)b( o . n i •0.005
Alls 0.143 -0.106 " 0 706
Altm 0.3DT 0465 0.611 . . .
Sdep -0.118 -0.156 •0J83:
Ndep •0.144 0.044 •0061 =0.307 0.936 ^
Riln •0.141 •OJ33 ^•0.290 0.278 t . 0.814 0-719
G l 0.179 •-.,0.311: 0.096 -0.033 0.320 ÿ 0.111 0.161 0.245
C Î -0.037 •0.169 | p J ) 2 0.180 ■0051 0.099 0.182 -0.2)9 -0.227
C3 -0.020 -0041 -0.170 0046 -0.170 -0.157 -0.237 -0126 •0.516 •0.264
G4 0004 0026 ■0.134 -0188 -0.270 0.102 0.022 -0.021 0064 -0 108 •0.188
SH 0.161 0.206 0.060 0.294 0.633 0.150 UJ91 0.388 0.424 •0.197 •0.055 -0.344
SM •0.104 •0.212 00)5 •0.0)3 :-0,472 -0.051 -0066 •0)65 -0448 0427 0176 0012 •0.595
SL 0.085 0004 0.025 -0 201 -0464 -0.270 •0.355 0335 0 124 0.332 4)671 0413
Bare •0.0)2 0046 0.130 -0064 -0.055 0 120 0 144 -0077 0 084 0.204 0.174 -0.189 0.014 0.307
SCLl 0.145 0.12! -0085 -0241 •0.289 -0.113 -0.180 -0199 •0079 0.003 0001 0.422 -0.414 0.737 OJOO
SCU 0.043 0.024 •0.186 .0369 •0.428 -0.067 -0.17) -0.154 -0.161 -0.127 0^77 •0.450 0.394 ^;. 0219
SC U •0.128 -0226 0 059 0.059 •0.438 0006 0(M9 -0.358 •0.691 0.432 0091 -0.588 0.500 0.127 0.259 0 171
SCL4 0.023 0.098 0.239 0.395 -0047 00)6 0.159 0.635 -0.250 ÜJ72 4)540 4)411 0046 •0.2)9 4)220 4)692
SCL5 0.060 0.043 -0076 0001 0151 0.102 0.246 -0006 0.244 -0,229 0.082 0.244 -0.191 4)285 •0065 -0.205 4)085 4)303 00)5
H* 0.214 0.221 0.194 0.372 0628 0.1)2 0.188 0.427 •0.18) -0122 -0.249 0.734 4)604 -0603 4)040 4)409 4)358 -0.592 0615 0.202
%hs -0.159 -0.169 ■0.148 -a392 -0.720 -OOJI -0 204 0.19V ■0.379 0.176 0.127 0.302 •0.792 0A19 0.695 007) 0.508 0.411 0.554 -0.506 4)350 4)827
SCL -0.086 -0.134 0.042 -0049 ■OJ85 -0 039 -0.165 -0.193 -0.517 0.227 0273 0.269 -a598 0.485 0,677 0.192 0636 OJ60 0.743 -0.474 41642 •0.589 0656
LC l 0.041 0 099 0.155 -0 211 •0.027 -0 007 -0 114 0054 0.064 •0127 -0.179 0.202 ■0.184 41.114 0156 -0.131 0.102 0.271 4)178 0079 -0086 0179 0,232 003)
LC2 0.013 -0.195 0.024 -0.150 -0.566 -0452 -0.546 -0657 -0.494 0.257 0255 0 120 -0.686 0.523 0679 0 080 0.336 O J l) 0585 ■0.480 ■0.293 -0627 0.658 0536 0.142
LCJ •0023 -0.168 •0.008 -0.098 -0J59 •0382 •0.379 -0.442 •0072 0.137 0059 0170 -0151 0.186 0248 01)7 0.224 0169 0.129 0.063 4).277 4)127 0.192 0.220 -0080 0J74
LC4 •0.138 -0.174 0067 ■0.1)2 -0.438', -0165 -0239 •0.271 ■0.063 0.0)8 0.060 0025 ■0J06 0J69 0.208 0.024 0.075 0.002 0.197 •0.067 4)288 -0.152 0.289 0.213 0014 02)3 0.419
LC5 0.168 0.087 0.218 0.564 4 0226 0J61 0.349 0.512 -0.323 -0.178 0.611 •0.309 ■0.486 4)224 •0392 4)509 0,489 0.300 ^ 0.432 4)494 •0,485 4)063 -0.521 4)208 -0.374
LC6 0.0)7 0.167 -0.099 0200 0.465 0.405 0.492 0.290 0043 -0.215 0.690 4)391 4)587 4)305 4)395 41.386 O J ll 0 .1 2 4 ^ 0.572 4)564 4).4 (M ^ 4)341 4)660 4)338 4)313 0<W
Area S_kn D ill Alls Alim Sdep Ndep Kain C l G4 SH SM SL SCLl SCLl SC U SC U SCL5 H * %hS SCL LC l L C l LC3 LC4 LCS
L’oaelation coefficients w ith p - values 5Ü.Ü5 arc shaded. These are signilicant at the 9596 level
Table 6.4: Matrix of Pearson product-moment correlations for 31 transformed catchment variables In = 7Q).
similar magnitude but in opposite directions. High sensitivity is thus associated with upland
moor while low arable land cover areas exhibit low sensitivity. Similarly, maximum catchment
altitude is positively correlated with variables representing greater sensitivity including SH
(r = 0.633) and H (r = 0.63) indicating an association between high sensitivity and
increasing altitude. Other relationships which may be expected intuitively include that
between N deposition and S deposition {r = 0.93). These are both highly correlated with
rainfall (r = 0.72 and 0.81, respectively) a consequence of the amount of wet deposition
relative to dry deposition in the areas from which sites were selected.
The correlation analysis has enabled an exploratory analysis of the structure of the
catchment data. There are clear indications that these explanatory variables may be
characterised by a strong gradient of variation similar to those exhibited by the chemistry
data. This supposition is now explored using principal components analysis (PCA).
6.2.2.S Principal Component Analysis (PCA) - catchment attributes
PCA was undertaken on the catchment attribute dataset to examine patterns of variation
among the catchment variables, identify gradients along which they vary and illustrate these
graphically. Table 6.5 presents the eigenvalues for the first four axes together with the
variable loadings for each catchment attribute. The relationships are illustrated in Figure 6.4,
a correlation biplot, which shows the position of the vectors for the catchment attributes
relative to the first two PCA axes and to each other.
In comparison with the water chemistry data, where Axes 1 and 2 account for 81.4% of the
variance, the importance of each axis in terms of explaining the variation in catchment
attributes is much reduced (Axes 1 and 2 explain 42% of the data). This is to be expected
given the complex and diverse nature of the catchment dataset. Unlike the water chemistry
174
description of variables)
1
PCA Axes
2 3 4
Eigenvalues .3163 .1047 .0906 .0700Cum. % variance 31.6 42.1 51.2 58.2
Variable loadings (correlations)Area -.1406 -.3497 .4185 .4449S_length -.2812 -.0452 .5397 .3152Dist -.0716 -.3160 -.0061 .6772Altm -.7276 -.1049 .0182 .4559Sdep -.3084 .8797 .1514 .0727Ndep -.4370 .7910 .0872 .2466Rain -.5390 .6973 .1511 -.2034G1 -.5994 -.1999 .4681 -.0958G2 .2412 .1397 -.1276 .6162G3 .2714 -.0135 -.4910 -.0764G4 .2723 .1370 .5560 -.2215SH -.8602 -.1154 -.0467 .0467SM .6811 .2026 -.2535 .2765SL .7956 -.0777 .3330 .1310Bare .1483 .0917 .3474 .3049SCLl .5313 .0193 .5668 .0720SCL2 .4707 .0448 .4762 -.1635SCL3 .7557 .2900 -.3116 .3314SCL4 -.6251 -.3792 .1761 -.0626SCL5 -.3871 .0327 .0719 -.1353H+ -.7905 -.3063 -.0780 .0962%BS .8398 .2338 .1739 -.1015SCL .7724 .2126 .0822 .2315Lei .1091 -.0895 .4490 -.1984Lc2 .8411 -.2439 -.0507 .0764Lc3 .3785 -.4330 -.0022 -.0680Lc4 .3847 -.1995 -.1450 -.1188Lc5 -.6987 .0161 .1324 .1896Lc6 -.7050 .2199 -.1947 .0426
are all negatively correlated data, there is not a single dominant gradient along which the
data vary. The highest variable loadings on axis 1 (and consequently the highest correlations
with that axis) are exhibited by the sensitivity classes SL and SH, the soil critical load
classes SCL3 and SCL4, %BS, H'", LC2, LC5 and LC6 and Altm. The soil and geology
variables associated with low sensitivity have high positive loadings whereas the converse
is true for those indicating high sensitivity. Altm, LC5 and LC6 with Axis 1 while LC2 is
positively correlated. Thus the principal gradient appears to relate to catchment sensitivity.
Sites with high positive scores along Axis 1 are characterised by high values for variables
representing low sensitivity and low values for variables representing low sensitivity. The
175
converse is true for sites with high negative scores along Axis 1. This is exemplified by
Table 6 . 6 which shows values for a number of the dominant Axis 1 variables for 5 sites
across the axis. From left to right across Axis 1 values for SL, %BS and LC2 increase
whereas Altm, SH, and LC6 tend to decrease.
Variable loadings on axis 2 are highest for S and N deposition along with rainfall. This axis
accounts for approximately 1 0 % of the variation in the catchment data indicating a
deposition and rainfall gradient. Sites mk75 and mk39 are at opposite ends of this gradient
and exhibit polarised values for N deposition (mk75, 1.58 keq H'" ha' y r \ mk39, 0.53keq
ha' yr ■'), S deposition (mk75,1.24keq ha' y r \ mk39, 0.38keq ha' yr and rainfall
(mk75, 3015mm y r \ mk39, 764mm yr'^). Axis 2 is not strongly associated with altitude
suggesting deposition and rainfall distribution along this axis is driven more by wider spatial
patterns than by altitude. This is possible due to the resolution at which these variables
mapped. This may be a result of the resolution at which these variables are mapped {i.e.
they do not relate to the actual catchment whereas altitude does).
Axis 3, which explains a similar level of the variance has high loadings for G4 (% coverage
of limestone) and SCLl (% coverage of soils where carbonates are the dominant mineral).
Axis 4 appears to bring out variation in distance from sea and G2. This component accounts
for approximately 7% of the variation in catchment data.
Figure 6.4 shows the correlation biplot with the sites superimposed onto the vectors for the
catchment attributes. This demonstrates how H'", %BS, LC5, LC2, SH and SCL are
characterised by long vectors which describe acute angles with the first PCA axis illustrating
the strength of the correlations with this axis. The second axis is also clearly discernable and
shows the dominance of S and N deposition together with rainfall previously indicated by
their high variable loadings (Table 6.5).
176
Figure 6.4: PCA correlation biplot of Phase 2 water chemistry (plotted using CALIBRATE - Juggins and Ter Braak, 1993) - vectors have been multiplied by three for clarity. (For key, see Table 6.3)
3.0
SoII
CMc /3
<<
2.0
1.0
0.0
- 1.0 —
- 2.0
InkTl \
P Omk75
Omk74O lo CD'
LC6
mk64 \ \ \ °O \ \ \
^ 0 ° O Q \ \
-LC5
- ‘ - . T n T " ' w ;
G L - ' - O O / '
Omk47SCL3
mklO mlloiO SM O
%BS
\\\
SCL4o f *"0:3+
G2 G4 ^-O
S O Jl o_ — -O- — *•5 ^ ^ ^ - " " ?--SCLJ ^MR03-
% LC4mk30 \
O Q\LC2
mk430Area ; o o o O o ^ \ L C 3
mkl4 Omkl5
mk39
-3 .0 - 2.0 - 1.0 0.0 1.0 2.0 3.0
PCA Axis 1 = 0 .3 1 6
Table 6 .6 : Values for dominant catchment variables for 5 sites along PCA Axis 1.
Alt(m) SH SL H" %BS LC2 LC6
mkOl 30 0.0 58.2 2.9 76.2 72.9 0.1
m k ll 90 0.0 78.0 6.4 52.1 78.6 0.0
mk30 470 85.4 0.0 10.8 19.7 50.5 10.1
mk70 488 100.0 0.0 13.4 22.7 0.0 3.2
mk25 933 100.0 0.0 24.3 6.2 0.2 58.0
A ltm = m axim um catchment a ltitude, S H = % high soil sensitivity, S L = % lo w soil sensitiv ity, H * = hydrogen ion concentration
in soil w ater, % B S = % base saturation, L C 2 = % cover o f agricultural grass and arable, L C 6 = % cover o f upland sem i-natural
grass/moor.
Thus PCA of the catchment attributes indicates that the primary axis of variation represents
177
a gradient along which catchment sensitivity varies. Given that soil mineralogy is derived
from weathering of primary parent materials (Creasey etal., 1986), where soils are of local
origin it is to be expected that sites dominated by sensitive soils should be highly correlated
with sensitive geology. However, following glacial and periglacial reworking it is possible for
sensitive soils to overlay non-sensitive geology and vice versa (Norton, 1980). Indeed the
surface water sensitivity classes defined by Hornung etal., (1995) included classes which
encompassed these circumstances. These soil/geology interactions are discussed further
in Section 6.3.4.
The following sections describe the results of direct gradient analyses on the chemistry and
catchment data. This seeks to establish whether the sensitivity gradient identified in PCA on
the chemistry data can be explained by the variation in catchment sensitivity as discussed
here. These subsequent analyses also address the collinearity that is likely to exist in both
datasets.
6.3 Direct gradient analysis - chemistry and catchment variables
In this section the direct gradient analysis techniques discussed in Chapter 4 are employed
to examine the degree to which variation in the chemical composition of surface waters can
be explained by external independent data. As in Chapter 5, the constrained ordination
method redundancy analysis (PDA) is used to quantify these relationships. Initially 29
catchment attributes (see Table 6.7) are used in these analyses. However, Chapter 4
identifies a number of other attributes, primarily soil and land cover classifications at different
resolutions to those used here. The 29 variables used in this section have been pre-selected
and the procedures used are described in Appendix 6.2. The aim of undertaking PDA using
all 29 catchment attributes is to explore holistically the catchment-chemistry relationships.
A forward selection procedure (see Chapter 4) is subsequently run to extract only those
178
catchment variables which are significant in explaining water chemistry variation.
6.3.1 Redundancy Analysis (PDA) on full catchment and chemistry datasets
Table 6.7 shows a summary of the PDA using the 29 catchment variables (Section 6.2.2.1)
and 16 water chemistry determinands (Section 6.2.1.1) that comprise the model calibration
dataset.
The sum of all canonical eigenvalues (E^) is 0.840 {i.e. all the catchment variables explain
84% of the variation in the water chemistry). Approximately 70% of the chemical variance
can be explained by the first two axes suggesting that it is responding primarily to catchment
variation along two gradients. This echoes the results of PCA analyses both on the
chemistry and catchment datasets. With regard to the water chemistry, those determinands
with the highest Axis 1 variable scores (e.g. conductivity, Mg '", Ca^ , NO ' and DCL) are
the same as those which dominate PCA axis 1 (Table 6.2). Similarly, PDA Axis 2 exhibits
high loadings for Na"", Cl' and Abs-250 for both the direct and indirect ordinations. The
structure of the water chemistry in response to the linear constraints of the catchment
attributes is, with the exception of reduced loadings on Axis 1 for the aluminium species,
very similar to that produced by the unconstrained PCA analysis. This similarity results from
the high level of response variance explained by the catchment attributes. Within the PDA
model only 16% of the statistical variation among the chemistry determinands cannot be
accounted for by the catchment variables. For each individual chemistry variable Table 6.7
presents values for the variance explained by the catchment data. This is expressed
cumulatively for Axis 1 and 2 and the PDA model as a whole. For most determinands the
latter is above 80% indicating that the strength of the relationship between catchment
179
Table 6.7: Results of RDA on chemistry and catchment variables (n=78)
Axes 1 2 3 4 Total variance
Eigenvalues (k) .543 .177 .052 .034 1.00
C um ulative percentage variance of;response data 54.3 72.0 77.1 80.5fitted relationship (Le. 64.6 85.6 91.8 95.9
Sum of a ll canonical eigenvalues .840Response variable scores (adjusted for variance)
cumulative variance explainedAxis 1 Axis 2 Total
pH .6520 -.4107 .4030 .1134 42.51 59.37 78.45A ik .7960 -.2506 .3058 .1396 63.37 69.65 83.66Cond .9369 .2301 .0899 .0343 87.79 93.89 95.30Na* .7156 .5494 .1696 -.2715 51.21 81.39 93.27K" .8172 .2620 -.3381 .0005 66.79 73.65 87.97M g:* .8932 .1376 .1453 .0245 79.78 81.67 91.01Ca:* .9239 -.0316 .0919 .1872 85.36 85.46 91.73Cl .7104 .5907 .1061 -.2657 50.47 85.36 95.27N O ; .8784 -.0685 -.3086 .0108 77.16 77.63 90.98S O / .8191 .2126 -.3892 .2392 67.09 71.61 95.56A l-T M -.5742 .5532 .1053 1964 32.97 63.57 70.44A l-N L -.6595 .5458 .0454 .0474 43.49 73.28 78.49
A l-L -.2086 .4447 .0826 .4038 4.35 24.13 48.90Abs-250 -.2746 .7745 .1019 .0232 7.45 67.52 75.12
H C L (passive) .7539 -.1264 .4265 .0453 56.83 58.43 90.28D C L (passive) .9156 -.0858 .1323 .1204 83.83 84.56 91.41
Biplot scores o f explanatory variables V IF
Area -.1212 -.0406 -.1930 .2013 15.5S_length -.3053 -.0506 -.1540 .1131 14.9Dist -.0924 -.5106 -.5029 .2910 8.2Altm -.6388 -.5729 -.0783 .0540 8.4Sdep -.4139 -.1866 .2116 -.0955 56.3
Ndep -.4978 -.2357 .1076 -.1184 54.0
Rain -.6379 -.1388 .4416 -.2074 30.2
G1 -.6067 .0953 -.2784 .1798 4.1
G2 .2675 -.1350 -.1765 .0920 3.5G3 .3029 -.0369 .3518 -.1017 4.0G4 .2803 -.0536 .1751 .3943 2.6SH -.7270 .0251 .1373 -.0275 7.6SM .6006 .0411 -.0545 -.2086 5.6SL .7590 -.0745 -.1243 .2170 7.0Bare .0758 -.0796 -.2467 .3102 1.9SC Ll .4981 -.0414 .0025 .2018 7.2SCL2 .4179 .1966 -.0760 .1318 2.5SCL3 .7185 -.1504 .1437 -.1007 22.1SCL4 -.5293 -.0684 -.1618 .1994 5.2SCL5 -.2739 .1866 -.1181 .1398 5.9H* -.7416 .0119 -.0640 .2061 5.8
%BS .6878 .0983 .0451 -.0779 11.2
SCL .6964 -.2735 .2635 .0681 22.8
LC l .0215 -.2704 -.3244 -.0541 2.6
LC2 .8617 -.0302 -.2209 -.0865 6.6
LC3 .3959 .2937 -.1988 -.0648 2.8
LC4 .1989 .3956 -.1908 .1539 2.8
LC5 -.6290 -.1369 -.0328 -.0799 3.4
LC6 -.6603 .0218 .3761 -.0403 4.3
(V IF = Variance Inflation Factor)
180
attribute and chemical response applies to most individual determinands. The exception is
labile aluminium where the RDA model explains only 49% of the variation. This variable is
most associated with RDA Axis 4, which is not strongly related to any of the catchment
variables. It is possible that the catchment processes which most influence freshwater
concentrations of labile aluminium are not represented by the catchment data used here.
Variation of specific soil properties such as hydraulic conductivity (Bauer and Feger, 1992)
and the soluble soil contents of soil solutions (Bache, 1985) which influence aluminium
mobilisation may not be adequately represented by the Phase 2 soil data.
The biplot scores for the explanatory variables are also similar to those produced by PCA
(Table 6.5). RDA Axis 1, which explains 54% of the variance in the chemistry data, is
dominated by altitude, rainfall, G1, SH, SL, SCL3, SCL, H"", %BS LC2, LC5 and LC6 . High
Axis 1 scores are associated with increasing proportions of low sensitivity soils and arable
land cover in the catchments and higher base saturation and critical load values. Low Axis
1 scores are associated with high proportions of granite and acid igneous rocks, highly
sensitive soils, moorland (LC5 and LC6 ) together with higher values for soil water H
concentration. Maximum catchment altitude and rainfall display inverse relationships with the
first axis. This primary axis appears to reflect the same sensitivity gradient within the
catchment dataset, based on soil, geology and land cover type as well as altitude and
rainfall, observed in the PCA. These attributes, which dominate Axis 1, are driving the
primary gradient of variation within the RDA model and as such are responsible for much
of the variation observed among the chemistry determinands. Approximately 84% of the
variation in DCL can be explained by Axis 1.
The second RDA axis accounts for 18% of the variance of the response data. Distance from
sea, altitude and LC4 (percent coverage of coniferous forest) exhibit the highest biplot
scores. With Na , Cl', AL-NL and Abs-250 having the highest variable loadings among the
181
chemistry data, Axis 2 appears to represent a sea-salt and organic acidity gradient. Peat
depths in low lying catchments close to the west coast of Scotland are greater than in
upland catchments further inland (Allott et al., 1995c). This may explain the positive
relationship between Abs-250 and the sea-salt concentration as these are inversely related
to distance from sea although, in Scotland, all freshwaters have a significant marine
component (Harriman and Pugh, 1994). Coniferous forestry (LC4) has a positive relationship
with Al-TM and Abs-250. Increased concentrations of toxic aluminium have been observed
in forested catchments in Scotland (Harriman and Morrison, 1982; Harriman eta!., 1995c),
the result of increased scavenging of sulphate which mobilises aluminium in the catchment.
The third and fourth RDA axes explain 5% and 4% of the response variance respectively.
The catchment variables closely associated with Axis 3 inciude distance from sea, rainfail,
and, to a lesser extent, upland moorland (LC6 ). HCL and pH exhibit the highest correiations
with this axis among the chemistry determinands. Axis 4 is not characterised by strong
relationships with any of the catchment variables. Unrestricted Monte Carlo permutation tests
on each of the four axis revealed that all were statistically significant at the 95% level.
Figure 6.5 illustrates the catchment/chemistry relationships quantified in Table 6.7. Most of
the solid vectors, representing the catchment attributes, describe acute angles with Axis 1.
This illustrates the importance of this axis in terms of interpreting catchment-chemistry
relationships, particularly those determinands most strongly associated with this primary axis
(e.g. Ca^ and DCL). The vectors representing LC2 (arable land cover), SL (low sensitivity)
and %BS are the longest of those which increase positively along Axis 1. Sites with high
values for these variables are characterised by high criticai loads and, as such, are not
sensitive to acidification. Conversely, the longest vectors increasing negatively along Axis
1 are exhibited by H"", SH (high sensitivity), G1 (the most sensitive geology class), rainfall
and maximum altitude. Sites which plot towards the arrowheads of these vectors will have
182
Figure 6.5. RDA biplot of chemistry and catchment data showing water chemistry (solid vectors) and catchment attributes (dashed vectors). See Tables 6 . 1 (chemistry) and 6 . 3
(catchments) for key
0.90
A b s -2 5 00.75 —
0.60 — Al—tm A l-n l K No
0.45 — A l-I LC4
0.30 — LC3Cc nd
II SCL2CN(AX<<Û
0.15 — Mg%BS
SH..........-fce2 —~ ^ - * “ C a
0.00 —SCL4
LC5 BareRain-0.15 —
Ndep AikLCl SCL-0.30 —
-0.45Dist
Altm
-0.60
-0.5 0.0 0.5 1.01.0
RDA Axis 1 = 0.543
low critical loads and more likely to be sensitive to acidification.
Comparison with the Phase 1 analysis reveals that the water chemistry responds in a similar
way to different catchment attributes. A gradient representing sensitivity to acidification has
been identified during both Phase 1 (Figure 5.1) and Phase 2 analyses. An analogous
sensitivity gradient also characterises the catchment variables in both the Phase 1 and
Phase 2 RDA biplots. However, only 51% of the chemistry variance was explained by
catchment attributes during the Phase 1 analysis, whereas this figure rises to 84% for Phase
2. The chemistry determinands used are similar for both datasets but there are substantial
differences between the two catchment variables in the two datasets. Phase 1 primarily
183
utilises data relating to the km^ in which the site is situated. The sites in the Phase 2 dataset
have been characterised according a number of attributes relating specifically to their
contributing catchments (see Chapter 4). Additionally, the Phase 2 analysis uses more
detailed catchment data, particularly in terms of geology and soil coverage.
These comparison suggest that the higher resolution, catchment specific data used in Phase
2 is more useful in terms of explaining variation among the water chemistry determinands.
Furthermore, with the dominant axis in this analysis representing a sensitivity gradient it is
likely that the Phase 2 catchment data can more accurately predict diatom critical load.
However, at this stage, the large number of catchment attributes included in the analysis is
likely to lead to spurious explanation. High variance inflation factors (VIPs) (Montgomery
and Peck, 1982) are exhibited by a number of the catchment variables, particularly among
rainfall and S and N deposition. A VIF above 20 is indicative that an explanatory variable
is very highly correlated with the other explanatory variables. As such, it does not contribute
uniquely to variation in the response variable(s) and its canonical coefficient is unstable and
should not be interpreted (ter Braak, 1986). The correlation matrix for the catchment data
(Figure 6.4) also shows that many of the variables are highly correlated and will not
independently account for a statistically significant amount of variation in the water chemistry
data. Consequently a forward selection procedure is employed to remove spurious variables
from the analyses and clarify the significant relationships between chemistry and catchment
attributes. These procedures are detailed in the next section.
6.3.2 Forward selection of catchment variables
Redundancy analysis incorporating forward selection (see Chapter 4) of the catchment
attributes is undertaken to assess which of these are statistically significant within the
framework of the RDA model, and which are most important with regard to explaining
184
variation in water chemistry. This procedure also eliminates spurious explanation by
removing collinearity from the analysis.
Following Monte Carlo permutation tests the RDA is run using only those variables which
have been selected on the basis of their significance at the 95% level. Each test is subject
to Bonferroni adjustment to avoid spurious significance (see Chapter 4). Table 6 .8 a lists
those variables identified as significant at the 95% level by the forward selection procedure.
The cumulative variance explained by selected variables, the number of permutations and
required significance level following Bonferroni adjustment are also included. Table 6 .8 b
summarises the RDA results. Axes 1 to 4 are significant at the 95% level as determined by
Monte Carlo permutation tests.
Table 6 .8 a : Catchment variables identified bv the forward selection procedure
Variable added cumulative variance explained by selected variables
number of permutations
(Bonferroni)requiredsignificance
significanceachieved
Land cover 2 (LC2) .41 99 .05 .01Maximum altitude (Altm) .52 99 .025 .01Soil Critical Load (SCL) .59 99 .0166 .01Rainfall (Rain) .63 999 0125 .001Distance from sea (Dist) .66 999 .01 .001Soil H* concentration (H*) .69 999 .008 .002
Variance explained by all variables .84
No other variables significant
Table 6 .8 b: RDA summarv using catchment variables identified bv forward selection
Axes 1 2 3 4 Total variance
Eigenvalues .508 .137 .030 .010 1.00
Cumulative % variance ofspecies data 50.8 64.5 67.5 68.5
Sum of all canonical eigenvalues .688
185
Figure 6 .6 , an RDA correlation biplot, illustrates the structure of the RDA model following
forward selection. This data reduction exercise has identified 6 catchment attributes as being
statistically significant in terms of explaining the response data. Soil water concentration
and rainfall are negatively correlated with Axis 1 (with biplot scores of -0.765 and -0.664
respectively) while the converse applies to SCL and LC2 (0.711 and 0.890 respectively).
Maximum altitude is negatively correlated both with Axis 1 and 2 (-0.675 and -0.593
respectively). Axis 2 is also associated with distance from sea (0.597).
Figure 6 .6 . RDA biplot of chemistry and catchment data showing water chemistry (solid vectors) and catchment parameters (dashed vectors) for soil map unit, the latter included following forward selection.
0.75 A b s - 2 5 0
0.50 — A l—nl I — tmNo
A l- I
0.25 — Cond
Mg
(N0.00 — ♦ C o
•DCL ■* LC2HCL
N 0 3
Aik-0.25 —
SCL
-0.50 —
AltmDist
-0.75
- 1.00 -0.75 -0.50 -0.25 0.00 0.25 0.50 0.75 1.00
RDA Axis 1 = 0.508
It is apparent that the chemical determinands closely associated with Axis 1 {i.e.
186
conductivity, K , Mg^ , Ca "", NO^, SO /' and DCL) are being driven by H"", rainfall, SCL, LC2
and maximum altitude, which are considered, in turn, below. However, the RDA model is
simply quantifying the statistical explanation within empirically derived relationships. Clearly
this approach does not involve the processes which operate within a catchment system and
which are used to paramaterise dynamic models (e.g. Christopherson et al., 1982; Cosby
etal., 1985b, 1990). However, dynamic modelling involving the quantification of process and
process rate require substantial amounts of input data. These are not available at a national
scale and, as such, are beyond the remit of this empirical model. The soil and land cover
explanatory variables selected are, in effect, surrogates for processes operating within the
catchment.
With regard to land cover, although LC2 (% cover of agricultural grass and arable land) does
not explicitly incorporate the processes which act upon surface and sub-surface water
passing through catchments in arable areas, it can be used to establish an empirical
relationship between the amount of arable land within a catchment and the nature of the
surface water chemistry in that catchment. Thus the strong positive relationships exhibited
here between the amount of arable land in a catchment and ionic concentrations and critical
load represent the processes by which agricultural practices ameliorate surface water acidity.
One of the most important of these is the addition of lime to counteract natural soil acidity.
This increases base saturation and replenishes the acid neutralizing capacity (ANC) of the
catchment, buffering the surface waters (Porcella et al., 1989). Experimental work has
confirmed this effect (Homung et al., 1986, 1990a). The amount of arable land within a
catchment is thus used as a surrogate for the effect of agricultural liming although this is not
quantified in terms of the elevated levels of carbonate minerals introduced into the soils.
The processes operating within soils which impact upon surface waters are represented here
by soil water H and SCL, a catchment soil critical load value derived from weighted
187
averaging of soil types. These variables are discussed in greater detail in Chapter 4. Given
the complex nature of processes operating at the soil/water interface it is clear that these
cannot be included explicitly if a predictive model is to have national applicability. These
processes are determined by the nature of the soil properties which in turn are dependent
to a great extent on the nature of the parent material. Surface water chemistry response to
variations in soil properties has been addressed by numerous authors (e.g. Reid etal., 1981 ;
Langan and Wilson, 1992; Matschullat et a!., 1992; - see Chapter 2 for more detailed
discussion). The critical load classification developed by Nilsson and Grennfelt (1988) and
Sverdrup and Warfvinge (1988) and modified by Hornung etal., (1994) divides soil materials
into classes based on the dominant primary weatherable materials of the parent materials.
A database held by the Macauley Land Use Research Institute (MLURI) ascribes a soil
critical load value to each soil map unit (Langan, pers. comm., see Chapter 4). The weighted
catchment soil critical load used in Phase 2 thus provides a surrogate for variations in the
properties of soil parent materials.
The second statistically significant soil variable (soil water H'" concentration) is the same as
soil pH. Soil pH is a reflection of the availability of free carbonates, the amount of
weatherable silicates and the base saturation of a system (Hornung, et al., 1990b). Soils
with high pH will tend to buffer incoming acidic precipitation and, as such, a strong positive
relationship may be expected between soils with low pH and the acidity of surface waters.
This is borne out by numerous studies (e.g. Schnoor and Stumm, 1985; Reuss etal., 1987;
Hooper et al., 1990). The weighted soil water H"" concentration variable used in Phase 2 is
based on soil pH values determined for each soil series throughout Scotland (Langan and
Wilson, 1991). As a consequence the Phase 2 analysis relates a representative soil pH to
surface water chemistry rather than establishing a direct relationship with soil data measured
in the catchment, an approach which has been adopted previously for regional studies
(Langan and Wilson, 1992; Hornung etal., 1995a).
188
The forward selection shows that rainfall has statistically significant explanatory role in
surface water chemistry variation. Sites with high annual rainfall values are characterised
by more sensitive water chemistry. This may be due, in part, to the fact that rainfall is higher
in upland areas which in turn are characterised by poorly buffered soils. However, a more
direct consequence of high rainfall levels is the removal of base cations and carbonates from
the soil through leaching (UKAWRG, 1986). Where soils and geology are of low base status
this may lead to reduced levels of base cation replenishment and a loss of buffering
capacity.
Distance form sea also contributes significantly to the variation in water chemistry. This is
based on its association with the second RDA axis (Table 6.7). As discussed, this primarily
represents a sea-salt gradient.
Although maximum catchment altitude offers significant explanation of variation in water
chemistry, the relationship is unlikely to be one based directly on processes operating across
a range of altitudes. Sites at high altitude have lower critical loads and are more sensitive
as evidenced by the obtuse vectors between the vectors for Altm and DCL in Figure 6 .6 .
However, a causal link is not suggested. As a result of the dominance in Scotland, at high
altitude of, base-poor, slow weathering soil parent materials, upland catchments are likely
to be characterised by acid soils with low to moderate buffering capacity (Hornung et al.,
1990b; UKAWRG, 1986). Where soils in catchments are similar it may be that the
catchments at higher altitudes are more sensitive as a result of the accumulated leaching
of base cations. This, in turn, is a consequence of the elevated levels of acid deposition
occurring at high altitudes due to increased rainfall levels and the effect of cap clouds
(Section 2.5.2 and 2.5.3). Although altitude may not directly explain variation in water
chemistry it is shown to be statistically significant in this context. Clearly this would not be
the case in uplands formed from limestone geology (e.g. the Dolomites).
189
A number of variables have been identified which, in statistical terms, significantly contribute
to variation in surface water chemistry in the Phase 2 calibration dataset. However, the use
of RDA with forward selection is primarily a data reduction exercise. It does not simply reject
variables which are not significant, rather it precludes the selection of variables which are
collinear with variables already selected. Thus, %BS may not be selected, even if it is
statistically significant, because H"" is selected preferentially and these two variables are
highly collinear. To illustrate this point a second RDA with forward selection has been
undertaken where variable choice is subjective and not based on maximum explanation. Six
catchment variables were selected, LC2, LC5, %BS, SCL, Altm and rainfall. Each variable
achieved the required Bonferroni adjusted significance level. These accounted for 64% of
the variation in water chemistry compared with 69% when forward selection is run
systematically (Table 6 .8 b). This shows that, using CANOCO, variables are selected on an
algorithmic basis and that variables omitted may still be significant.
While the variance in the water chemistry explained by the full catchment dataset is 84%,
the reduced catchment dataset following fon^/ard selection accounts for approximately 69%,
a difference of only 15%. By eliminating collinearity and reducing spurious explanation it is
therefore possible, using six catchment variables, to explain approximately 70% of the
variation in water chemistry. Furthermore these variables account for 51% of chemistry
variation along the sensitivity gradient represented by Axis 1 compared with 57% without
forward selection. This Axis is closely associated with diatom critical load (DCL). The next
stage in model development is to assess which are most important in explaining variation
in DCL
6.3.3 Redundancy analysis (RDA) using Diatom critical load (DCL) as a single response variable
RDA using DCL as a single response variable and multiple explanatory variables is
190
equivalent to multiple regression (see Chapter 5). Similarly, the use of forward selection is
akin to stepwise regression. However, at this stage, the purpose is to explore relationships
between DCL and catchment attributes. The use of RDA here also enables direct
comparison with analyses on the full dataset. In Chapter 7 this approach is extended to
allow prediction of DCL using multiple regression techniques.
The RDA uses the same analytical options as previous analyses. The response data are
centred (it was not necessary to standardise these data given that there is only one unit of
measurement) and the explanatory scores are scaled as in a correlation biplot (ter Braak,
1987). Table 6.9 summarises the RDA with biplot scores along the first axis for the water
chemistry and the catchment attributes. The use of a single response variable means there
is only one constrained axis. An unrestricted Monte Carlo permutation test shows that this
axis is significant at the 95% level. All other chemistry determinands are passive.
Table 6.9: Results of an RDA on DCL and catchment attributes.
E ig e n v a lu e f o r 1st ax is .9 1 4
P e rc e n ta g e v a r ia n c e o f species d a ta 9 1 .4
S u m o f a l l c a n o n ic a l e ig en va lu es .9 1 4
c a tc h m e n t a t tr ib u te
b ip lo t scores
V I F V I F
Catchment area -.1381 15.5 SCLl 0.5052 7.2Stream length -.2989 14.9 SCL2 0.4368 2.5Distance from sea -.0997 8.2 SCL3 0.6937 22.1Maximum Altitude -.6127 8.5 SCL4 -0.5085 5.2S deposition -.3004 56.3 SCL5 -0.2398 5.9N deposition -.4139 54.0 H* Concentration -0.7491 5.8Rainfall -.5132 31.2 % Base saturation 0.7295 11.2G1 -.5853 4.1 SCL 0.7219 22.8G2 .2211 3.5 LCl 0.0832 2.6G3 .2882 4.0 LC2 0.7969 6.6G4 .4416 2.6 LC3 0.2816 2.8SH -.7379 7.6 LC4 0.1474 2.8SM .5718 5.6 LC5 0.5952 3.8SL .7405 7.0 LC6 -0.6476 4.3Bare .0329 1.9
191
The sum of all canonical values (which, given the sole response variable equates to Axis
1 and is the same as the cumulative response variable score for DCL in RDA using all water
chemistry variables - Table 6.7) is 0.914. Thus 91% of the variation in DCL is explained by
the 39 variables comprising the full catchment dataset. The attributes with the highest biplot
scores (positive and negative) include LC2 (0.797), (-0.749), SL (0.741), SH (-0.738),
%BS (0.730), SCL (0.722), SCL3 (0.694), LC6 (-0.648) and maximum altitude (-0.613).
These are similar to the scores obtained along RDA Axis 1 using the full chemistry dataset
(Table 6.7) and reflects the close association between DCL and the sensitivity gradient along
this axis. However, the fact that the catchment data account for such a high proportion of
DCL variation is, to a degree, due to the number of explanatory variables. Some of these
are collinear as evidenced by the high variance inflation factors (VIFs) exhibited by several
catchment and there may be a spurious level of explanation. Forward selection, with
Bonferroni adjusted Monte Carlo permutation testing, is employed to reduce the number of
variables to only those which are statistically significant (Table 6.10).
The five variables indicated by the fonA/ard selection procedure (LC2, SCL, G4, G1, H I
explain 81% of the variation in DCL, 10% less than when all variables are included. The
relationships between DCL and these catchment variables are further illustrated by
scatterplots (Figure 6.7). It should be noted that the data transformations discussed at the
beginning of this chapter remain applicable and as such, all relationships discussed
hereafter relate to the transformed variables. There are similarities with the results of the
RDA using forward selection on the full chemistry dataset (Table 6 .8 a). LC2, SCL and H*
are significant in both instances with LC2 selected first both times. Altitude and distance
from sea have been omitted here though while G1 and G4 (at either end of the geological
sensitivity spectrum) are included. It is possible that altitude and distance from sea are more
important in explaining other aspects of water chemistry (e.g. sea-salts). The results here
suggest that DCL can be explained and, within certain confidence limits, predicted by a
192
combination of soil, geology and landcover factors.
Table 6.10: Forward selection with DCL as a sole response variable
Variable added cumulative variance explained by
selected variables
number of permutations
Bonferronirequiredsignificance
significanceachieved
Lc2 .58 99 .05 .01SCL .69 99 .025 .01G4 .76 99 .0166 .01Gi .79 999 .0125 .004
.81 999 .01 .010
Variance explained by allvariables .91
No other variables significant
The inverse relationship between DCL and is the most straightfon/vard. As soil water H""
concentration decreases the base critical freshwater critical load increases. Conversely, as
the derived catchment soil critical load (SCL) increases DCL also increases. This echoes
work relating soil sensitivity to freshwater sensitivity (Langan and Wilson 1992; Hornung et
al., 1995a; Hall, 1995a) which has demonstrated that soil sensitivity is a precursor to
freshwater sensitivity. G4 (the least sensitive class) is completely absent from many sites.
Where present it varies positively with DCL showing that catchments rich in limestone are
well buffered and have high critical load values. Conversely, although the trend is more
complex, lower DCL values are found where G1 (granite and gneisses) dominate the
catchment geology. Catchments where the dominant land use is arable (LC2) tend to exhibit
the highest critical loads reflecting the increased buffering capacity afforded by agricultural
liming. The scatterplot of LC2 against DCL shows that sites are arranged into two clusters
exhibiting different forms of relationship. At high DCL sites the relationship with LC2 is linear.
A greater scatter is observed where LC2 is less dominant. This is likely to indicate the
influence of other variables.
193
Figure 6.7 Scatterplots showing DCL against variables selected by the forward selection procedure
O
3.0
s#
• •
0.0
d0 0
1.25
0.00
Ü
+X
6.0
1.0
1.650.15
g
2.0
• • >
0.00.15 1.65
D C L
The scatterplots showing DCL against the catchment variables indicated by forward selection
show that relationships with individual variables are not straightforward. Only exhibits a
simple linear relationship with DCL. It appears that each variable explains some part of DCL
variation. This is to be expected for variables measured as percentage coverage given there
will be sites where coverage is 0%. Thus, where arable land is present in relatively low
proportions it is likely that other catchment attributes will be more influential. The scatter of
194
sites when DCL is plotted against G1 suggests that sites with sensitive geology can be
characterised by both high and low critical load values. In the absence of other factors it is
possible that geological sensitivity may vary linearly with surface water sensitivity. However,
in this case, if some catchments characterised by high proportion of G1 also have varying
proportions of LC2 (which may increases buffering capacity) the DCL response to G1 is
unlikely to be linear. Similarly, catchments where sensitive geology is overlain by non
sensitive soils will be buffered to a greater degree than if sensitive soils and geology are
juxtaposed. This and the non-sensitive geology/sensitive-soil situation is incorporated into
the surface water sensitivity classification (see Table 4.6) which is based on the assumption
that base flow is primarily influenced by bedrock chemistry and mineralogy while peak flow
comprises a greater proportion of water which has been subject to chemical processes at
the soil\water interface (Hornung etal., 1995a). The implication here is that the interaction
between these various catchment attributes is likely to explain significant levels of DCL
variation. The next section uses (partial) redundancy analysis to quantify this interaction and
assess its importance within the context of a predictive model.
This section has identified a number of key catchment variables which substantively explain
variation in diatom critical load. This explanation is based on statistical techniques applied
to empirically derived data and, as such, do not relate directly to within catchment
processes. Additionally, although forward selection was employed to determine the statistical
significance of the catchment parameters, variables omitted may also be significant. This is
because the selection criterion for the procedure dictates that variables correlated with other
variables become non-significant once the latter have been selected. (In this way %BS is
not selected once H"" is included). Additionally, the moorland variables (LC5 and LOS) which
exhibit strong inverse relationships with arable land cover are not selected due to the initial
selection of LC2. This suggests that many of the catchment variables may be of value in
explaining DCL variation and that choice of these variables can be based on practical criteria
195
(e.g. accessibility and availability).
6.3.4 Variance partitioning
The use of (partial) redundancy analysis is discussed in Chapter 4. The purpose of this
analysis is to partition the variation of a response dataset into a number of independent
components (Borcard et al., 1992). Variance partitioning using (partial) RDA is more often
been used in an ecological context. It has illustrated, for example, the importance that
covariances between explanatory variables in accounting for variation in mountain plant
distribution (Birks, 1993), aquatic ecosystems (Letter and Birks, 1993) and fish species
composition (Albert and Bergstad, 1993). This technique has not been widely outside
ecology. Here, (partial) RDA is used to examine the covariances between catchment
attributes which govern surface water sensitivity as measured by diatom critical load.
It is acknowledged that no one single catchment attribute drives surface water sensitivity.
Combinations of soil and geology variables have been used previously to predict regional
sensitivity to surface water acidification in Wales (Hornung et ai., 1990b) and throughout
Great Britain (Hornung etal., 1995a). Similarly superimposing different types of land use on
similar lithologies can result in significant differences in surface water chemistry. Previous
authors have discussed the effect of catchment management on surface water acidity
(Hornung et al., 1990c; Bird et al., 1990a) and, more specifically, the ameliorating and
exacerbating effects of pasture improvement (Hornung et al., 1986) and forestry (Harriman
and Morrison, 1982). The high eigenvalues for covariances between components in the
analysis here emphasise the importance of these variable covariances in determining diatom
critical load. The magnitude of each eigenvalue indicate those covariances which are most
important. In this instance the catchment data are initially divided into four components, an
extrinsic component (comprising variables which do not represent processes operating
196
within the catchment), geology, soil and land use. The variables in each component are
listed below.
1. Extrinsic : Area, SJength, Dist, Alts, Altm, Ndep and Sdep.
2. Geology : G1, G2, G3 and G4.
3. Soil : SH, SM, SL, Bare, SCL1, SCL2, SCL3, SCL4, SCL5, SCL, H+ and %BS.
4. Land use : LC1, LC2, LC3, LC4, LC5, LC6 .
The fraction of variation in DCL explained by each of these components individually was
determined by undertaking four separate RDA’s. A forward selection was employed for each
component in tandem with a Monte Carlo Permutation Test (0kland and Eilertsen, 1994) to
remove those variables which do not significantly contribute to chemical variation. The
results of these analyses are shown in Table 6.11. In each case the eigenvalue for Axis 1
is equivalent to the sum of all canonical eigenvalues as there is only one response variable.
Table 6.11: Redundancv Analvses on Catchment Variable Components
Catchment variable component Variables selected and Cumulative variance explained
Extrinsic 0.460 Altm (35%), Rain (46%)Geology 0.524 01 (31%), 0 4 (52%)Soil 0.684 H+ (51%), SL (63%), SCL3 (68%)Land use 0.621 LC2 (58%), LC5 (62%)
Within the extrinsic component only maximum catchment altitude and rainfall are selected
while S and N deposition are omitted. The most and least sensitive geology classes (GI and
G4 respectively) are significant whereas the intermediate classes are not. The soil
component comprises H"", SL from the sensitivity classification and SCL3 from the soil critical
load classification. LC2 and LC5 represent the land cover data. The eigenvalues for each
197
component show that the soil (1 = 0.684) and land use (k = 0.621) components, when
analysed independently, explain variation in DCL better than geology (k = 0.524) and the
extrinsic component {k = 0.460).
The use of (partial) RDA requires that certain explanatory variables are used as covariables
so that their influence on the response data can be eliminated. In this way the proportion of
the response variance that is uniquely attributable to a specific component can be
determined, with the effects of the other components considered. For example, if the
extrinsic, soil and land cover components are used as covariables, the unique influence that
the soil component has on DCL can be quantified. Additionaily, using various combinations
of covariables, it is possible to establish how much of the DCL variance can be accounted
for by unique interactions between components. For example, to establish the variation in
DCL explained by a unique combination of soil and geology variables, the land cover and
extrinsic components are used as covariabies. In RDA the ’interaction’ is actualiy the extent
to which variation in, for example, geology covaries with variation in, for exampie soils and
is thus the geology caused covariation in soils. To quantify the soil/geology covariance
component the eigenvalues computed for the soil and geology components individually are
subtracted from that using soil and geology as active variables with iand cover and extrinsic
as covariables. Thus;
^soiflgeology ~ ^soilTlgeology " (^soil ^geology)
where k is the eigenvalue
n indicates the unique covariance between variables.
With four sets of explanatory variables there are, including covariance terms, fifteen
components. The level of DCL variation explained by these is subtracted from the total DCL
198
variance (unity in the case of RDA) to calculate the unexplained variance. Figure 6 . 8 is a
schematic representation of the explanatory variable components and all possible
covariances between them. Each numbered segment represents a unique combination of
catchment attributes based on the initial division of the data into four independent
components. For example, segment 11 represents the unique effect of the covariance
between soil and land use variables with the influence of the geology and extrinsic variables
accounted for by their use as covariables. The individual influences of soil and land use are
also removed. Table 6.12 shows the results of partitioning the DCL variance into 15
separate components. The each component and covariance term can be related to a
numbered segment in Figure 6 .8 . The eigenvalue for each component is also included and
these are related to the percentage of the variation explained in Table 6.12. Additionally,
active variables and covariables for each (partial) RDA, the eigenvalue and the percentage
variation in DCL explained by the component/covariance are shown. This is represented
graphically in the form of a bar chart (Figure 6.9) which breaks down the variation in DCL
according to the unique combinations of variables represented in Table 6.12.
From Figure 6.9 and Table 6.12 it is clear that combinations of components (or sets of
environmental variables) explain most of the variation in DCL. The covariance between all
four components explains 27.8% of DCL variation. The combination of soil, geology and
extrinsic components account for 1 2 .8 % while the soil, land cover and extrinsic combination
explains 10.4%. Other covariance components with important contributions include soil and
geology (6 %) and soil and land use (5.09%). Taken individually with all other components
partialled out, the components most important in explaining variation in DCL are soil (4.8%)
and geology (7.6%). Conversely the extrinsic (0.9%) and land use components (1.3%)
account for very little in terms of DCL variation when the influence of other components is
taken into account. The combination of land cover and extrinsic components explains 3.9%
of DCL variation. No other single covariance accounts for more than 0.5% of DCL variation
199
Figure 6 .8 : Schematic representation of the explanatory variable components and covariances used in (partial) RDA on significant catchment variables and DCL. AflB is the unigue covariation between A and B. ATIBnc between A, B and C etc..
LEGEND
1 = A2 = AnC3 = C4 = B5 = AnBnC6 = A nB nC nD7 = A nC nD8 = CoD9 = BnCnD 10= D 11=B nD 12= B13= AnBnD 14= AnD 15= BnC
D=EXTRINSIC
B=GEOLOGY
A=SOIL
C=LANDUSE
200
Table 6.12: Results of (partial) RDA (A=Soil. B=Geology. C=Land use. D=Extrinsic. AnBnc=Unique covariance between A.B and C.
Component Set Active Covariable Eigenvalue % DCL explained
A 1 A B, C & D 0.048 4.8%
B 12 B A, C & D 0.076 7.6%
C 3 C A, B & D 0.013 1.3%
D 10 D A, B & C 0.009 0.9%
AflC 2 A & C B & D 0.059 5.9%
criD 8 C & D A & B 0.039 3.9%
BflD 11 B & D A & C -0.003 -0.3%
ATIB 4 A & B C & D 0.06 6.0%
ATID 14 A & D B & C -0.001 -0.1%
b d c 15 B & C A & D 0.005 0.5%
BncTlD 9 B, C & D A -0.001 -0.1%
AflBnO 13 A, B & D C 0.005 0.5%
AflBDC 5 A, B & C D 0.104 10.4%
ATlCnD 7 A, C & D B 0.128 12.8%
AfiBncnD 6 A, B, C & D N/A 0.278 27.8%
Total Explained DCL Variance 81.9%
Total unexplained DCL variance 18.1%
Total DCL variance 100.0%
eigenvalues for individual components. Thus, to calculate the eigenvalue for the unique covariance between soil and geology;^''«iflgeology *“ ^soUDgeology " (^ s o il ^eeolocv)
Several (all involving the extrinsic component) exhibit small negative eigenvalues (see Table
6.12) although this is theoretically possible (Whittaker, 1984).
The explanation offered by component D (the extrinsic component comprising rainfall and
maximum altitude) is <1% when the influence of all other components is considered. Thus
rainfall and maximum altitude (rain/altm) explain very little in terms of DCL variation
independently of soil geology and land use. Similarly, the land use component explains only
201
Figure 6.9: Bar chart showing the results of (partial) RDA on catchment attributes and DCL.
27.8%
10.4%
ABD« BCD
5 10 15% variation in DCL explained
iSszA = Soil B = Soil C = Land use D = Extrinsic
AC = SoillXand use AD = SoilTlExtrinsicCD = Land useHExtrinsic EC = GeologyTlLand useBD = GeologyfExtrinsic BCD = GeologyTlLand useTExtrinsicAB = SoilDGeology ABD = SoiinGcologyTlExtrinsic
ABC = SoiinOeologyTlLand use BCD = SoilTILand useTExtrinsic ABCD = SoilAGeologyTILand useTExtrinsic UX = Unexplained variance
X indicates a unique covariance between comptinents. the effect of all other components having been partialled out of the analysis
1.3% of DCL variation once the effects of the other components are included. Unique
covariances between alt/rain and soil (X,ad = c.0 %) and alt/rain and geology (Igg = c.0 %) are
also negligible, as is the covariance between all three of these components ( . bd = 0 -5 %).
However, covariance between land use and rain/altm (ylco = 3.9%) accounts for nearly 4%
of the variation in DCL. Thus, while high rainfall levels are associated with high sensitivity
202
sites, where catchment land use incorporates an agricultural component the effect of the
high rainfall will be ameliorated. Both the soil and geology components have relatively high
eigenvalues (0.048 and 0.076 respectively) these indicating that, individually, these are
important explanatory components. The importance of the covariance between soils and
geology has been discussed previously and here, explains 6 % of the variation in DCL. This
is likely to be the result of soils and geology with different sensitivities juxtaposed at the
catchment scale. The covariances between the soil/geology/land use (k = 0.104) and
soil/land use/extrinsic {k = 0.128) components explain 10.4% and 12.8% of the variation in
DCL respectively. The most important combination is that involving all four components
which explains accounts for almost 28% of DCL variation. The complexity of these statistical
relationships reflects the complex nature of surface water response to varying catchment
characteristics. These covariances cannot be explicitly incorporated into a predictive model
but their importance is implicit by virtue of the variables selected.. This is discussed further
in Section 6.5.
6.4 Analysis of a reduced dataset of more sensitive sites (Ca '^<400peq 1 )
The development of a predictive model has hitherto been based on response data which
occur across a wide gradient of chemical variation. Thus it has been possible to assess
which catchment attributes might be used to predict surface water chemistry (and,
specifically, diatom critical load) across a broad range of surface water conditions. However,
as discussed in Chapter 5, it is of interest to establish whether the relationships identified
for this broad gradient are the same when the chemical variation is confined to poorly
buffered, or more sensitive, sites. In Chapter 5 the sensitivity of a site was based not on
catchment attributes but on the water chemistry. The Critical Loads Advisory Group (CLAG)
national database shows that approximately 5% of sites where Ca^"^>200|ieq \' have
acidified according to critical loads models (C.Curtis, pers. comm.). This value was used to
203
delineate a sensitive subset of sites during Phase 1 anaiysis. However the Phase 2 model
calibration dataset has far fewer sites and only 23 with Ca^ concentration less than 200paq
\'\ This is not felt to be sufficient to establish meaningful chemistry/catchment relationships.
Further examination of the CLAG national database shows that less than 1% of sites with
Ca^"^400peq 1' have a critical load exceedance (Curtis, pers. comm.). Many of these are
located in the Pennines where deposition levels are much higher than in Scotland. Thus,
although critical load exceedance is not necessarily a guide to sensitivity, given the
deposition levels in Scotland it is felt that the Ca<400 threshold adequately defines potential
sensitivity.
Given that the sampling strategy resulted in a paucity of very low Ca "" sites this section
focuses on identifying gradients of variation in the chemistry dataset and how these are
accounted for by the catchment data. The lack of very sensitive sites inhibits the
development of a workable predictive model at the sensitive end of the spectrum although
regression analysis is undertaken on the sensitive subset for comparative purposes in
Chapter 7. Exploratory analyses here examine the chemistry and catchment data separately
using indirect gradient analysis techniques. Redundancy analyses will then examine the
extent that the variation in water chemistry for this reduced sensitivity gradient can be
explained by the catchment attributes. Variation in diatom critical load (DCL) will then be
related to changing catchment attributes using the same approach.
The Ca^^<400peq 1' dataset (sensitive subset) comprises 46 sites. The chemistry and
catchment data are transformed according to the same criteria as the 78 site dataset (full
dataset).
6.4.1 Exploratory analysis of response (water chemistry) variables
204
Summary statistics for water chemistry from the sensitive subset are presented in Appendix
6.3. The pattern whereby the gradients of variation are substantially reduced for most
determinands when compared with the full dataset, observed in the Phase 1 analysis of a
sensitive subset, is echoed here. Obviously the Ca " maximum has been reduced to less
than 400peq \'\ Most other determinands exhibit reduced means and maxima compared
with the full dataset. The exceptions are the aluminium species which have higher means
in the sensitive subset. These differences reflect the fact that very well buffered sites have
been omitted from the sensitivity subset as a result of the Ca "^<400p.eq 1' cut-off. The
reduction in the mean NOg' value is particularly dramatic (from 72peq 1' in the full dataset
to 5peq r ) and may be due to the omission of catchments containing relatively large areas
of arable land. Of the 46 sites in the sensitive subset, 40 contain less than 5% arable land.
Correlation coefficients calculated for the transformed chemistry determinands in the
sensitive subset and these are shown in a correlation matrix (Appendix 6.4). Those
coefficients significant at the 5% level are highlighted. There are fewer significant correiation
coefficients in the sensitive subset. The strength of the relationships between NOg' and other
determinands are particularly reduced as a result of the shortened NOg' gradient. Correlation
coefficients between Ca "" and the other determinands are also reduced but not to the same
extent as that witnessed during the Phase 1 analysis, a result of the higher cut-off value.
The results of a PCA undertaken on the sensitive dataset are shown in Table 6.13. As with
the PCA on the full data-set, both derived critical load values are made passive. Table 6.13
shows that the PCA is not characterised by a single dominant axis. Axis 1 and 2 both
account for similar levels of the variation in water chemistry (39% and 32% respectively).
The highest variable loadings on Axis 1 are exhibited by Na^ (0.7066), Cl' (0.7541), Abs-250
(0.9230) and the aluminium species (AI-TM, 0.8611; AI-NL, 0.7524; Al-L, 0.7524). This
indicates that the primary PCA axis 1 represents a sea-salt and organic acidity gradient. Axis
205
2 is dominated by alkalinity, conductivity, Ca "" and DCL (variable loadings of 0.7661,
0.8594, 0.8405, 0.9164 and 0.8599 respectively). This second axis represents an acid
sensitivity gradient. On Axis 3, SO/+ and NOg' have the highest variable loadings. The PCA
correlation biplot (Figure 6.10) illustrates the results presented in Table 6.13 and shows the
switch between the gradients represented by Axes 1 and 2 compared with the full dataset.
The omission of sites which are not susceptible to acidification {i.e., with Ca>400peq I' ) has
reduced the importance of the sensitivity gradient which was dominant following PCA of the
full data-set (X^O.578). Conversely the determinands which had the highest Axis 2 variable
loadings from the full dataset are those which are dominant along Axis 1 here although this
axis itself does not dominate the PCA. The reduced importance of the sensitivity gradient
following unconstrained ordination of the chemistry data is not surprising. The redundancy
analyses in the following section shows how the relationships between water chemistry
Ca '" <= 400uea 1'
1Axes
2 3 4
Eigenvalue .392 .324 .129 .051Cum % variance 39.2 71.6 84.4 89.6
Variable loadings (correlations)
pH -.6333 .6563 -.2660 -.2014Aik -.4673 .7661 -.2003 -.2706Cond .4684 .8594 -.0759 .0545Na+ .7066 .5659 -.2923 .2601K+ .6160 .2787 .5598 0285Mg^* .3557 .8405 -.1819 .0298Ca2+ -.1512 .9164 .0642 -.2681CI .7541 .5164 -.2283 .2550NQ,- -.1839 .4237 .7273 .2098s o /- .4049 .2541 .7919 -.2140Al-TM .8611 -.3677 -.1301 -.2221Al-NL .8304 -.4197 -.1248 -.1143Al-L .7524 -.1288 -.0312 -.4903Abs-250 .9230 -.0143 -.0716 .0608
HCL (passive) -.2567 .7250 -.4676 -.0352DCL (passive) -.3168 .8599 -.1990 -.1145
206
Figure 6.10: PCA plot of water chemistry determinands from sites where <=4Q0ueq 1'
3.0Co
DCL Mg CondAik
mk3HCL2.0 —
Nomk}6
mkSOCNd 1.0 —
mk40SO.
mk79mk210 mk53
v^b&-25.0...0.0 —
mk36
mk70 A i-I
mk46 : mk65- 1.0 — [—tm
Al—n!
mk67
- 2.0
- 2.0-3.0 - 1.0 0.0 1.0 2.0 3.0
PCA Axis 1 = 0.392
and the catchment attributes are affected by the altered gradients within the sensitive
subset. Additionally, analysis with DCL as a sole response variable shows how the reduced
sensitivity gradient has affected the potential for predicting DCL from catchment variables
where only sensitive sites are included. Figure 6.11a illustrates how the strong relationship
between DCL and Axis 1 revealed by PCA on the full dataset is not reproduced for the
sensitive subset. However, there is a sensitivity gradient associated with PCA Axis 2 which
exhibits a strong relationship with DCL (Figure 6.11b). The anomalous position of site mk40
in this plot is addressed in Section 6.2.1.2. The relationships between water chemistry and
catchment characteristics are discussed following exploratory analysis of the latter.
207
Figure 6 .11: Scatterplot of Diatom Critical Load (DCL) against PCA Axes scores for sites where Ca%400ueg \'\
a: DCL vs Axis 1 site scores
Omk40
mk39
mk45
. 2 0.0 -
mkS6
mk 7
Omk22
T ra n s fo rm e d d ia to m c r itic a l load
b: DCL vs Axis 2 site scores
mk56
Ofnk40
mk39
o inkAf)
mk46
T ran s fo rm ed d ia to m c r itic a l load
208
6.4.2 Exploratory data analysis - explanatory variables
Catchment attribute summary statistics are presented for the untransformed data from the
sensitive subset in Appendix 6.5. The omission of sites with high Ca "" concentrations has
increased the mean (617m) and minimum (172m) values for maximum catchment altitude
in the explanatory data-set in comparison with the full dataset (515m and 30m respectively).
Similarly mean rainfall (2158mm yr^) is also increased relative to the full dataset (1750mm
yr^). This suggests that sites most prone to acidification are more likely to be in upland
areas which will also be characterised by higher rainfall levels. These relationships are
discussed in greater detail in the following section. Mean values for % coverage of the more
sensitive soil (SH, SCL4 and SCL5) and geology types (G2) are higher for the sensitive
subset. Additionally the mean soil water concentration is higher for sites in the sensitive
subset while the converse is the case for % base saturation. Compared to the full dataset,
relatively fewer catchments are covered by arable land (LC2) while areas characterised by
moorland (LC5 and LC6 ) are more widespread. By screening out sites where Ca^^400peq
the catchment dataset is more heavily weighted towards poorly buffered moorland sites.
The implications of this in terms of the relationships between chemistry and catchment
variables are discussed in the following section.
The results of a PCA on the catchment data from the sensitive subset are summarised in
Table 6.14. Correlation coefficients between catchment variables are shown in a correlation
matrix in Appendix 6 .6 . The structure of the catchment dataset is further illustrated by a PCA
biplot (Figure 6.12). The eigenvalues for each of the first four axes are of similar magnitude
to those exhibited following PCA of the full dataset. However the structure of the explanatory
data in the sensitive subset appears to be less coherent than in the full dataset, in particular,
the sensitivity gradient identified in the latter is not fully reproduced in the sensitive subset.
The highest (positive and negative) variable loadings on Axis 1 are exhibited by SCL3 (soils
209
Table 6.14: Results of a PCA on transformed catchment attributes on sites with Ca "" <400eq 1'
1
PCA Axes
2 3 4
Eigenvalues .2696 .1315 .0985 .0808Cum. % varianceVariable loadings (correlations)
27.0 40.1 50.0 58.0
Area -.1806 .2174 -.1135 .5630S_length -.1039 .2085 .1587 .5609Dist -.0753 .7289 -.2150 .2856Alt(s) .0847 .4282 -.6010 .3543Alt(m) -.1606 .4130 -.3182 .7116Sdep .5254 -.0294 .5655 .0855Ndep .5247 .1225 .6122 .1777Rain .3379 -.2771 .3685 .0660G1 -.7701 .3998 .3555 -.1284G2 .2934 .6062 .1294 -.1270G3 .5568 -.5166 -.4587 -.0127G4 .0755 .1236 .0587 .0565SH -.5544 -.3826 -.3556 .2410SM .6213 .2152 .3917 -.1400SL .4446 .6410 .2041 -.0292Bare .2081 .5188 .0740 -.0828SCLl .3992 .5982 .1846 -.0249SCL3 .9287 -.0100 -.0956 .0765SCL4 -.6838 .5731 -.0985 -.2164SCL5 -.2483 -.2740 .5039 .3286H+eq -.4915 .2618 -.5171 .0324%BS .4601 .0724 .3897 -.3692SCL .7853 .2940 -.3566 -.0924Lei -.3982 .0374 .1203 .2633Lc2 .1130 -.0608 -.1748 -.3450Lc3 -.2632 .0915 -2986 -.5992Lc4 -.2057 .1711 -.0014 -.6289Lc5 -.3250 .2267 3960 .3052Lc6 .3789 -.2041 -.1598 -.2720
with granodiorite, greywacke, schist or gabbro parent materials, 0.9287), SOL (the composite
soil critical load value for each catchment based on applying weighted averaging to the
lower class limit for each SOL class, 0.7853), G1 (% coverage of granite and igneous rock, -
0.7701), SCL4 (soils from granite/gneiss parent materials, -0.6838 and SM (soils of medium
sensitivity with pH >4.5 and <5.5, base saturation >20% and < 60%, 0.6213). Thus although
G1, the most sensitive geology class, is inversely related to Axis 1, and soil critical load
(SOL) varies positively along this axis, it does not appear to represent a sensitivity gradient.
By omitting less sensitive sites the result is a dataset which comprises catchments
210
dominated by geology, soils and land cover types which provide a limited buffering capacity.
Variation in these catchments appears to be greatest among variables with medium
sensitivities (e.g. SM, SCL3 and SCL4). This may be a consequence of using percentage
data based on classifications where a high value for one class precludes a high value for
others. The consequence is that the polarity of variation in the full dataset which produces
the sensitivity gradient (section 6.3.1) is not apparent here. This applies not simply to the
primary axis but also to Axes 2 to 4, none of which appear to represent immediately
apparent gradients. The implications of this in terms of quantifying the relationships between
water chemistry and catchment attributes are discussed in the following section.
Figure 6.12: PCA correlation biplot of transformed catchment attributes on sites where Ca"+<400ueg I '
3.0
D ls t
SL2.0 — G 2S C L 4
B a re /
AltmCN1.0 — SCLo
Ndep
0.0 —
\xNSL C 6
SCL5 °
w•Bain
- 1.0 —
SH
G 3
mk58- 2.0
- 2.0 - 1.0 0.0 3.01.0 2.0-3.0
PCA Axis 1 = 0.270
211
6.4.3. Direct gradient analysis
6.4.3.1 Redundancy anaiysis (RDA) on sites where Ca^"<400|ieq l"*
The results of an RDA undertaken on the sensitive subset, using the same criteria as those
employed on the full dataset are summarised in Table 6.15. An RDA biplot (Figure 6.13)
illustrates the structure of the relationships between the catchment attributes, the chemistry
determinands and the canonical axis for the sensitive subset.
The sum of all canonical eigenvalues (LX) is 0.867. Thus for the sensitive subset, 87% of
the variation in water chemistry is explained by the catchment data. This is very similar to
that produced during RDA on the full dataset (84%). However, the eigenvalues for each
individual axis show that while Axis 1 dominates the structure of the RDA model on the full
dataset (À=0.543), the sensitive subset does not appear to be dominated by a single
gradient. Axis 1 (1=0.346) and Axis 2 (A^O.288) account for similar levels of variation in the
water chemistry data. This reflects the results obtained using PCA both on the response
data and the explanatory data (Sections 6.4.1 and 6.4.2 respectively). With regard to the
response variable scores (Table 6.15), when the chemistry data are constrained by linear
combination of the explanatory variables, the gradients identified during PCA of the former
remain in place. RDA Axis 1 is correlated with conductivity, Na", Cl' and Abs-250,
representing a sea-salt and organic acidity response gradient. Axis 2 is highly correlated with
pH, alkalinity, Ca "", and both critical load values and represents a sensitivity response
gradient. Table 6.15 also provides the cumulative percentage, by axis, for each determinand.
Axis 1 explains most of the variation in Na "", Cl' and Abs-250 (75%, 79% and 6 8 %
respectively) and very little for pH, alkalinity Ca^ and DCL (11%, 2%, 4% and <1%
respectively). However the cumulative explanation provided by Axes 1 and 2 for these
determinands reaches 73%, 71% and 75% respectively. Axis 3, which explains
212
Ca^'"<400uea 1'
Axes 1 2 3 4 Total variance
Eigenvalues .3463 .2879 .1228 .0389 1.00
Cumulative percentage variance ofspecies data 34.6 63.4 75.7 79.6species-environment correlations 40.0 73.2 87.4 91.8Sum of all canonical eigenvalues .867
Response variable scores (adjusted for variance) % explained
Axis 1 Axis 2 Axis 3
pH -.3301 .7832 -.2617 -.1703 10.89 72.73 85.83Aik -.1428 .8278 -.1754 -.2294 2.04 70.57 93.70Cond .7432 .6000 -.0693 .0436 55.23 91.28 94.46Na+ .8642 .2708 -.2888 .2315 74.68 82.02 97.55K+ .5868 .0720 .5379 .0934 34.43 34.97 76.33
.6411 .6250 -.1787 -.0019 41.10 80.16 95.50Ca + .1891 .8583 .0803 -.2627 3.58 77.24 88.41CI .8880 .1986 -.2278 .2295 78.85 82.79 96.82NQ,- .0024 .4366 .6899 .2303 0.00 19.06 87.99SO / .4552 .0674 .8036 -.2501 20.72 21.18 96.70Al-tm .5848 -.5869 -.1220 -.1950 34.20 68.65 78.65Al-nl .5555 -.6197 -.1080 -.1331 30.85 69.26 80.02Al-1 .5456 -.3425 -.0253 -.3177 29.77 41.50 64.18Abs-250 .8273 -.2928 -.0564 -.0660 68.44 77.02 87.02HCL .0246 .7433 -.4723 .0005 6.00 55.30 94.42DCL .0324 .8654 -.1954 -.1260 0.11 75.00 88.71
Biplot scores of explanatory variables V IF
Area -.1525 -.1265 .1689 .0517 42.03SJength -.2488 -.1955 .1146 .1839 40.08Dist -.5569 -.1119 .4811 -.4587 14.57Alt(m) -.7610 -.0790 .1703 -.1015 8.58Sdep -.2502 .0484 -.1289 .1463 208.34Ndep -.2877 .0677 .0343 .0970 177.20Rain -.3496 -.0657 -.5868 .2766 28.31G1 -.0780 -.6179 .4870 .1242 32.07G2 -.0152 .1783 .5302 -.3925 9.80G3 .1423 .6119 -.5123 .1743 29.51G4 -.1172 .2315 .2277 -.0679 2.86SH .0113 -.2856 -.2082 .0211 5.65SM .1922 .3264 .1531 .1371 14.93SL -.0870 .2697 .2325 .0094 21.80Bare -.1639 .1530 .3850 -.2493 5.80SCLl -.0484 .2797 .2405 .0815 11.19SCL3 .0034 .7653 -.0812 -.1582 56.80SCL4 -.2752 -.3546 .3761 -.1598 13.38SCL5 .2407 -.1710 .1805 .1520 12.60H+eq -.0701 -.2494 .1684 -.3396 7.17%BS .1554 .0983 -.1922 .1044 11.46SCL -.2630 .6533 -.1543 -.3055 56.29Lcl -.5123 -.4019 .0112 .2407 4.20Lc2 .1799 .4121 .0994 .1929 2.30Lc3 .2697 -.0391 .1441 .0256 4.83Lc4 .3302 -.2453 .2039 -.1909 6.20Lc5 -.2489 -.3218 .1253 .0863 2.87Lc6 .3017 .2715 -.1497 .1423 5.53
213
Figure 6.13; RDA biplot of chemistry and chemistry attributes for sites where Ca^'^<400ueq
DCL CaAik
SCL Mg CondG3
0.5 —
.SL' No0000csdII
(S
<<§
0.0 —LC3
Altm Dist
L-250SH/iC5 A I- I
SCL4LCl
-0.5 —
Al—tmG1 A l-n l
-0.50 -0.25-0.75 0.00 0.25 0.50 0.75 1.00- 1.00
R D A Axis 1 = 0.346
12% of the variation in water chemistry ()u=0.1228) is highly correlated with NOg’ and S O /’
replicating the suggested gradient observed following PCA on the chemistry determinands.
Closer examination of the explanatory bi-plot scores (Table 6.15) reveals that explanation
is not dominated by any single or set of catchment attributes along any axis. The highest
scores on Axis 1 are exhibited by maximum catchment altitude (Altm) and distance from sea
(dist). These are negatively correlated with Axis 1 and are thus inversely related to the sea-
salt/organic gradient suggested by the response variable scores. The explanatory variables
most clearly associated with Axis 2 (and, as a consequence, the sensitivity gradient) are G1,
G3 (% coverage basic and ultrabasic rocks), SCL3 and SCL. The RDA correlation biplot
(Figure 6.13) illustrates these relationships. Alt(m) and dist have long vectors which describe
214
acute angles with Axis 1. Equally, acute angles characterise the relationship between the
SCL3 and G1 vectors and Axis 2. The only high explanatory biplot score associated with
Axis 3 is that of rainfall (-0.5868) with which it is negatively correlated. Lower rainfall values
are thus associated with high NOg' and SO /' values. S and N deposition are not strongly
associated with any of the first four axes.
Unrestricted Monte Carlo permutation tests show that Axes 1 to 4 are significant at the 95%
level. However, it is clear from the number of high variance inflation factor values (VIFs) that
many of the explanatory variables are collinear and do not contribute significantly to the
variation in water chemistry. Therefore, in the next section, forward selection is used to
reduce the number of explanatory variables.
6.4.S.2 Forward selection of environmental variables
Table 6.16a shows the results of the forward selection procedure including the variables
which are significant and the level of explanation offered by each. The use of forward
selection reduces from 0.867 to 0.614, the latter explaining 25% less of the variation in
water chemistry than the former. This suggests that there was a substantial degree of
spurious explanation when all catchment variables were included in the model. The variables
which significantly account for variation are, in order of their selection, maximum catchment
altitude, SCL3, rainfall, S deposition (Sdep) and SCL4.
Table 6.16b provides a summary of the axes eigenvalues (Monte Carlo Permutation testing
showed that each of these are significant at the 95% level) together with axes scores for the
response variables and the selected explanatory variables. There are no substantial changes
in the scores for the response variables. The reduction in explanatory variables has not
215
caused major changes in the response structure of the RDA model. Of the explanatory
variables, Altm has the highest biplot score on Axis 1 (-0.8247) indicating that altitude is the
catchment parameter driving variation along this gradient. The high scores for Na^", Cl' and
Abs-250 suggests that once sites insensitive to acidification are omitted, the dominant axis
becomes one representing a sea-salt/organic gradient. The previously dominant sensitivity
to acidification gradient now occupies the second axis. Many variables important in driving
the sensitivity gradient apparent during analyses of the full dataset (e.g. soil water H*
concentration, % arable land cover) are no longer influential. With this gradient reduced,
altitude explains most of the chemical variation. Axis 2, which explains 21% of chemical
variation, exhibits a strong relationship with SCL3 suggesting that, for this sensitive subset,
the % coverage of soils formed from granodiorite, greywacke, schist or gabbro (with soil
critical loads between 1 and 2 keq H"" ha'\r'^) is the most important catchment attribute along
the (reduced) sensitivity gradient. The reduced influence of the most important explanatory
variables from the full dataset may be a result of a reduction in the variety of catchment
types in the sensitive subset. By omitted sites with Ca^00p.eq 1' catchments characterised
by non-sensitive soil, geology and land cover are likely to be screened out and sensitive
catchments will dominate the data. This is supported by Appendix 6.5, summary statistics
for the catchment attributes from the sensitive subset, which shows that mean values for all
’sensitive’ variables are higher than those for the full dataset (Table 6.3). If the catchment
variables are characterised gradients of variation that are too short their use as predictors
are limited. The implications of this for the application of the predictive model are discussed
further in Chapter 8 .
Axis 3 (explaining 9% of water chemistry variation) is associated with rainfall which appears
to be driving SO /' variation. In sensitive areas elevated acidity in rainfall are less likely to
be buffered by the catchment. Higher rainfall levels in such areas are liable to lead to
increased SO/' leached to the surface waters. Axis 4 explains only 1 % of the chemistry
216
Table 6.16a : Forward selection of catchment variables at sites where Ca <400ueq I
Variable added cumulative variance of selected variables
number of permutations
Bonferronirequiredsignificance
significanceachieved
Maximum altitude .21 99 .05 .01Soil Critical Load Class 3 .38 99 .025 .01Rainfall .50 99 .0166 .01S deposition .56 999 .0125 .001Soil Critical Load Class 4 .61 999 .01 .002
Variance explained by allvariables .87
No other variables significant
Table 6.16b: RDA summarv usina variables from forward selection at sites whereCa^^<400ueq 1'
Axes 1 2 3 4 Total variance
Eigenvalues .296 .214 .090 .012 1.00
Cumulative % variance ofspecies data 29.6 51.0 60.0 61.2
Sum of all canonical eigenvalues .614
Response variable scores (adjusted for variance) % explainedAxis 1 Axis 2 Axis 3
pH -.1675 -.8034 -.1424 -.0970 2.80 67.35 70.39Aik .0207 -.8124 -.0414 -.1866 0.04 66.04 69.71Cond .7801 -.3853 -.0371 .1152 60.86 75.71 77.23Na+ .8263 -.0981 -.3471 .0004 68.29 69.29 81.34K+ .5501 .1088 .4625 .0761 30.26 31.44 53.42Mg^* .6450 -.4990 -.0456 .1361 41.60 66.50 69.85Ca"+ .3610 -.7324 .1738 -.0258 10.03 66.67 69.77CI .8345 -.0155 -.2901 -.0174 69.64 69.67 78.73NQ,- .1107 -.2220 .4120 -.1242 1.23 6.16 24.67SO / .4449 .1134 .7623 -.0557 19.79 21.08 79.50Al-tm .4335 .4703 -.1342 -.1413 18.79 40.91 44.88Al-nl .4329 .5130 -.1366 -.1868 18.74 45.06 50.65Al-1 .3731 .2807 .0036 .0763 13.92 21.80 22.39Abs-250 .7449 .3312 -.0291 -.0755 55.49 66.46 67.11HCL .1207 -.7398 -.3793 .1710 1.46 56.19 82.39DCL .2093 -.8002 -.0395 .0544 4.38 68.57 69.25
Biplot scores of explanatory variables V IF
Alt(m) -.8247 -.0616 .2584 .1598 1.08Sdep -.2622 -.1107 -.1291 .4372 3.36Rain -.3912 -.0457 -.6951 .1981 2.93SCL3 .1310 -.8753 .0840 .3664 2.22SCL4 -.3447 .3689 .3794 -.7752 1.93
217
response and is highly correlated with SCL4 although none of the chemistry determinands
has a high variable score for this axis. Figure 6.14 illustrates the results of the analyses
using an RDA biplot. This confirms the importance of Alt(m) and SCL3 with regard to driving
variation along Axes 1 and 2 respectively.
Figure 6.14: RDA biplot using variables from forward selection at sites where Ca^~'<400ueq I'
0.60Al—nl
Al—tm0.45 —SCL4
A b s -2 5 00.30 — A I-I
0.15 — S 0 4
CNd 0.00 —
RainAltmIIN o(N
(A"x<<§
Sdep-0.15 —
i N 0 3
-0.30 —
Cond
-0.45 —Mg
-0.60 —
Co-0.75 — HÇLDCL
AikSC,L3
-0.90
- 0.6 -0.4 - 0.2 1.0■0.8 0.0 0.2 0.4 0.6 0.8- 1.0
RDA Axis 1 = 0.296
e.4.3.3 RDA using DCL as a single response variable
This section describes the results of RDA undertaken on the sensitive subset using DCL as
a sole response variable. This analysis seeks to establish whether the catchment attributes
218
which account for DCL variation at sensitive sites are the same as those for a much broader
chemical gradient. To eliminate spurious explanation fon/vard selection with Monte Carlo
permutation testing was employed with Bonferroni correction. Table 6.17 shows the results
of the RDA. A single response variable results in only one canonical axis.
for sites where Ca^ < 400ueq 1-1
Variable added cumulative variance of number of Bonferroni required achievedselected variables permutations significance significance
Soil Critical Load Class 3 .58 99 .05 .01Landcover Class 2 .68 99 .025 .01
Variance explained by allvariables .89
No other variables significant
Table 6.17 shows that 89% of the variance in DCL is explained by all the catchment
variables. However, given the number of explanatory variables it is probable that a
substantial proportion of this explanation is spurious. Following forward selection only two
of the catchment variables (SCL3 and LC2) were statistically significant at the 95% level.
The cumulative variance in DCL explained by the selected variables is 6 8 %. SCL3 is
selected initially and explains 58% while adding LC2 explains a further 10%. The single RDA
axis is thus dominated by SCL3 (biplot score 0.9266). This variable dominates Axis 2 (the
sensitivity gradient) when RDA with forward selection is undertaken using all chemistry
determinands and as such might be expected to account for a substantial part of the
variation in DCL. More surprising is the selection of % coverage of arable land. This variable
was not selected as significant during analysis of the sensitive subset with all chemistry
determinands included and, intuitively, with the screening out of non-sensitive sites this
might be expected. LC2 may be the only variable uncorrelated with SCL3 which explains
additional variance in DCL. Scatterplots showing DCL these variables (Figure 6.15) illustrate
219
Figure 6.15: Scatterplots showing diatom critical load against variables selected by forward selection procedure (sites where Ca<400ueg 1' )
a) Soil critical load class 3 (SCL3)3.0
0.3 • 0 4 0.5 0.6 0.7
Transformed diatom critical load
b) Land cover class 21.50
1.20 —
0.90 —
2 0 . 6 0 -
0.30 —
Transformed diatom critical load
(RDA on these data omitting site mk31 from the analysis altered the results by a negligible amount. The same two variables were selected and the cumulative variance o f D C L explained was reduced by 1%. This was undertaken to ensure that m k3I (which has 27% arable land in the catchment and a N O / concentration o f 95peq I ' ) did not exert an disproportionate effect on the outcome.)
220
the explanatory/response relationships. Sites with increasing % coverage of SCL3 are
characterised by higher critical loads. The relationship between DCL and LC2 exhibits a
wider scatter although sites where the proportion of arable land in the catchment are
characterised by concomitantly low critical loads.
It is apparent that reducing the chemical response gradient, causes the importance of certain
catchment attributes to diminish as others become more influential. RDA using forward
selection was undertaken on two further subsets, one where Ca^''<300peq 1' and one where
Ca^^<200|Lieq l'\ Table 6.18 shows the results of these together with the results of the
Ca^"^<400peq 1' subset and the full dataset for comparison.
Table 6.18: Redundancy Analysis (with forward selection) on datasets of varying sensitivity
Dataset X Variable selected and cumulative variance explained no of sites
All sites 0.810 LC2, 58%; SCL, 69%; G4, 76%; Gl, 79%; 81% 78Ca^*<400^eq 1' 0.680 SCL3, 61%; LC2, 68% 46Ca"+<300;ieq 1' 0.574 SCL3, 33%; LC2, 57% 32Ca"+<200 teq 1' 0.551 SCL3, 26%; Sdep 55% 23
Table 6.18 shows that, as the datasets encompass increasing proportions of sensitive sites,
the level of explanation offered by the catchment attributes decreases as does the number
of significant catchment variables. The importance of variables associated with soil and
geological sensitivity decline with a reduced site sensitivity gradient while the % coverage
of soils of medium sensitivity (SCL3) assumes greater importance. The proportion of arable
land in sensitive catchments accounts for a significant fraction of DCL variance. Where all
sites are very sensitive (Ca<200peq 1' ) it appears that the amount of S deposition becomes
significant. Overall, the pattern is one of decreasing eigenvalues (hence, level of
explanation) with shortened gradients of variation in the water chemistry. Intuitively, as the
221
response variation becomes smaller, there is less to explain.
A comparative analysis of the full dataset and the sensitive (Ca ''<400p,eq 1' ) subset reveals
that the catchment attributes account for slightly more of the variation in the response data
in the former case than in the latter. Approximately 69% of the variation in water chemistry
is significantly explained by catchment attributes in the full dataset (Table 6 .8 b) compared
to 61% for the sensitive subset (Table 6.16b). The difference in explanatory power when
DCL is used as a sole response variable is slightly greater with 81% of DCL variation
explained for the full dataset and 6 8 % for the sensitive subset. During the Phase 1 analyses
of more sensitive sites 40% of the variation in water chemistry was significantly explained
(at the 99% level) by catchment attributes (Table 5.14a). With DCL as the sole response
variable the catchment data explained only 10% (Table 5.16). However, the sensitivity
screening exercise for Phase 1 used sites where Ca^''<200|ieq 1' as a threshold rather than
the 400peq M cut-off used in Phase 2 and as a consequence the results of the respective
analyses are not directly comparable. It is possible that the reduced gradient of variation in
the sensitivity subset precludes the use of catchment attributes as water chemistry
predictors. It is not certain whether the increasing level of explanation in the Phase 2
sensitive subset is the result of a broader gradient of variation in the response variable or
is due to the fact that the explanatory variables are at a higher resolution and relate more
specifically to the catchment. Additionally, there are far fewer sites in the Phase 2 sensitive
subset. As such, the fact that the Phase 2 catchment data exhibit stronger relationships with
the water chemistry (and, more specifically, diatom critical load) may be an artefact of the
statistical technique. These issues are explored further in Chapter 7, which uses multiple
regression to produce equations which predict DCL response to variations in catchment
attributes and in Chapter 8 .
222
6.5 Summary
This chapter has described multivariate statistical analyses undertaken on a dataset relating
to 78 catchments identified as part of a sampling strategy devised to encompass a gradient
of sensitivity to surface water acidification. A water sample taken from each catchment
enabled a suite of chemical determinands to be analysed and these comprised the response
data. A range of catchment attributes were quantified for each catchment, from a variety
of sources, and these constituted the explanatory data. The derivation of these data-sets is
discussed in greater detail in Chapter 4. The objective was to explore the empirical
relationships between the response surface water chemistry and the explanatory catchment
attributes in an attempt to develop a statistical model which will allow diatom critical load to
be predicted from catchment data. The methodology was developed from that used in the
preliminary (Phase 1) analysis described in Chapter 5 by utilising higher resolution
catchment specific explanatory data.
Initially, unconstrained and constrained ordination techniques were used to identify
relationships within, and between surface water chemistry and the catchment attributes. A
gradient representing sensitivity to acidification was identified, following principal components
analysis (PCA), both for the response (Figure 6.2) and the explanatory variables (Figure
6.4). This gradient was maintained when the former were constrained by linear combinations
of the latter following redundancy analysis (RDA) (Figure 6.5). RDA showed that,
statistically, the variation in water chemistry was largely explained (84%) by the catchment
attributes, particularly the soil, geology and land cover variables (Table 6.7). The use of a
fonvard selection procedure enabled a substantial reduction of the number of catchment
variables with comparatively little decrease in explanation (Table 6 .8 a). This showed that two
gradients within the water chemistry were being explained by the catchment attribute, one
representing geochemical acid sensitivity and another associated with sea-salts and organic
223
acidity (Figure 6.6).
Having identified the importance of the primary acidity gradient, attention was focused on
the relationships between diatom critical load (DCL), an indicator of surface water sensitivity,
and catchment attributes. With this single gradient as a response, RDA (with forward
selection) indicated that the sensitivity of catchment soils and geology were important in
explaining variation in DCL. The amount of arable land in a catchment was, however, the
most important factor (Table 6.10). The variables identified following forward selection are
subsequently used in multiple regression analysis in Chapter 7. When (partial) RDA enabled
the variance in DCL to be partitioned into broader, ’variable type’ components it was
revealed that the greatest explanation was offered by covariances between explanatory
variables (Figure 6.9).
Further analysis was undertaken on a more sensitive subset of the data, defined as sites
where Ca^'"<400peq \'\ to assess whether the relationships between water chemistry and
catchment data were the same as those across a broader sensitivity gradient. The water
chemistry for this reduced dataset was dominated by a sea-salt rather than an acidity
gradient (Figure 6.10). This was reflected in the results of RDAs on these data. The level
of explanation provided by the catchment attributes for the sensitive subset is similar to that
offered by the full dataset when the response data comprises the full suite of chemistry
(Table 6.16a). It is substantially reduced when DCL is the sole response variable (Table
6.17). This reduction continues as the sensitivity gradient contracts and is accompanied by
a change in the nature of the significant explanatory variables.
Chapter 6 shows that the catchment attributes employed here can explain a large proportion
of the variation in surface water critical loads across a broad sensitivity. The relationships
upon which the strength of this explanation is founded are expressed in predictive terms in
224
Chapter 7. Less certain are the relationships at the more sensitive end of the scale. The
problems associated with this, in terms of the applicability of a predictive model, are
discussed in Chapter 8 .
225
CHAPTER 7 : PHASE 2 - MODEL CALIBRATION
7.1 Introduction
The previous chapter described exploratory stages in the development of an empirical
statistical model to predict surface water critical load using catchment characteristics. Each
of the 78 sites in the calibration dataset was characterised according to a number of
attributes. These related to soil, geology, land cover, topography, geography and acid
loading. Direct gradient analysis was employed to explore the relationships between the
response data (initially water chemistry and subsequently diatom critical load, (DCL)) and
the explanatory variables (catchment attributes). Redundancy analysis (RDA), in tandem with
forward selection enabled the catchment variables to be ranked according to their influence
on surface water critical load. This chapter seeks to build on the developmental work in
Chapter 6 , and focuses on calibrating multiple linear regression models. The aim is to go
beyond exploring the general associations between variables demonstrated in the previous
chapter by predicting a response variable (DCL) from selected catchment attributes. The
results of regression analyses are presented together with regression diagnostics which
enable the statistical validity of the models presented to be tested. The analysis is
undertaken on the 78 site dataset initially, and subsequently using a sensitive sub-set of the
data (sites where Ca^^^OO|ieq 1' - see Chapter 6 ). The results presented are discussed in
Section 7.4.
7.2 Regression results and model diagnostics
The conceptual basis behind regression analysis is detailed in Chapter 4 where it is shown
that the dependent or response variable Y, can be predicted from values of n explanatory
variables, X,, X„ (Equation 4.3). The results of forward selection (Section 6.3.3)
226
indicate that LC2 (% cover of arable land), SCL (catchment critical load), G4 (% coverage
of limestone and chalk, the least sensitive geology class), G 1 (% coverage of granite and
acid igneous rock, the most sensitive geology class) and catchment weighted soil water
concentration are significant explanatory variables with regard to variation in DCL. These
transformed catchment attributes (see Chapter 6 ) are regressed onto log-transformed DCL
and the results of this are shown in Table 7.1. This summarises the output from running a
multiple regression in SAS/INSIGHT (SAS Institute Inc., 1993).
The R- square ( /f) value of the model is 0.809 which means that approximately 80% of the
variation in DCL is explained by the catchment parameters. As expected, this mirrors the
results from the exploratory analysis using RDA with forward selection undertaken in
Chapter 6 where DCL is the sole response variable. Subsequent discussion will focus on
the adjusted FF (R adj) this accounts for the number of parameters in the model and as
such is more comparable over models which use different numbers of explanatory variables
(SAS Institute Inc., 1993). The RMSE for Equation 7.1 is 0.1698. This means (after
calculation of the antilog value and subtraction of the constant) that, for the calibration
dataset, predicted DCL will lie within O.Skeq ha' yr' of the actual value for c.6 8 % of sites
and within 1.2keq ha' yr' for 95% of sites. Further examination of the regression
diagnostics shows that each explanatory parameter is statistically significant at the 95% level
{p-<0.05). With regard to the collinearity diagnostics, values for tolerance and VIF show that
the explanatory variables are not collinear thus fulfilling one of the major assumptions of the
regression model that it should not be biased. Collinearity can also lead to unreliable
estimates of the partial regression coefficients and their standard errors. The validity of
further assumptions can be assessed by an examination of the residuals or errors. The
residual for the Ah observation is the observed value for the response variable minus the
predicted value. Plotting residual values {i.e. the difference between the observed and
predicted DCL) against predicted DCL can provide an indication whether the model is poorly
227
Table 7.1: Multiple linear regression output with G1, G2. SCL. and LC2 as predictors.
Predictive Equation
log,,, (D C L + 1 ) = 0 .804 - 0.107 lo g ,o (G l+ l) + 0 .206 log,o (G 4+l) + 0.263 log ,„(S C L+ l) - 0.083 + 0.213 lo g ,o (LC 2+ l) (7 .1 )
Regression diagnostics
Type III Tests Parameter Estimates 95% ConfidenceIntervalPredictor F Statistic p-value Estimate Tolerance VIF Lower Upper
INTERCEPT n/a n/a 0.8038 n/a 0.0000 0.4474 1.1601G1 9.5674 0.0028 -0.1073 0.6206 1.6113 -0.1765 -0.0382G4 26.9283 0.0001 0.2056 0.8461 1.1819 0.1266 0.2846SCL 7.5653 0.0207 0.2625 0.5167 1.9535 0.0414 0.4836H* 5.6000 0.0057 -0.0827 0.5018 1.9927 -0.1427 -0.0228LC2 33.6219 0.0001 0.2133 0.5282 1.8931 0.1400 0.2867
RMSE = 0.1698 R = 0.8090 Adjusted R = 0.7958
Type II I Tests: Type II I tests in SAS provide sums of squares associated with each of the estimated coefficients in the regression model (not reported). This is the ’extra sum of squares principle’ (Manly, 1992) which considers the variation in the response variable accounted for by an explanatory variable after the effect of all other variables are considered and is similar to the use of covariables (see Chapter 6). A type I test is more akin to using forward selection (see Chapter 6) which examines the sequential incremental improvement as each variable is added. A partial f-test (Draper and Smith, 1981) can be made for all regression coefficients to examine the unique effect of a variable, in addition to that of the others. This does not depend on the order specified in the model. The F statistic can be used to test the null hypothesis that a parameter in the regression model is 0. The p-value is the probability of obtaining an f-statistic greater than the computed F-statistic if the null hypothesis is true. The significance level set for these analyses is 95%. Thus i f p t 0.05 then the coefficient is not statistically significant and the variable does not improve the fit of the regression model.
Parameter Estimates: The parameter estimates are used to paramaterise the fitted model. The intercept parameter provides a value for in equation 4.3. This is the base constant which is an estimate of the value of the response variable when all explanatory variables are zero. The estimates for each of the predictors provides values forj5, The tolerance and variance statistics provide a measure of the strength of any collinearity between the predictor variables (Rawlings, 1988). Tolerances close to 0 and VDF’s (variance inflation factor) greater than 20 (ter Braak, 1987) are indicative of collinearity. Tolerance is the reciprocal of the that results from the regression of an explanatory variable on the other explanatory variables.
Confidence Interval: A 95% confidence interval is shown for each parameter with upper and lower limits specified for each variable.
RMSE: RMSE is the square root of the mean square error equating to the estimate of the standard deviation of the error term and indicates the range of error that might be expected when using the equation to predict Y. If the residuals are normally distributed approximately 68% of predictions will produce a residual less than 1 standard deviation from the true value and 95% less than 2 standard deviations. (Taylor, 1993). Thus, in this instance, RMSE is 0.17 so approximately two thirds of predictions will be within log„,0.17keq H* ha ' y f ' of DCL + 1 (nb. Y = DCL+1) (ie, 1.5 keq H* ha * yr ‘). Removing the constant used during transformation gives a value of 0.5keq ha"' y f')
: R^ is an indication of the proportion of the (corrected) total variation in the response data that can be attributed to the fitted model.
Adjusted R^ (F^,.;): is adjusted according to the degrees of freedom in the model. As such, it can be used to compare modelswith different numbers of parameters (SAS Institute Inc, 1993).
specified or exhibits heterogeneity of variance (Myers, 1986). The assumption required here
228
is that the errors are not autocorrelated {i.e. they are independently distributed). Figure 7.1,
a plot of the residuals against predicted DCL, is characterised by a random scatter. The
absence of any systematic pattern suggests that the regression model is not autocorrelated
within the model. However, two outliers are identified at both extremes of the distribution.
These are sites mk05 (which is over-predicted) and mk37 (which is under-predicted). Closer
examination of the raw data for these sites showed that there appeared to be no grounds
for omitting them from the regression (e.g. there were no obvious measurement error).
However, if these sites are removed from the analysis rises to 0.843.
Figure 7.1: Residuals plotted against predicted DCL
05-
0 .3--
0 .2- -
:2 0 .1
U Q 0-
3K3K
3K
3*
3*3* JtC_________
3K 3K
3K3*
3t€
3K
3 *
3*
X X
--------X .......3K
f#
3* 3K)K3*
3*
3*
-------
3K
' ^ ' 3 .....................................
3K
.............3 t< *"3*
3*
3K
3K
3*3*
3*
0.2 0'.4 0'.6 d.8 1 l'.2 l'.4 1
- 0 . 1 - -
- 0 .2 - ■
-0 .3 --
-0 .4 "
-0.5-
Predicted transformed DCL
229
A further requirement of the regression model is that the errors are normally distributed. This
is necessary to justify tests for significance and confidence levels used in regression.
Studentized residuals {i.e. standardised with the estimate of the residual variance obtained
after deleting the Ah observation) are used to assess this assumption as they are more
useful than ordinary residuals, particularly if the data include outliers (Wiesberg, 1985).
Figure 7.2 provides information relating to the distribution of the studentized residuals. The
shape and the cumulative distribution function suggest that these are normally distributed.
This is confirmed by reference to the Kolmogorov D statistic which can be used to compare
an empirical distribution with a theoretical distribution (the normal distribution in this
instance).’/7 here is 0.09, which is less than the critical value of D (0.16 for 78 degrees of
freedom).
Figure 7.2: Distribution analyses of studentized residuals
-3 .5 t o -2 .5 ■ -2 .5 t o -1 .5 -1 .5 t o -0 .5 -0 .5 to 0 .5 0 .5 to 1.5 1.5 to 2 .5Studentized residuals
2.5 to 3.5
230
Given that the major assumptions of the regression model are met by this analysis, the high
/fadj suggests that, within the stated confidence limits, the model described by equation 7.1
accurately predicts DCL across a broad chemical gradient. This is illustrated by a plot of
observed DCL against predicted DCL (Figure 7.3) which describes, with some scatter, a
strong linear relationship. The cluster of sites which are underpredicted towards the origin
is a manifestation of the difficulties encountered when attempting to predict DCL at more
sensitive sites. This is discussed further in Chapter 8 .
Figure 7.3: Plot of observed DCL against predicted DCL
1 .6 - -
1.4- ■
1.2 - -
mk37
mk05
■a 0 8- -
0 .6- -
0 .4--
0.2- -
0.60.4Predicted transformed DCL
It is possible that the strength of the predictive relationship may not be substantially reduced
by removing one or more of the predictors, thus simplifying the model and reducing the data
231
requirements. The predictor which adds least to the fit of the model is the log transformed
soil critical load variable (SCL) which has the lowest sum of squares (8 8 , not reported) and
the lowest F- statistic. Given that this variable was derived somewhat artificially and is based
on an overestimation of sensitivity (see Chapter 4) its removal at this stage might be
appropriate. The results of a second multiple regression, removing 8 CL as a predictor, are
shown in Table 7.2.
Table 7.2: Multiple linear regression output with G1, G2, and LC2 as predictors.
Predictive Equation
log,,, (DCL+1) = 1.106 - 0.136 log,„(Gl+l) + 0.229 log,„(G4+l) - 0.104 / h * + 0.228 Iog,„(LC2+l) (7.2)
Regression diagnostics
IntervalPredictor
Type III Tests
F Statistic /?-value
Parameter Estimates
Estimate Tolerance VIF
95% Confidence
Lower Upper
INTCRCEPT n/a n/a 1.1055 n/a 0.0000 0.8481 1.3630G1 16.4750 0.0001 -0.1360 0.7071 1.4143 -0.2028 -0.0692G4 33.5265 0.0001 0.2291 0.9025 1.1081 0.1502 0.3079H+ 12.2688 0.0008 -0.1038 0.5499 1.8187 -0.0477 -0.0447LC2 37.1748 0.0001 0.2280 0.5436 1.8395 0.3025 0.3025
RMSE = 0.1750 = 0.7942 Adjusted = 0.7829
The adjusted value can indicate whether a simplification of the model has resulted in loss
of explanatory power. for equation 7.2 is 0.7829 which represents a reduction of
approximately 1% in explanatory power compared to the model presented in Table 7.1. It
is apparent that 8 CL accounts for very little of the variation in DCL once the other variables
are considered, despite being selected second in the forward selection procedure (Table
6.10). Chapter 6 showed that 8 CL is significantly correlated with G1, H"" and LC2. Although
the magnitudes of the correlation coefficients (Table 6.4) between 8 CL and these variables
and 8 CL are insufficient to cause collinearity in the model, they are large enough to suggest
232
that the explanation offered by SCL can be accounted for by the other variables.
The regression diagnostics for the second model (Table 7.2) indicate that root soil water
concentration now has the lowest F- statistic of all the predictors and adds least to the fit of
the model when the effects of all other parameters are considered. A third multiple
regression was performed omitting H"" and the results are presented in Table 7.3.
Table 7.3: Multiple linear regression output with G1, G2, and LC2 as predictors.
Predictive Equation
log,,, (DCL+1) = 0.712 - 0.161 log,„(Gl+l) + 0.265 log,„(G4+l) + 0.295 log,„(LC2+l) (7.3)
Regression diagnostics
Type III TestsIntervalPredictor F Statistic /?-value
Parameter Estimates
Estimate Tolerance VIF
95%
Lower
Confidence
Upper
INTERCEPT n/a n/a 0.7122 n/a 0.0000 0.5757 1.8487G! 20.9599 0.0001 -0.1609 0.7404 1.3505 -0.2310 -0.0909G4 41.4970 0.0001 0.2645 0.9655 1.0357 0.1827 0.3463LC2 72.5843 0.0001 0.2945 0.7328 1.3646 0.2256 0.3634
RMSE = 0.1879 = 0.7596 Adjusted = 0.7499
The R adj value for Equation 7.3 is 0.7499 and thus this model explains 3% less of the
variation than Equation 7.2 and 4% less than Equation 7.1. The omission of two predictor
variables has not led to a substantial reduction in the fit of the model. The regression
diagnostics for Equation 7.3 reveals that the F- statistic for LC2 has considerably increased
relative to those for G1 and G4. It appears that 102 adds more to the fit of the model when
H'" is omitted. This is likely to be a consequence of a correlation between these two
predictors (although statistically they are not collinear). The increased tolerance exhibited
by LC2 bears this out.
233
Having shown through Equations 7.1, 7.2 and 7.3 that removing predictors from the
regression model does not necessarily result in significant loss of predictive power it can
also be demonstrated that the choice of predictors need not be limited to those indicated by
the forward selection procedure in RDA. Given that, at each step, the forward selection
algorithm in CANOCO selects the variable that adds most to the explained variance of the
response data it is likely that other variables which have been eliminated as a result of
collinearity with the variables already selected. Regression analyses were therefore
undertaken on a variety of explanatory variables selected on a purely subjective basis. Radj
and RMSE for each analysis are presented in Table 7.4.
Table 7.4: Results of regression analvses on a varietv of catchment attribute combinations
Predictors RMSE
LC2, G l, H* 0.211 0.670 0.688LC2, 04. H+ 0.193 0.748 0.738LC2, 04 0.211 0.692 0.683LC2, 01 0.233 0.625 0.615
01, 04 0.214 0.690 0.677%BS, LC2, 01, 04 0.182 0.778 0.766%BS, LC2 0.226 0.648 0.639%BS, LC5, 01 0.240 0.608 0.592LC2 0.245 0.580 0.575%BS 0.271 0.486 0.480H* 0.264 0.513 0.507SCL 0.274 0.476 0.46901 0.313 0.313 0.30404 0.343 0.178 0.168
The explanatory power of the equations (/^adj) produced by regressing these variables onto
DCL varies between 0.168 (for simple linear regression with G4 and DCL) and 0.766 when
the explanatory variables are %BS, LC2, G1 and G4. Taken with the values for
Equations 7.1, 7.2 and 7.3, it is clear that LC2 is the most significant single variable in terms
of explaining DCL variation confirming the results from the forward selection exercise (Table
6 .1 0 ). The implication here is that the percentage of arable land in a catchment will drive the
sensitivity of the surface waters. However, the most sensitive sites, those in upland areas
234
with poorly buffered soils, will contain no arable land and here, a predictive model based
solely on LC2 will fail.
The importance of LC2 is therefore a reflection of the length of the chemical gradient
exhibited by the calibration dataset. Figure 6.7 shows that, although the general trend is for
DCL to increase with LC2, the relationship is more complex in that two clusters of sites are
evident. One, where DCL and LC2 are relatively low exhibits a random scatter of points. The
other represents a linear relationship between LC2 and DCL at non-sensitive sites. The
suggestion is that the use of LC2 as a predictor can differentiate between sensitive and non
sensitive sites. For accurate prediction at the sensitive end other variables are required and
the soil and geology variables become more important. With regard to soil. Table 7.4 shows
that, individually, the three continuous soil variables (H", %BS and SCL) exhibit similar
explanatory power with /^gdj values of 0.507, 0.480 and 0.469 respectively. Each varies
linearly with DCL. Thus, although is identified as most significant following RDA with
forward selection it appears that any of these variables might effectively be included in a
predictive model. Comparison between value where %BS, LC2, G1 and G4 are
predictors (0.766) and Equation 7.2, where is the soil variable (0.783) bears out this
supposition. The highest A gdj values are produced using combinations of land cover, soil
and geology. The specific explanatory variables, however, appear to be interchangeable
although G4, despite exhibiting a lower individual A gdj than G1, adds more to the
explanatory power in multiple models. G4 is the least sensitive geology class and has zero
coverage in most catchments. For the 13 sites within which it occurs, G4 exhibits a linear
relationship with increasing DCL (see Figure 6.7). As such, G4, as well as LC2, can be used
to differentiate between sensitive and non-sensitive sites although the former is less suitable
for predictions across a broad sensitivity gradient. G 1 occurs at many more sites and in
numerous catchments exhibits maximum {i.e. 100%) coverage. These sites tend to have the
lowest critical loads. Across the most sensitive gradients G1 has 100% coverage, and this
235
lack of variation precludes the use of the geological classification as predictors at more
sensitive sites. The next section follows on from the developmental work in Section 6.4. and
assesses the predictive capability of the catchment across a shorter, more sensitive
chemical gradient {i.e. sites where Ca^*<400|ieq 1*^
7.3 Multiple Regression with more sensitive sites (Ca^^<400|ieq 1' )
The lack of highly sensitive sites in the Phase 2 calibration dataset is discussed in Chapter
6 . Nevertheless, an attempt was made to examine the relationships between DCL and
catchment attributes for a reduced sensitivity gradient (sites where Ca^"<400|ieq 1' ).
Redundancy analysis (with forward selection and Monte Carlo permutation testing) only
selected SCL3 (granodiorite, greywacke, schist and gabbro parent materials) and LC2 to
explain variation in diatom critical load (DCL) for sites where Ca^'"<400|ieq \'\ Multiple
regression using these variables as predictors explains 65% of DCL variation, somewhat
less than the explanation offered across the longer sensitivity gradient. Equation 7.4
describes this relationship and the regression diagnostics are presented in Table 7.5.
Further regression analyses were run using other variables in tandem with SCL3.
Regressing SCL3 and H*, in the first instance, and then %BS, produced an of 0.560
in both cases. Neither or %BS significantly contributed to the goodness- of-fit in their
respective equations. The major limitation with these continuous variables in this context is
that they do not exhibit sufficient variation across the more sensitive dataset, and as such
cannot successfully act as predictors. In the most sensitive catchments, for example, G1 will
be characterised by 100% coverage. It appears that the catchment attributes which best
explain DCL within a confined sensitivity gradient are not adequately reproduced in the
calibration dataset (e.g. the more detailed data relating to catchment processes throughout
the soil profile which are used to calibrate dynamic models) . The implications for the
applicability of the predictive model are discussed in Chapter 8 .
236
Table 7.5: Multiple linear regression output with SCL3 and LC2 as predictors - sites where Ca^~'<40Queq 1' .
Predictive Equation
log,,, (DCL+1) = 0.325 + 0.168 log,„(SCL3+I) + 0.187 (LC2)
Regression diagnostics
(7.6)
IntervalPredictor
Type II I Tests
F Statistic p-value
Parameter Estimates
Estimate Tolerance V IF
95% Confidence
Lower Upper
INTERCEPT n/a SCL3 66.6997LC2 12.1321
RMSE = 0.1098
n/a0.00010.0012
= 0.6686
0.3247 n/a 0.00000.1682 0.9917 1.00840.1867 0.9917 1.0084
Adjusted /?’ = 0.6528
0.2759 0.37340.1266 0.20980.0785 0.2948
7.4 Discussion
A number of predictive models have been produced regressing a variety of catchment
attributes against DCL. For many of these combinations is very similar indicating
comparable levels of explanation. The model with the highest fî adj (Equation 7.1)
incorporates the effects of five explanatory variables yet only explains 4% more of the
variation in DCL than a model with two of those variabies removed (Equation 7.3). The
variance partitioning analysis (Section 6.4) demonstrated that covariances between
explanatory variables accounted for the greatest proportion of DCL variance. As a
consequence of this interaction between predictors the regression analyses generates
several acceptable 'minimal' variable models which have similar explanatory power for
predicting DCL. These include a mixture of soil, geology and land use variables. Whittaker
(1984) reviewed a number of analyses, based on automatic selection procedures,
undertaken by several authors (Hamaker, 1962; Aitkin, 1974; Draper and Smith, 1980) of
a dataset published by Hald (1952), concluding that the concept of a "best" regression
equation is a fallacy. Within this context, the choice of predictors for inclusion in a model can
237
be based on subjective judgement. This may involve a priori knowledge of processes
operating within the catchment system. For example, the enormous buffering effects offered
by limestone and chalk, discussed throughout this thesis, may justify the inclusion G4 as a
predictor rather than G1. Similarly, some knowledge of the derivation of the explanatory data
may influence the choice of variables. It may be that some data are characterised by more
inherent inaccuracies than others and, given similar predictive power, the former can be
favoured over the latter. In terms of national application, the similarity between some of the
equations means that selection of catchment attributes for input into the model can be driven
by practical considerations, particularly data availability. For example soil attribute data for
Scottish soils would be of little value for sites in Engiand and Wales whereas land cover and
geology have been mapped using the same classification for the UK as a whole. The next
chapter examines the predictive model presented here in a catchment management context
and also considers the difficulties inherent in predicting DCL at sensitive sites.
238
CHAPTER 8 : DISCUSSION AND CONCLUSION
8.1 Introduction
This chapter comprises a general discussion of results presented in earlier chapters.
Problems stemming from the methodology employed (Chapter 4) are addressed in detail
together with implications for the development of the statistical models to predict critical
loads. Finally, potential improvements to the model are recommended along with the
implications these may have for catchment management strategies.
8.2 Summary of results
The Phase 1 analyses described in Chapter 5 sought to identify relationships between
catchment data and surface water chemistry. The intention was, using a national water
chemistry database (CLAG Freshwaters, 1995) and catchment data from a variety of
sources (see Chapter 4), to assess the feasibility of developing a predictive model to enable
freshwater critical loads to be predicted from catchment attributes. Multivariate statistical
techniques showed that water chemistry was characterised by a 'sensitivity to acidification'
gradient dominating the data structure. The diatom (or baseline) critical load (DCL) is
closely associated with this gradient. It was subsequently shown that approximately 60% of
the variation in DCL could be accounted for by the catchment data. The degree of
explanation offered during the Phase 1 analyses, particularly given the coarse nature of the
predictor variables (see Chapters 4 and 5) indicated that a predictive model might be viable.
The approach was developed further in Chapter 6 (Phase 2) in which the same statistical
techniques were used to identify and quantify empirical relationships between catchment
attributes and water chemistry for a 78 site calibration dataset. The relationships were
239
improved by the use of higher resolution, catchment specific data. The analyses
subsequently focused on relationships between catchment attributes and DCL and a
number of key predictor variables were identified which accounted for over 80% of DCL
variation. These were a combination of soil, geology and land use variables. Specifically,
the following variables were indicated (by forward selection) as the most influential;
■ LC2 : % coverage of arable land
■ SCL : Soil critical load value (averaged over the catchment)
■ G1 : % coverage of the most sensitive geology (granite and gneisses)
■ G4 : % coverage of the least sensitive geology (limestones and chalk)
■ : soil water H* concentration (averaged over the catchment)
The catchment characteristics these variables represent, namely, type of land cover,
geology and soil, are widely acknowledged to be the most important factors in determining
catchment sensitivity to acidification (Section 6.3.2). Chapter 2 shows that surface water
chemistry is primarily governed by solution reactions between geology and soils within the
catchment together with the ameliorating or exacerbating effects of land use. The
relationships exhibited in Chapters 6 and 7 support this. The importance of the interaction
between these variables was also quantified and found to be highly significant (Section 6.4).
Regression analysis was used in Chapter 7 to quantify the relationships between DCL and
catchment attributes within a predictive context. A number of regression equations, using
different combinations of predictor variables, were produced which exhibited similar
predictive power. High adjusted Ff values for these equations are testament to the
predictive capabilities of the catchment data. However, given the distribution of DCL within
the calibration dataset, it is argued that the model can only operate across a wide sensitivity
gradient. Effectively, this limits the models functionality (in its present form) to differentiation
240
between sensitive and non-sensitive sites. Prediction of critical load at high resolution,
towards the sensitive end of the scale, is not possible with the data used here.
8.3 Implications of results for model development
The analyses have shown that catchment attributes can be used to predict surface water
critical loads. The predictive model calibrated in Chapter 7, hereafter referred to as the
'catchment model' or 'model', uses nationally available mapped data (both digital and hard
copy) to characterise the land use, soil and geology types within each catchment. The soil
and geology data relating to each catchment were aggregated according to a number of
sensitivity classifications, also available nationally. The model utilises empirical data relating
to soil properties which were available for Scottish soils. The position occupied by the
model, in terms of critical loads applications, thus lies somewhere between the national
CLAG mapping approach (where DCL for a 10 km grid square is represented by DCL at a
single site within the grid square) and the dynamic modelling approach (which requires
large scale, catchment specific, paramaterisation). As such, it offers an opportunity to
assess, by desk study, sensitivity to acidification at a catchment scale using data available
nationally. Therefore DCL can be predicted for sites where chemical data is not available.
This potentially enables the critical loads approach to be applied for catchment
management purposes in addition to its contemporary use as a mapping aid to national
emission reduction strategies.
However, it should be noted that the model, as described here, applies only to the diatom
critical load for sulphur. It has not been calibrated for the Henriksen steady state water
chemistry (SSWC) model and as such, is limited to predicting the base critical load {i.e. that
which identifies the threshold at which long term change begins to affect diatom
241
communities, the most sensitive in freshwater ecosystems). Target critical loads, calculated
by varying ANC in the Henriksen model (Henrikesen et al., 1992; Harriman et al., 1995b)
are presently beyond the scope of the catchment model as presented here. Consequently
critical loads for individual species cannot be predicted. The scope for adopting the
approach developed here for predicting SSWC critical load is discussed in Section 8 .6 .
Similarly, although the diatom critical load used to calibrate the catchment model is for
sulphur deposition alone it has since been modified to encompass nitrogen and, as a
consequence, total acidity (Allott et al., 1995a). Further development of the catchment
model using empirically derived N and total acidity critical loads is another of the further
research possibilities discussed in Section 8.7.
In its current form, the catchment model operates within two major constraints. It cannot yet
reliably predict DCL at highly sensitive sites and it is limited to predicting the diatom critical
load for sulphur. However within those constraints, the model functions well as a statistical
tool and further development of the model may enable these constraints to be overcome
(Section 8.7). In terms of model application there are numerous caveats which may have
implications for the validity of the model. These are primarily related to the methodology
adopted for model development, and the data used for calibration.
8.4 Methodologlcal Issues
The methodology behind the development and calibration of the catchment model is
presented in Chapter 4 where a number of problems were identified both with the approach
and the data used (see also Chapter 6 ). Some of the difficulties encountered during Phase 2
reflect those experienced during the Phase 1 preliminary analysis. In addition, specific
deficiencies were noted with the Phase 1 catchment data. However, these are dealt with
242
substantively in Chapter 5 and, given the exploratory nature of the Phase 1 analyses, it is
not proposed to elaborate on these here. Emphasis is instead placed on the issues arising
during the Phase 2 model development and calibration.
8.4.1 Sampling strategy
In Chapter 4 some discussion was given over to temporal variation in surface water
chemistry and the problems inherent in using a single sample to characterise the chemistry
of a stream system. In addition, following examination of summary statistics for the
calibration dataset, a paucity of sites with very low critical loads was revealed. These issues
are now discussed more fully.
Temporal variation in water chemistry
The problems encountered in developing a predictive critical loads model based on
empirical relationships when the response data are subject to seasonal variability and, due
to logistical constraints, only one sample can be taken, cannot be divorced from the
temporal assumptions of the critical loads approach. The diatom model was initially
calibrated using the ratio for a training set of lakes using mean water chemistry data
(Battarbee et ai., 1996). Subsequently diatom critical loads have been calculated from spot
water chemistry measurements (Harriman et ai., 1995b). As such, the actual DCL value,
calculated using water chemistry from the site will be dependent on conditions at the time
the sample was taken. Thus during summer low flow, when freshwater Ca * concentration is
greatest as a result of increased proportions of the highly buffered baseflow component
(see Chapter 2), DCL values will be higher than in winter when peak flow comprises a
greater proportion of precipitation based chemistry. Although the conceptual idea of the
critical load is not dependent on season, the timescales of ecosystem change in response
243
to acid inputs can vary from the very short term {i.e. acidic episodes following storm events)
to the very long term {i.e. modified weathering rates in response to climate change) (UN
ECE, 1993). The ecosystem response can thus be measured in the short term or the long
term. Critical loads definitions require that the critical load is geared towards long term
ecosystem change (e.g. Nilsson and Grennfelt, 1988). However, a critical load based on a
water sample taken in the summer may be higher than one taken in the winter. It is possible
that the autumn derived critical load is exceeded seasonally during the winter which may, of
course, result in long term ecosystem change. The problem of how to define critical loads in
this temporal context is further exacerbated if critical load values are used to calibrate an
empirical model. The time of sampling becomes important here because if the calibration
dataset is sampled during peak flow periods the Ca * concentrations will be higher in some
sites (see Section 2.7.4.4) and this may produce different regression coefficients. For the
purposes of this model, sampling was undertaken in the autumn to approximate mean
conditions (see Chapter 4). Future research could include calibrating the model using water
chemistry sampled at the most sensitive time {i.e. during high/peak flow) to produce a
model which predicts critical load at the most sensitive water chemistry. The CLAG
database (CLAG, 1995) from which the Phase 1 analyses drew heavily was weighted more
towards lake sites than stream sites on the assumption that lake chemistry during autumn
or spring overturn was fairly representative of chemistry throughout the year (Forsius et ai.,
1992; Kreiser et al., 1995). It has also been demonstrated that seasonal variations of stream
critical loads are greater than those exhibited by lakes (Harriman et ai., 1995b). Chapter 4,
however, reviews work which demonstrated that critical loads at medium and peak flow
were much more closely associated than those at medium and low flow. This suggests that
autumn sampling offers sufficient representivity for model calibration, given the logistical
constraints inhibiting more regular sampling. The use of secondary sources of time series
water chemistry to validate this approach would form a useful part of any future research.
244
Lack of highly sensitive sites
The site selection criteria used for the Phase 2 model calibration are presented in Section
4.2. The major considerations of the sampling strategy were to ensure that a range of
catchment sensitivities were included in the calibration dataset and, to ensure wide
applicability, the dominant soil, geology and land cover types should be well represented.
However, examination of the summary statistics for the water chemistry response data
shows that sites with very low ionic concentrations are not well represented in the
calibration dataset. This is reflected in an absence of sites with very low critical loads {i.e.
<0.5 keq ha^ yr ’), illustrated by Figure 6.1. As such, it is not possible to build on the
conclusions of the Phase 1 sensitivity analysis {i.e. RDA on sites where Ca ^<200|n.eqr ) in
Chapter 5 which shows a relatively weak relationship between water chemistry (particularly
DCL) and the 'surrogate' catchment attributes. Although a similar analysis undertaken with
the Phase 2 calibration dataset shows a much stronger relationship, it is not possible to
deduce whether this is a result of an increase in the explanatory power of the catchment
data or a lengthening of the response gradient. In the Phase 2 analysis the lack of sites with
Ca ^<2 0 0 peql' required the sensitivity threshold to be increased to 400)ieqr^ The overall
consequence of the lack of highly sensitive sites is that the reliability of the predictive model
at sensitive sites is unknown. Further research using an expanded dataset, including more
sites with DCL <0.5keq H* ha yr \ is required to establish whether the catchment attributes
used here can predict DCL as accurately at the very sensitive end of the scale as they can
across a wide gradient. It is possible, given the narrow range at this sensitive end (at sites
where DCL<2.0keq H* ha y r \ Ca reaches a maximum of approximately 200peql l‘ ) that
prediction is not possible without more detailed model paramaterisation.
245
This section has dealt with methodological problems arising from the sampling strategy and
has considered the sites (or observations) within the predictive model. The next two
sections deal with the measurement and derivation of the response (water chemistry and
specifically DCL) and explanatory (catchment attributes) variables.
8.4.2 Water chemistry
Errors in analytical procedures can be random or systematic. The aim of the Analytical
Quality Control (AQC) is to minimise the errors resulting from the analysis. Precision and
accuracy estimates for the chemical analyses undertaken at the Freshwater Fisheries
Laboratory, Pitlochry are provided in Table 4.2. This shows that there are no systematic
errors and that random errors have been minimised.
In the previous section the diatom critical load was discussed within the context of temporal
variation of water chemistry. The accuracy and applicability of the Ca:S ratio is crucial to the
diatom model and validation of this ratio has been undertaken using analysis of sediment
cores (Battarbee et al., 1996). However the sites used to the calibrate the diatom model
were all from non-afforested catchments and additionally, no site had a Ca concentration
greater than 200peqr\ The diatom model was calibrated for lake sites only, as it is not
possible to derive diatom histories for stream sites although the same assumptions should
hold given that, generally, the same processes of acidification operate in lakes and streams.
It is not possible, therefore, to test the assumption that the ratio can be applied to streams
as well as lakes. The calibration of the diatom model is thus limited to particular site types
246
{i.e. sensitive, moorland, lake catchments) and it is not possible, without further
palaeoecological analysis, to assess whether it can be applied to other types of catchment.
A further uncertainty with the diatom model is the use of the Henriksen F-factor to calculate
pre-industrial calcium concentration (CaJ. This is discussed in more detail in Chapter 3.
The temporal variations in DCL in response to seasonality and changing flow add more
uncertainty and is a particular concern in stream. Additionally, the diatom model as used
here does not incorporate a N component despite the importance of N deposition in surface
water acidification, as evidenced by high NOg' values in upland waters (Reynolds and
Edwards, 1995; Allott eta!., 1995c)
Validation of the diatom model in the light of these uncertainties is beyond the scope of this
research. The uncertainties described do not influence the validity of the predictive
relationships between catchment attributes and DCL, and are only important in terms of the
utility of the model. Indeed the critical loads approach is undergoing constant modification
including the adoption of the more complex FAB model (Hornung at a!., 1995b). Although
the predictive model developed in Chapter 6 and calibrated in Chapter 7 relates diatom
critical load to catchment characteristics, the strength of the empirical relationships
(particularly with soil, geology and land use variables) on which it is based suggest that it
can be calibrated to incorporate modifications of the critical loads approach.
8.4.3 Catchment characterisation
The catchment dataset comprises 29 variables from a variety of sources (see Chapter 4)
Catchments were characterised according to a combination of discrete values (e.g.
altitude), classification systems (e.g. geological sensitivity) and catchment sensitivity values
247
derived from weighted averaging (e.g. soil water concentration). Although the
determination and derivation of some of these variables is fairly straightforward, many
require complex calculation and some are dependent on restrictive assumptions. The
implications of these for the validity of the catchment model are briefly addressed in Chapter
4 and discussed in greater detail here.
GIS and Map derived variables
Catchment area and stream length are subject to errors resulting from digitising procedures
(including interpretation of catchment boundaries) and inaccuracies inherent in the OS
1:25,000 scale maps. Where relief is pronounced the former point is unlikely to be
problematic. Low relief catchments are more difficult to delineate. It is difficult to quantify the
likely magnitude of these potential errors. However, interpretation of catchment boundaries
is a necessarily a subjective exercise and the 1:25,000 0.8 maps are the finest resoiution
available nationally. Additionally, catchment area and stream length account for very little of
the variation on water chemistry at this scale (see Table 6.7).
Deposition and rainfall data
The difficulties arising from applying data mapped using 20km grid squares (based on
interpolations from a sparse network of monitoring sites to individual catchments which may
receive substantially different deposition inputs is discussed in Chapter 6 . Models used to
produce deposition maps are limited by a lack of monitored data and the uncertainties
relating to deposition processes. Despite corrections for elevation and vegetation type the
deposition data simply provide an average value for each 20km grid square. As such the
models cannot currently be used to estimate deposition at higher resolutions because they
are unable to reflect variation at the local scale (Smith et ai, 1995). Similar problems affect
the use of rainfall data. However, at a national scale, these data are the best currently
248
available, and national critical load exceedance maps are determined using data at this
resolution. Deposition (S and N) and rainfall were not among the most important variables
in the development of the catchment model (see Table 6.9). The problems with the
resolution of deposition data are likely to be of more concern if the model is to be used to
predict critical load exceedance (see Section 8.7.5)
Geology
To incorporate a geological component into the model, solid geology maps covering the
sites in the catchment dataset were digitised. It was necessary to use a combination of
1 -.50,000/1 ;63,360 and 1:250,000 scale maps as the former did not provide comprehensive
national coverage. For most catchments, geological sensitivity is based on the
1 ;50,000/163.360 scale maps but at others it is derived at a much lower resolution. This is a
less than ideal situation for comparative analysis and, in retrospect, it may have been better
to incorporate a data availability criterion in the sampling strategy. A further consideration is
the loss of data which occurs with decreasing resolution. This was addressed briefly in
Section 4.8.3. Streamwater chemistry may be influenced considerably by small scale
variations in bedrock geology not shown on low resolution maps. The presence of small
scale calcite veins in a more sensitive bedrock can result in a considerable increase in the
catchment buffering capacity (see Reynolds et a!., 1986). These would not feature on the
very low resolution maps and may also be missing at higher resolutions. Although
geological units may be classified according to the most influential rock type with the
highest buffering capacity, it is still possible that these will not be mapped at the scale used.
Similarly, although geological features such as doleritic dykes may appear on
1:50,000/1:63,360 scale maps these are located schematically rather than with reference to
their actual location. Whether the geological maps at either of these scales are truly
representative of the geological divisions in the catchment is thus questionable. It should
249
also be noted that the geological map units describe the formations at the surface of the
earth and as such they provide a one-dimensional guide to catchment geology. For
example, with dipping sedimentary strata and unconformities the sub-surface geology may
be significantly different to the surface geology indicated by the map and there may be
concomitant differences in the buffering capacity of individual catchments. Although these
sub-surface complexities are indicated on 1:50,000/1:63,360 maps (but not at 1:250,000
scale) by conventional symbols it is beyond the scope of the catchment model to
incorporate them. It is clear that, at the catchment scale, high resolution geology data is
required if the geological input to the catchment model is to truly reflect catchment
conditions. Although higher resolution data should improve the predictive capability of the
model the analyses shows that geology data at the resolution used is statistically important
(Table 6.10).
Once the geology maps were digitised the sensitivity classification developed by Kinniburgh
and Edmunds (1984) was used to aggregate the geology data into four sensitivity classes.
Each catchment can thus be characterised according to the percentage of its surface area
covered by each sensitivity class. Kinniburgh and Edmunds (1984) expressed reservations
concerning the stratigraphical nature of their classification. Despite these, the system was
used for model development because it offered a way of reducing the number of potential
geological variables from 8 8 (the number of units on in the 1:625,000 Geological Survey,
North Sheet) to 4. This data reduction exercise facilitated the application of the multivariate
statistical techniques using a classification based criterion meaningful in terms of the
relationships between catchment attributes and surface waters {i.e. buffering capacity - see
Table 4.7)
Soil
250
Variation in soil type in the catchment has been represented by a number of derived
variables as well as using classification systems. The Phase 1 analyses used a 'site
sensitivity' classification based on a combination of soil, geology and land use data
(Hornung et al., 1994b) as part of the surrogate catchment dataset. During Phase 2
analyses a sensitivity classification was developed initially at a soil association level and,
subsequently for soil series (and by conversion, to soil map unit). This was based on mean
percentage saturation values determined from analysis of 2 0 0 0 individual soil profiles held
on a database at the Macauley Land Use Research Institute (MLURI) (Langan and Wilson,
1991). However, the number of observations used to produce this mean value varies
considerably and for some series only one data point is available. Given the range of values
observed for series where numerous samples are used, this may be unrepresentative.
There is, for example, data from only one profile from the Strathfinella series (Strathfinella
association) on the MLURI database and this has a base saturation value of 44%.
Conversely the mean base saturation for the Strichen series (14%) from the Strichen
association is calculated from 8 6 profiles. However, the distribution of values ranges from
0% to 100% (standard deviation, 21). Thus, although each polygon representing the
Strichen series (or the appropriate soil map unit) is classed as highly sensitive {I.e. base
saturation < 2 0 %), from soil profile information it is apparent that this does not apply globally
for this series. Each catchment can be defined by the percentage cover of each soil
sensitivity class. The way these classes are determined clearly introduces scope for error in
the sensitivity level ascribed to the catchment. It is possible that the sensitivity class
allocated a particular soil series (or map unit) may be inappropriate for specific catchments
(where base saturation diverges from the mean value on which the class is based). This
could make model predictions unreliable. However the series classification is a reflection of
the sensitivities that might be expected given the parent materials and drainage status of
the series (Langan and Wilson, 1981). This is also translated to the redundancy analysis in
251
Chapter 5 which shows that catchments dominated by high sensitivity soils are associated
with surface waters of low ionic strength, while the converse is true where low sensitivity
soils predominate.
Soil water concentration (H*) and % base saturation (%BS) are also affected by the use
of mean values for soil attributes which vary across wide ranges. Using weighted averaging,
a single H* and %BS value was given to each catchment (see Chapter 4) by reference to
the MLURI database. Consequently the use of a mean value may over- or underestimate
the true conditions existing in the catchment. Despite this uncertainty, the empirical
relationships between DCL and H* (strongly negative) and %BS (strongly positive) (Section
6.3.3) are as might be expected given the nature of catchment processes (see Chapter 2).
For some minor series, H* and %BS were not available. Where this was the case, a zero
weighting was given to the series. This did not occur regularly and the areas in the affected
catchments was minimal.
The soil critical load classification is used both in the Phase 1 and Phase 2 analyses. Soil
critical load classes are ascribed to each soil map unit based on the dominant primary
weatherable minerals of the parent material (Nilsson and Grennfelt, 1988). As with the soil
sensitivity classification, a soil critical load class is ascribed to each soil map unit. The
percentage of each class in each catchment provides five explanatory variables, SCL1 to
SCL5. A sixth variable based on this classification gives a single soil critical load value for
each site. This used the lower limit of each class band (as opposed to the upper limit used
by ITE for mapping purposes) and, following weighted averaging, provided a single soil
critical load value for each catchment (SCL). Although SCL is derived on this rather ad hoc
basis it enables the relationship between soil critical loads and freshwater critical loads to
be expressed across a continuum, whereas for the individual SCL classes at many sites the
252
coverage is 0%. In non-sensitive areas there are likely to be no soils in SCL class 1 while in
sensitive areas most catchments will not contain SCL5 soils. The use of SCL as an
explanatory variable overcomes the imbalances that occur using percentage coverages in
this way and it was more important in explaining variation in DCL than any of the individual
SCL classes (Table 6.10).
Catchment characterisation according to the soil variables began with digitising soil map
units (or series) within each catchment. The use of map units to represent variation within a
catchment assumes uniformity of properties within mapping units. Soil maps are produced
by different survey procedures and to different scales and this will have a considerable
effect on utility in terms of unit heterogeneity (Beckett and Burrough, 1971). Examination of
the choice of mapping unit in soil surveys indicates that soil maps do not predict soil
conditions at a particular site with certainty (Bie and Beckett, 1971). Clearly the mapping
unit cannot be assumed to represent heterogenous conditions within its boundaries.
Employing classifications and variables based on characteristics ascribed to mapping units
introduce a further source of error. This error is likely to be compounded as the scale
decreases. However, in the absence of high resolution soils and geology data for national
scale modelling the highest resolution data available is at 1:50,000/1:63,360.
A further problem with using mapped soil data at this scale is that it is not viable to weight
areas nearer stream channels more heavily than those further away. The former are more
likely to have a greater impact on the nature of the stream chemistry than the latter (Billett
and Cresser, 1992) although a study regressing catchment attributes onto alkalinity
suggests othenA/ise (Lynch and Dise, 1985). To enable a heavier weighting to be given to
areas proximal to the stream the soil data would need to be at a much higher resolution.
253
Land cover
The land cover database (Fuller and Groom, 1993a) aggregated to produce explanatory
land use variables has certain limitations in terms of the accuracy with which the data is
representative of land cover on the ground (Fuller and Groom, 1993a). One of the most
important predictors, LC2, is more likely to be classed erroneously at the scale used than
other classes (Section 4.8.2). Nevertheless, in statistical terms the relationship between
DCL and LC2 clearly discerns between sensitive and non-sensitive sites while the
association between low ionic strength surface waters and the moorland land cover classes
(LC5 and LC6 ) is as might be expected. Once again, in national terms, there is no other
database at this resolution that would provide the equivalent, or more accurate information
that might improve the fit of the model.
Summary
The explanatory data has been shown to exhibit numerous limitations in terms of accuracy
and representivity. Given the complex nature of the catchment systems being examined this
is to be expected. The purpose of the model is to enable prediction of freshwater critical
loads for individual catchments for any site throughout Great Britain, and for national
applicability it is necessary to be able to utilise national datasets. The catchment data used
here were the best available in the circumstances (although other data may be available to
various potential users now, and some time in the future) and despite the limitations the
model highlights the strong predictive relationships that exist within these data.
8.5 Model representation of catchment processes
As discussed in Chapter 2 the chemical composition of surface waters depends primarily
upon the hydrological pathways followed by percolating water and the chemical.
254
mineralogical and biological nature of the materials with which it comes into contact. These
factors are closely integrated. The geochemical and biochemical reactions with incoming
rainwater are mediated by factors such as soil type, bedrock and drift geology and
vegetation. The importance of residence time and the nature of the hydrological pathways
stems from the extent to which they allow acid neutralising reactions to proceed. In
freshwaters, the concentration of hydrogen ions (H*) is determined by reactions between
dissolved ions in precipitation and the buffering systems in soil and vegetation. Solution H*
concentration represents the differences between resulting from ionic exchange
reactions and the H* consumed by weathering and exchange processes (see below).
The predictive model described here has been developed using the best data available at a
national scale. The input parameters are intended to act as surrogates for catchment
processes which influence surface water chemistry. The purpose of this section is to
examine the efficacy with which key catchment processes are represented by the variables
selected for use in the empirical model. The fundamental mechanisms which regulate the
acidity of freshwaters are summarised building on the introduction presented in Chapter 2.
An assessment is then made of whether these are adequately represented by the data
used and whether, by improving model paramaterisation (whilst retaining the requirement
for nationally available data), catchment processes might be represented more fully.
Beyond this, an attempt is made to identify an ‘optimal’ dataset which might further improve
the performance of the model .The ‘pragmatic’ approach {i.e. that using national data) seeks
to use data which fall between the high resolution, catchment specific, measured
information required to parameterise dynamic models such as MAGIC (Cosby eta!., 1985b)
(and which might form part of an ‘optimal’ dataset) and the data used to produce regional
maps of acid sensitivity exemplified by Kinniburgh and Edmunds (1984). Within this context
a number of key questions need to be addressed.
255
i) What are the key catchment processes and attributes which determine surface water
chemistry?
ii) How are these key processes represented in the model as currently formulated and do
the variables used as catchment attributes adequately reflect the nature of processes
operating within the catchment {i.e. those involved in buffering S driven acidity)?
iii) Are the data used the most appropriate given the potential availability of other national
datasets which might incorporate both the conceptual requirement that processes be
adequately represented by the input parameters, and the practical requirement that such
data are available nationally.
iv) Which data might ideally be used to represent key processes more accurately?
In Chapter 2, four key factors were identified which dominate catchment response to
acidified precipitation (geology, soil, land use and hydrological pathways). The structure of
this discussion is based on an examination of the important processes relating to each of
these individually.
8.5.1. Geology
The influence that the underlying geology will exert on catchment response to acid
precipitation is based on;
i) the neutralising capacity of the bedrock when in contact with water percolating through the
sub-surface, and;
ii) the rate at which new soil material can be created through weathering.
256
The dominant processes governing the response of water coming into contact with bedrock
(and drift) geology are solution and hydrolysis (Hornung et al., 1990b). The primary
weathering agent is carbonated water.
Solution
Solution, or carbonation, is the reaction of carbonate or bicarbonate ions with minerals. The
formation of carbonates is part of a weathering sequence. Carbon dioxide in water (HgCOg)
is an acid which greatly facilitates the base exchange process and is vital in the solution of
carbonates themselves. Equation 8.1 shows how H* is neutralised by solution during the
dissolution of limestone, the product of the reaction with calcite being calcium and
bicarbonate (Bache, 1983).
CaCOg (calcite) + HT= Ca" + HCOg 8.1
Catchments on calcareous bedrock are likely to yield well buffered surface waters as a
consequence of the rapid kinetics of the solution of CaCOg (Braake et al, 1990). Generally,
solution affects a limited number of minerals including halite, gypsum and amorphous silica
(Bache , 1983). The only common rock which is routinely affected by solution weathering is
limestone and, as such, this process is not important in acid sensitive systems.
Hydrolysis
Hydrolysis is the primary mechanism by which silicate mineral weathering proceeds
(oxidation and solution may also be significant). This involves the replacement of the
dominant base metal cations from the mineral matrix by hydrogen ions which may have
originated from acidified precipitation and can be simply expressed as;
257
MSiO + H" -> M" + HSiO 8.2silicate hydrogen metal cation silicon hydroxide
or, using an example;
Mg^SiO, + 4H* -> 2Mg"* + H.SiO, 8.3olivine hydrogen magnesium silicon hydroxide
The weathering rate and the type of processes acting on a rock are governed to a large
degree by the porosity and permeability of the rock. These in turn determine the ease
withwhich water can enter and weathering products can be removed which, together with
the mineralogy of the rock, are the main factors responsible for acid neutralisation by
geology. In highly porous granular sediments almost all grains may be exposed to
weathering. As the surface area of the rock matrix increases the area available for solution
reactions increases. Thus in shattered, jointed or finely bedded bedrock the greater surface
area is likely to result in a greater degree of buffering than in massive bedrock (Hornung et
al., 1990). Massive rocks can only be weathered at the surface or along joints as they have
no intergranular porosity. Similarly the greater the amount of time the solute is in contact
with the bedrock the more likely that chemical equilibrium between the two will be reached
and the more likely it is that the geology will impart an ameliorating effect on acidified
deposition (Hornung at a!., 1990). The structure of the bedrock determines both contact
area and residence time and will play some part in the solute-rock buffering processes.
Numerous authors have ranked minerals into weathering series based on a general ease of
weathering. These weathering indices incorporate physical, chemical and bulk properties as
well as mineralogy. Goldich (1938) identified a weathering sequence for igneous minerals
related to their ‘basicity’ (the ratio of silica to other cations). The most weatherable minerals
are those with the greatest number of cations which can be replaced by hydrogen. The
sequence described (according to decreasing stability) reads quartz > muscovite > potash
258
feldspar > biotite > alkalic plagioclase > hornblende > alkali-calcic plagioclase > calcic-
alkalic plagioclase > augite > calcic plagioclase > olivine. Weathering sequences for fine
grained minerals have also been attempted (e.g. Jackson et al., 1948) ranging from
secondary minerals which weather extremely slowly (e.g. anatase, gibbsite, kaolinite)
through easily and rapidly weathered minerals (e.g. albite, biotite and olivine) to the soluble
minerals (e.g. calcite and gypsum). These minerals have been observed to be important in
studies comparing water chemistry of streams draining catchments with different underlying
geology. Reid at a/.,(1981) observed that drainage water from gneiss contained higher
amounts of calcium and sodium than that from granite. He attributed this to the more rapidly
weathering andesine plagioclase feldspar in the gneiss compared with the oligoclase
plagioclase feldspar from the granite. Higher concentrations of potassium were assumed to
originate from differences in the weatherability of the main potassium bearing minerals
(biotite in the gneiss and orthoclase feldspar in the granite).
Several other factors determine the rate of mineral weathering as well as the structure and
composition of the mineral, the main ones being crystal size, crystal shape, crystal
perfection and access of weathering agent and removal of weathered product. However,
the overriding influence exerted by geology on water chemistry is based on the mineralogy
and geochemistry of the rock (Hornung at a!., 1990). The most important geological
property of bedrock is the capacity to assimilate protons (Norton, 1980) although the
solubility of the phases and the kinetics of solution are also important. In geological terms,
the factors which most determine sensitivity are the amount of carbonate and weatherable
silicates available. Where CaCOg and silicate minerals are in short supply then the pool of
cations available for substitution is reduced. This is the case when the rate at which
weathering can replace depleted cations is insufficient to keep up with losses incurred
through hydrolysis and solution (Paces, 1986).
259
The geological variables used to parameterise the current model are based on a sensitivity
classification which categorises UK bedrock geology into one of four classes based on
buffering capacity (Kinniburgh and Edmunds, 1986). Each geological map unit within a
catchment is allocated one class and the proportion of each class is calculated and
constitutes a single variable. The limitations of this approach are acknowledged by the
authors and foremost among these is the lithological and stratigraphical basis of the
classification. Geochemistry is not explicitly incorporated. Thus there is no attempt to
quantify carbonate or weatherable silica content. Similarly, the amount of reactive minerals
(e.g. quartz, calcite, biotite) which are also important in weathering reactions are not
included. Nevertheless, to apply the classification requires only the geological map units
within the catchments to be known. These are available, albeit at varying resolutions,
across the UK whereas there are currently no such data relating to
geochemistry/mineralogy. Geochemical atlases are currently being produced by the British
Geological Survey (Green, pers. comm.) and these will provide a national picture of
geochemistry in the UK, based on point source geochemical data. Future model
development could incorporate these data rather than use a lithological classification.
In terms of modelling catchment response to acid deposition, it would be useful to
incorporate some explicit measure of the key processes (hydrolysis, solution) based on
information derived from individual catchments. For ‘optimal model’ paramaterisation,
quantification of fluxes of carbonate, silicate, hydrogen and cationic species could enable
some measure of geological buffering to be used as a predictor of surface water chemistry.
Very few catchment specific studies have quantified rock weathering processes (e.g.
Bricker et al. 1967; Likens et al., 1877; Reid, et al., 1981b). Indeed, quantification of these
processes are difficult at a catchment scale and therefore inappropriate for a generalised
260
model for national application. The data required are not available for large numbers of
catchments. Similarly, the composition of the underlying geology in a catchment can be
determined by a detailed petrological examination of the rock itself (e.g. Hatch, 1973) or by
the weathering product which would give some idea as to the content of carbonate and
weatherable silicate . However, these too can only done on a catchment specific basis and
this approach cannot be scaled up to the national scale.
A number of attempts to relate geology to surface water chemistry based on the content of
carbonates and silicate have been undertaken without explicitly requiring quantified
amounts of these minerals to be determined. The predictive schemes cited in Hornung,
(1990b) {i.e. Galloway and Cowling, 1978; Likens et a!., 1979) and Norton (1980) classify
rocks according to buffering capacity and the classifications employ relationships between
rock type and carbonate and weatherable silica content. Hornung used a combination of
the Norton (1980) and Kinniburgh and Edmunds (1984) classifications and applied it to the
solid geology of Wales using published maps and data on the lithology and mineralogy of
the main stratigraphie units. Norton (1980) identified a spectrum of mineral (and rock)
response to acid precipitation and argued that geology can be classified according to the
buffering capacity they accorded the surface waters. The classification was based on map
explanations stratigraphie lexicons and the author’s personal knowledge of certain
geological areas. This approach may offer a more practical means of improving process
representation. However the collation of detailed geological information from memoirs and
lexicons would require substantial effort if the model is to have national relevance.
The major assumption underpinning the use of the sensitivity classification used is that, at a
coarse level, geological formation can be used as a guide as to the composition of the rock.
This may be sufficient when attempting to relate geology to surface waters at a national
261
scale, for all types of rock. Reuss etaL, (1976) noted that the most sensitive surface waters
tend to drain areas overlaying granite or other highly siliceous rocks while Kramer (1976)
observed that calcareous rock can buffer even the most intense acid loading. The analyses
presented in Chapters 6 and 7 reflect these observations. The most sensitive catchments in
the calibration dataset are associated with high proportions of G1, the sensitivity class
comprising granite and acid igneous rocks whereas catchments dominated by limestone
(G4) tended to be non sensitive, and are characterised by high critical loads. This echoes
other catchment based approaches relating geology to surface water sensitivity (Lynch and
Dise, 1985; Bricker, 1986). The use of this sensitivity classification applied to UK geological
maps (by definition, nationally available) distinguishes between those catchments which are
susceptible to surface water acidification and those which are well buffered.
While some authors have attempted to relate geology to surface water chemistry by
quantifying rock weathering processes and mineralogy, this approach is not appropriate for
calibrating or validating a model for general application at a national scale. If geochemical
attributes were available for all geological formations it might be feasible to produce
catchment weighted mineralogical parameters for use as explanatory variables. However,
this is not currently the case and the data used in Chapters 6 and 7 appear to offer the most
realistic paramaterisation of a geological component at this scale. It would be useful
however to compare this approach with those using some of the more detailed geochemical
data (described above), at catchments for which data is available at both resolutions, to
assess whether the model performs better using higher resolution data.
8.5.2 Soil
262
Chapter 2 examined the chemical, physical and biological processes which occur when soil
and water interact and how they affect the acidity of the drainage water. Some of these
processes are also associated with the catchment geology (see above) which, with the soil,
forms part of a highly integrated system. This section identifies a number of key processes
and attributes which are central to the role of the soil in buffering surface waters from
acidity. Four key process areas are discussed.
Soil processes
1. Chemical weathering.
Weathering processes consume hydrogen ions and release other cations (e.g. Ca , Mg^\
a P ) from the soil matrix leading to higher concentrations of cations in the soil solution
(Henriksen, 1984). Catchments are generally sources of these ions, although Mg and Ca
can originate from sea-salts. Silicate weathering proceeds by hydrolysis (see above) which
is the replacement of one or more cationic species from the mineral matrix by hydrogen
ions. These may have originated from carbonic acid (which is dominant in the absence of
anthropogenic effects), organic acids of biological origin or wet/dry deposited acid
components. Weathering is most advanced for the most weatherable mineral close to the
surface and negligible at depth because the descending water becomes progressively more
saturated with the most weatherable mineral. The weathering rate is influenced by soil pH
and the soluble salts present (Cresser and Edwards, 1983).
When a dilute acid solution interacts with a mineral assemblage, hydrolysis commences
according to the appropriate mineral weathering equation (e.g.. Equation 8.3). Assuming
there is no additional acid input available, the H* is neutralised and hydrolysis slows. If the
contact time is of sufficient length, equilibrium is reached. If the acid solution is replenished
further weathering occurs until mineral equilibrium appropriate to the pH of the water/COg
263
system is attained. In soil, the pH of the buffered soil/water system is the crucial factor.
Water draining soil often has a pH near or greater than 7 because of partial or complete
acid neutralisation by weathering or due to outgassing of dissolved CO . However,
thermodynamic equilibrium may not be reached in the field because the kinetics of mineral
weathering are often slow. Also, the overall process is complicated by the involvement of
ion exchange, redox and adsorption - desorption reactions (see below), testament to the
integrated process-response systems operating within catchments. Peat catchments are
particularly vulnerable to acid deposition because the cations leached by exchange with H*
ions are not replenished by mineral weathering.
Many catchment processes, mechanisms and reactions can be described, for many soils, in
terms of rapid surface adsorption and exchange processes (see below). These are
ultimately linked to long-term mineral weathering which is responsible for the creation of
fine-grained highly reactive minerals that dominate adsorption and exchange in mineral
soils. Additionally, mineral weathering may act as a proton sink in its own right and,
depending on the rate of the processes involved, contribute significantly to the neutralisation
of acid deposition (Bache, 1983). As with geology, the long-term capacity of a soil to
neutralise incoming acidity is primarily determined by it’s content of weatherable minerals
and on the access of percolating solutions to the surfaces of these minerals. Even in short
term catchment experiments, weathering has accounted for 70-100% of proton
neutralisation in soils with moderate to high amounts of weatherable materials and up to
70% in soils such as podsols containing smaller amounts of weatherable material (Van
Breeman et al., 1984). To predict sensitivity, even using non-dynamic steady-state models,
there is clearly a need to incorporate the buffering role of mineral weathering given its
capacity to neutralise acidity both in the short and long term and this is discussed further
below.
264
2. Ion exchange
Cation exchange reactions constitute one of the most important physico-chemical
processes within the soil (Cresser and Edwards, 1987). These occur at negatively charged
surface exchange sites on soil organic matter (primarily carboxyl acid or phenolic OH
groups) and clay minerals. The extent of the dissociation of the former (and therefore the
cation exchange capacity) is pH dependent. The negative charge of clay minerals arises
due to the clay mineral lattice effects and isomorphous substitution (the displacement of one
lattice cation by another of similar size by ions with lower positive charge). Cation exchange
reactions in soils are very rapid and occur at much faster rates than mineral weathering
reactions. As such they are responsible for short term buffering of drainage water
chemistry.
Following rainfall, the existing soil solution is diluted by infiltrating water and new solute is
introduced. This equilibrates rapidly with cations on the exchange complex. If precipitation
pH is less than soil pH, cations will be displaced by H from exchange sites. The proportion
of individual base cations (and aluminium and iron species) displaced depends on their
relative distribution on exchange sites and their relative and absolute concentrations in
incoming precipitation. Soils with low base content and low sulphate adsorption capacity
(see below) are moderately sensitive to accelerated base cation leaching and acidification
(Johnson, 1981). As the fraction of exchange sites taken up by calcium falls below 0.1, the
contribution of aluminium to the cation concentration of the equilibrating soil solution
increases and there is a concomitant reduction in calcium concentration (Reuss, 1983). The
composition of the resulting drainage water depends on the nature of the exchangeable
cations mixture and the initial interacting water chemistry. Some cations displaced from the
exchange sites are leached out of the soil profile together with the associated anions. In
265
longer intervals between storm events, geochemical weathering proceeds, replenishing all
or some of the cations leached from exchange sites. This process is assisted by elevated
CO2 building up in the soil atmosphere.
An important aspect of buffering in many acid soils is the exchangeable acidity which refers
to the and Al^ ions adsorbed onto clay or organic exchange surfaces . Of importance
here is the displacement of and Al^ by (for example) a neutral salt. Soil solution pH is
buffered against rapid fluctuation by the exchange acidity as it acts as a reservoir for these
acid ions (Wilson, et al., 1988). When soil weathering and atmospheric inputs produce
sufficient base cations and ion exchange equilibria are such that the base saturation of the
exchange complex remains high, soil pH remains relatively stable. Once geochemical
weathering is incapable of adequately replenishing base cations leached from the exchange
sites then soil solution pH and base saturation falls.
3. Anion/sulphate adsorption
Sulphate can be retained in catchments by;
i) Biological uptake as organic S. The SO/ uptake and cycling capacities of the
biotic component of the ecosystem can buffer the impact of sulphate deposition.
ii) Adsorption reactions in mineral salts (Walker at al., 1990). Some soils have the
capacity to adsorb SO/ on particle surfaces which buffers against solution sulphate
concentrations (Reuss and Johnson, 1986). These reactions are more rapid than those
involving biological uptake.
The buffering role of sulphate adsorption is twofold. A SO/ adsorbing soil delays the cation
leaching effects of dilute H^SO inputs until adsorbing capacity is satisfied. Additionally the
adsorption process can directly neutralise applied acid because soil adsorption of SO/ from
266
HgSO solutions is often accompanied by a rise in pH which results from the replacement of
-OH groups on mineral surfaces by SO/ . With regard to the biotic component of SO/
adsorption, accumulation in the forest biomass represents a relatively minor part of the
capacity of the system to accumulate S. The soil organic matter contains the largest pool of
S in most forest ecosystems (Reuss and Johnson, 1986). However, adsorbed SO/
assumes increasing importance in soils heavily loaded by acid deposition and may exceed
organic S in soils receiving large atmospheric inputs.
The degree of adsorption depends on the sulphate concentration in solution and the pH. As
solution pH decreases in response to increased S deposition, the positive charge of the
solid phase increases allowing greater anion retention (Harriman eta i, 1995a). Sulphate is
not strongly adsorbed on exchange complexes in UK upland soils as these are relatively
young and not deeply weathered (Jenkins et a!., 1977). Indeed, in dynamic models such as
MAGIC (Cosby et ai, 1985b, 1990), applied primarily to upland catchments, SO/ is
effectively assumed to act in steady state. In these upland soils SO/ and 01 tend to be
leached vertically through the soil profile. SO/ is generally adsorbed at depth with
concomitant adsorption of base cations leached from upper horizons (Cresser and
Edwards, 1987). In this instance chloride will normally be the predominant anion in the
solute. If, however, the SO/' adsorption sites at depth are saturated as a consequence of
continued S deposition in precipitation, SO/ will become the dominant anion and leaching
will increase. In the medium to long term soil and, ultimately, freshwater acidification may
follow. The importance of SO/ anions in the transportation of cations through and out of the
soil (the mobile anion concept) is central to understanding of the acidification of soils and
waters (Johnson and Cole, 1977; Cronan et al., 1987; Seip, 1980). Elevated levels of SO/
in throughput following acidic deposition are accompanied by increased leaching of base
cations from ion exchange surfaces. Where the supply of base cations from soil, geology or
267
biological exchange sites is unable to keep pace with SO/' levels the cation deficit is met by
and Al^ ions (Harriman ,1988). It is this process which fundamentally determines the
sensitivity of surface waters in a catchment.
4. Processes involving fluxes of organic acids
The acidity, both natural and anthropogenic, of surface waters is influenced by processes
acting on fluxes of organic acids (OA). These include solution, deprotanation and
degradation. In freshwater systems organic acids originate from the degradation of
biomass, predominantly from terrestrial sources. In the context of acidification, these mainly
comprise compounds with low chemical reactivity (Hemond, 1994). Many studies have
highlighted the contribution of OA to acid freshwaters (e.g. Theis, 1990; Kamari et al.,
1991).
However, it has also been noted that where the pH of acid waters is driven by OA these
systems are buffered against both increased and decreased concentrations of strong
minerals acids (Hemond, 1984). Protanation of organic acids provide a buffering capacity
(Reuss and Johnson, 1983; Krug and Frink, 1983). The impact of strong minerals acids in
precipitation can be buffered by concomitant losses in organic acids which results in little or
no reduction in pH. In peat or wetland catchments the increased reduction and storage of
s o / can buffer against the acidifying potential of the mobile SO/ anion (Hemond 1990).
However, the capacity of this form of buffering is limited and can be overcome by sustained
high levels of sulphur deposition (Hemond, 1994).
Fluxes of organic acids are often associated with episodic events. This reflects the routing
of soil water through organic upper soil horizons following, for example, intensive rainfall.
Such events will often result in a dominance of lateral flow (Section 8.6.4). Subsequently,
268
drainage water does not react with exchange complexes on the mineral soils and as such,
bypasses the acid neutralising environment associated with mineral soils. The fall in pH of a
stream during storm conditions can also be due to the increased significance of soluble
organic acids from the surface horizons. The importance of hydrological pathways and their
representation in the catchment model is discussed below.
The role of organic acids in freshwater acidification relative to anthropogenic sources of
acidity (e.g. HgSO and HNO3 ) is not fully understood (Hemond, 1994). Such uncertainty
also applies to the biological pathways associated with the production and consumption of
OA together with the physical means of their transportation to surface waters (Hemond,
1990) Further complexity is added by the potential of OA to ameliorate or exacerbate
freshwater acidity. In circumneutral systems organic acids from catchment sources dissolve
in water leading to discoloration and acidification (Krug and Frink, 1983). However, when
strong mineral acids create acidified systems, pH decreases and organic acids perform a
buffering role by complexing aluminium thus reducing the concentration of ionic aluminium
species in solution (Reuss and Johnson, 1983). Thus the confusing situation exists
whereby the catchment processes producing organic acids can increase or decrease the
potential for acidity, depending on pH conditions. Model calibration in these circumstances
presents obvious difficulties, in terms of incorporating organic acid effects it would be useful
to examine the relationships between critical load and catchment attributes under different
pH conditions, an approach which is discussed further in Section 8.6.3. One possible
approach might be to use the amount of peat within the catchment as a surrogate for the
organic contribution to freshwater acidity.
269
The processes discussed {i.e., those involving weathering, ion exchange, adsorption and
organic acids) are driven, or limited by a number of soil properties or catchment attributes
which are now examined.
Soil Attributes
1. Hydraulic residence time.
The retention time of water in the soil will dictate the amount of time that the water is in
contact with the soil matrix and therefore the extent to which chemical reactions can occur.
The role of hydrological pathways within this context is discussed in more detail below.
Residence time tends to be short in thin soils. If the chemical composition and geology of
the bedrock and soil is similar then variations in CM (Ca + Mg) concentration in surface
water can be due to the retention time in watersheds (Henriksen, 1984).
2. Content of carbonate and weatherable minerals.
The importance of these echoes their role in bedrock buffering of acid precipitation. Both are
low in thin acid soils. Through processes similar to those operating at the bedrock/solution
interface, calcareous soils will usually yield well buffered waters.
3. Horizon/soil depth
The depth of the soil profile and of the individual horizons through which the water flows are
key factors as they determine to a great extent the potential for neutralising processes to
occur. Deep, well drained soils will favour acid neutralisation through ion exchange in
addition to providing a reservoir of base cations to replenish exchange sites.
Kinniburgh and Edmunds (1986) observed increases in acidity in the surface, organic
horizons of the soil where the dominant proton sources involve dissociation of organic acids
270
and ion exchange reactions. These layers appear to be a net source of acidity. In the
mineral horizons solution acidity may gradually decrease down the profile. Clearly the
presence of deep organic horizons increase the possibility that lateral throughflow will
bypass the exchange sites in the mineral horizon. These sensitive upper organic horizons
underpin soil critical loads maps (Hornung etal., 1995). However, the relative importance of
different horizons is, to a large extent, dictated by flow conditions and these form part of the
discussion in Section 8.5.4.
4. Base cations in the catchment
In alkaline or neutral soils the cation exchange complex is dominated by base cations
whereas aluminium species and H* are prevalent in acid mineral and acid organic soils,
respectively. The total amount of base cations in the catchments is based on;
i) the product of the exchangeable fraction;
ii) total cation exchange capacity of catchment soils;
iii) bulk density of the soil;
iv) aqueous concentration and pore water volume of soil per unit area.
5. Base saturation.
Base saturation is the extent to which the exchange sites within a soil are saturated with
cations other than hydrogen and aluminium. This is expressed as a percentage of the total
cation exchange capacity. As discussed above, solution draining through the soil matrix will
equilibrate quickly with the ion exchange sites. As a consequence, the solution pH will be
driven by soil pH rather than precipitation pH (unless the latter is very acid, Cresser et al.,
1986). Acidified precipitation will temporarily lower base saturation by removing cations
from exchange sites. These may subsequently be replenished by mineral weathering. If
weathering rate cannot replace cations as fast as they are removed, base saturation will fall
271
to a level dictated by residual mineral weathering and precipitation chemistry (Cresser and
Edwards, 1987). Over time this may lead to a lowering of soil pH and, subsequently pH of
surface waters. In soils with moderate to high base saturation but low in readily weatherable
silicate minerals the buffering capacity may be lowered rapidly as proton inputs and cation
exchange reduce base saturation, increasing the equilibrium concentration of hydrogen and
aluminium ions (Berden etal., 1988).
Summary of key soil processes and attributes
The capacity of a soil to neutralise incoming acidity is dependent on the interactions
between the various buffering systems {i.e. carbonate dissolution, silicate weathering,
cation exchange and, depending on pH, aluminium (Ulrich, 1983)), the ions on the
exchange sites and in solution and the weatherable minerals in the soil. The interactions
involving the processes and the nature of the attributes discussed determine the acidity of
both the soil and drainage waters. As acidified deposition enters a catchment system the
s o / concentration of the soil solution increases. Initially, the sulphate may be adsorbed on
soil particle surfaces or taken up by the biosphere (see above) which will delay the onset of
increased sulphate concentrations. Subsequently the soil system may reach an equilibrium
at which elevated levels of SO/ become sufficiently high so that the outgoing flux of
sulphate in drainage water is equal to that incoming as acid deposition. As sulphate in soil
solution and leachate increases, charge balance considerations dictate that SO/ ions are
accompanied by an equivalent amount of cations. In neutral or moderately acidic soils the
dominant anion in solution is HCOg'. In catchments where a moderate supply of bases are
available {i.e. where >15% of negatively charged exchange sites are occupied by the base
cations Ca^\ IVIg \ K* and Na ), the base cations will comprise most of the increase in soil
solute concentration (Reuss and Johnson, 1986). The increase in the rate of base cation
removal may ultimately lead to the acidification of the soil system. However, because the
272
total supply is generally large compared with the annual input of H* in acid deposition, this
will be a very slow process in deep or base rich soils. In soils with low base status a
significant fraction of increased cations in solution and leachate may consist of H* and ionic
Al species. The latter is toxic and both will tend to reduce alkalinity of discharge water. The
key difference between naturally acid systems (where anion supply is likely to be limited by
protanation of bicarbonate and organic acids) and anthropogenically acidified affected
systems is that the latter are supplied by an outside source of SO/ . Galloway et al., (1983)
identify a seven stage scenario which presents a simplified summary of the processes
described hitherto. Stage 1 has the system prior to the onset of acidification. Adsorption of
s o / occurs in Stage 2. Stage 3 sees SO/ breakthrough, initially accompanied by base
cations but, as base saturation falls, by or Al^ increasing solution acidity. During stage 4
the system is in steady state with base saturation near zero. Subsequent stages describe
recovery periods following reduced SO/ input and desorption of SO/ from soils.
Representing processes and attributes within the model
The sensitivity of surface waters to acidification is therefore determined by a formidable list
of soil mediated processes and parameters. In highly complex process oriented models
such as ILWAS (Gherini et ai., 1985) attempts are made to incorporate as many of these as
possible. For the purposes of empirical modelling using available data for parameterisation,
the key question is how well can these processes and attributes be represented as input
variables. Chapter 4 details the soil variables used to parameterise the current model. A soil
sensitivity classification is employed (Langan and Wilson, 1991; Hornung et ai., 1995a)
based on soil water pH (H ) and % base saturation (%BS). A second classification, the soil
critical load (SCL) (Nilsson and Grennfelt, 1988; Sverdrup and Warfvinge, 1988; Hornung
etai., 1994) is based on the dominant weatherable materials and incorporates the effects of
precipitation, vegetation, texture, drainage and soil depth together with the role of secondary
273
minerals. In addition, explicit values for H \ and %BS based on soil profile data are
calculated for each catchment. These variables would appear to represent some of the
more important aspects of the soil buffering system. The potential for base cation
replenishment is represented by the soil critical load classification which also incorporates
key drainage, depth and texture factors which, to a large degree, determine the efficacy with
which neutralising processes will occur. %BS is a measure of the neutralising potential of
the soil whereas H* is a reflection of the availability of free carbonates and the amount of
weatherable silicates (Hornung et al., 1990b). There is considerable collinearity between
these soil variables as they all provide some indication of catchment sensitivity. SCL and H*
emerge as key predictors of DCL. Nevertheless, in general terms, the representation of
processes and attributes within the soil remains fairly coarse.
Ideally, in an ‘optimal’ dataset, the soils component would incorporate some quantified
measures of each of the key soil processes or properties outlined above. This kind of model
paramaterisation is already undertaken, to some extent, when applying dynamic models
such as MAGIC (Cosby et a!., 1985b, 1990), which is primarily driven by soils data. Key
soil parameters required for MAGIC include
• depth
• bulk density
• porosity
• CEC (measured at pH)
• Fraction of exchange sites occupied by Ca^\ Mg , Na"" and K\
The total amount of each base cation within the catchment is the product of these variables
(Jenkins et a!., 1997). Values for each of these parameters are established on a catchment
specific basis and are aggregated spatially and with depth at each catchment to obtain a
274
single value for each parameter. Generally MAGIC is a ‘two compartment’ representation of
a catchment comprising soil and stream compartments. Where extensively developed A
and B horizons characterise the catchment (e.g. in the case of a podsol) three
compartments may be used. This allows for the different physical and chemical attributes of
both horizons to be incorporated into the model. Therefore the influence of processes
operating within respective horizons can be examined in the way described by
Christopherson et al. (1982) where a two reservoir model comprises an upper reservoir
supplying quickflow in contact with the upper soil horizons and a lower reservoir supplying
baseflow. Other catchment specific models such as BIRKENES (Christopherson at a/.,
1982; Christopherson and Wright, 1981) have incorporated cation exchange, weathering,
dissolution/precipitation of gibbsite, SO/ adsorption/desorption and SO/ mineralisation.
ILWAS incorporates an even more comprehensive array of hydrological and chemical
processes (Reuss et a!., 1986) and as such it is very complicated to parameterise and
difficult to apply.
Although an ‘optimal’ dataset might include data relating to ionic fluxes involving hydrolysis,
cation exchange and anion adsorption and desorption for specific catchments, it is not
generally feasible to quantitatively define acid/base consuming reactions in complex soil
systems (Kinninburgh, 1986). The process oriented dynamic models referred to above
employ relatively detailed soil attribute data collected from specific catchments. These
models cannot be applied to catchments for which such data are not available and are
therefore inappropriate at regional or national scales. However it may be possible to
develop a more detailed empirical model by incorporating the physical and chemical
attribute data held on national soil databases, for England and Wales by the Soil Survey and
Land Use Research Centre (SSLRC) and for Scotland by the Macauley Land Use Research
Institute (MLURI). Soil profile data are available for a range of properties including, horizon
275
depth, organic carbon, bulk density, total porosity, pH, cation exchange capacity and
exchangeable bases for each of the soil series on the 1:250,000 soil map of England and
Wales. Similar soil attribute data (with the addition of base saturation and C:N ratio) is
available for a profile database at MLURI which also includes representative profiles for
each soil series. Using digital catchment boundaries, the proportion of each soil series in
each catchment can be determined as described in Chapter 4. Reference to the soil profile
databases would enable some representative value for the soil attribute data for each soil
series identified in the calibration dataset to be determined. Using the relative proportions of
series within a catchment it would then be possible to calculate a weighted value for each
attribute in much the same way as MAGIC is calibrated (Jenkins et al., 1987). Given that
data are available for each horizon, it would be possible to generate a single value for each
parameter, weighted spatially and for depth. Similarly, weathering rates Scotland have been
determined for the major soil associations in (Langan et a!., 1995). These data, introduced
earlier (Section 4.6.2.6), would be of considerable use if coverage extended beyond the 18
major soil associations. However, all these parameters vary across a wide range within the
same soil series. If the national profile databases were employed in an attempt to move
towards a more detailed dataset this problem would need to be addressed. The use of a
median value or centred data may overcome the disproportionate influence of extreme
outliers which can be encountered when using means. The use of these data offers a
pragmatic solution to model improvement in a way that the hypothetical, catchment specific,
process oriented variables discussed within the context of an optimal dataset do not.
8.5.3 Land use
Chapter 2 examined how the effects of land use and catchment management can influence
the sensitivity of freshwaters to acidification. The role of forestry and upland agricultural
276
improvement on exacerbating and ameliorating, respectively, the consequences of acidic
deposition is stressed. It is emphasised that the focus of this discussion is on processes
involving precipitation acidified by sulphur rather than nitrogen deposition, as the predictive
model being developed is based on the critical loads models hitherto published.
Forestry
Studies have shown that conifer afforestation enhances the acidity of water draining from
catchments when compared with moorland sites (Harriman and Morrison, 1982; Nilsson at
al., 1982; Stoner and Gee 1985). A number of reasons for this have been cited. Base cation
uptake by the forest biomass can lead to acidification of surface waters due to the release of
H* ions in exchange for base cations (or of OHVHCOg' ions in exchange for anions taken up)
or to the reduction in the pool of exchangeable bases (Reuss and Johnson, 1986).
Decreased pH of streams draining forest catchments has also been attributed to fact that
trees are characterised by larger surface areas in comparison with grass and moorland
vegetation which enhances their efficiency in terms of scavenging marine salts and gaseous
pollutants from the atmosphere. The base saturation and pH of the soil surface organic
horizons can be lowered by forest growth and litter fall and equilibrating water draining
laterally through or over these horizons is more acidic regardless of rainfall acidity (Cresser
and Edwards, 1984). Additionally, changes in soil physical features associated with initial
drainage improvements or subsequent growth following afforestation are such that drainage
water is less likely to enter the mineral rich horizons. The bypassing of these buffered layers
leads to the entry of acidified runoff into the surface water system (Miller 1985a). Forest
clearance can cause drastic changes in catchment response to acid deposition, impinging
on a range of physical processes. Mechanical erosion following clearfelling may facilitate
acidification in a number of ways including a loss of storage water capacity on upper slopes,
reduction in the depth of the soil profile, the acidification of mineral soil due to increased
277
leaching and SO/' saturation and an increased likelihood of rapid throughflow (Cresser and
Edwards, 1987). Coniferous forest growth or regeneration on a clearfelled or burned forest
site may lead to the creation of acidic surface organic horizons (Krug and Frink, 1983). The
underlying soil may also be acidified as a result (Skeffington, 1983)
Liming
In lowland agricultural areas, or where grassland is managed, problems with acid waters
are much less likely to occur because the soil is often routinely limed to maintain pH and
base saturation. The nature of catchment response to such treatment depends on the
amount and type of lime added (Hornung et al., 1986). Generally, calcium and magnesium
concentrations increase with concomitant reductions in the concentrations of hydrogen and
aluminium ions. Drainage waters from all horizons will have higher calcium, magnesium
and bicarbonate levels than precipitation and will be considerably less acid (Hornung,
1990c). Catchments undergoing pasture improvement may be buffered against acidic
deposition regardless of the nature of their soils and geology.
Model representation of land cover
It is a requirement for the development of a nationally applicable model that, initially, it
should be calibrated to encompass the fullest range of potential sensitivity. Therefore,
variables reflecting the broad range of land use categories and processes should be
employed. Dynamic acidification models now incorporate the impact of afforestation on soil
and water acidification in geologically sensitive areas by simulating forest growth (Neal et
al., 1986; Whitehead et al., 1988; Cosby et al., 1990). MAGIC, for example, incorporates
three processes relating to the impact of afforestation on acidification.
• mineral uptake by growing forests (Miller, 1981 )
• enhanced dry and occult deposition (Mayer and Ulrich, 1977)
278
• decreased water yield concentrating pollutants in surface waters (Neal et al., 1986;
Whitehead et a!., 1988)
The data used to parameterise MAGIC in this respect include the age and percentage of
mature forest cover (from local Forest Enterprise stock maps) and the amount of mature
canopy cover in the catchment (Jenkins eta!., 1997).
The data used to represent land use in the current model are derived from digital land cover
data (Fuller eta!., 1994). Using GIS the percentage of land cover type in each catchment is
determined and subsequently aggregated into 6 classes (Section 4.6.1.4). As such, there
are no process oriented data relating to land cover included in the model. Similarly, the
simple areal coverage of, for example, coniferous forestry does not include details relating,
for example, to stand age.
To optimise the land use component of the model it would be necessary to incorporate
some measure of base cation uptake, enhanced deposition and the amount of reduced
water yield among other forestry driven parameters discussed above. Ideally, these could
be quantified for each catchment. Clearly, measured data are not available on this scale.
The ability to derive a catchment weighted base cation uptake based on a priori knowledge
for different species and at different stages of growth would also be of use. The amount of
each type of vegetation in each catchment would then be sufficient to provide a measure of
base cation uptake. A similar approach might be applied to other vegetation dependant
processes and attributes. This would echo the weighting techniques proposed for the soil
attribute data described in the previous section. For the influence of liming to be explicitly
quantified the amount of lime added to each catchment would be required which is clearly
not a realistic objective.
279
More pragmatically, the model could be improved by using empirical relationships between
vegetation/land use and parameters such as base cation uptake, in tandem with the national
land cover database held at the Institute of Terrestrial Ecology. Base cation uptake has
been calculated for conifers and deciduous (on Ca rich and Ca * poor soils) woodland
(Hall, et ai, 1996) and, if available, weighted values could be determined for each
catchment using these data. Use might also be made of more detailed forestry data relating
to stand age if these data are available nationally. Forestry Authority maps could be
employed in this respect. The use of the ITE land cover data represents a practical
approach to model calibration by virtue of its’ geographical scope. However, these data are
only used to define areal coverage within a catchment. Thus the maturity of a forest stand
which, as much as the fact that the catchment is forested, determines the nature of the
response to acidic deposition. The % coniferous forest is not significant in any of the
analyses presented in Chapter 6 but the model might, however, benefit from the use of
more detailed data relating to base cation uptake from forestry and access to Forest
Authority maps.
Conversely, the % agricultural grass and arable land (LC2) drives a number of the
relationships identified in Chapters 6 and 7. This suggests that the data used can
distinguish between highly/moderately sensitive catchments and those which are non
sensitive, possibly as a result of pastoral or arable improvement. Obviously the use of LC2
as a variable does not explicitly incorporate the processes operating in grassland and
arable areas. However, given the strength of the relationship between LC2 and DCL it is
unlikely that the acquisition of more detailed data to account for the effect of liming on
catchments would improve model fit.
8.5.4 Hydrology
280
“The importance of pathways taken by percolating water through the soil to each horizon
cannot be stressed enough” (Cresser and Edwards, 1987 p. 27). Reaction products at one
depth are the reactants for the next horizon. The importance of the processes described in
the previous three sections are dependent on drainage water coming into contact with the
soil, geology and biosphere and the length of time it remains in contact. These two factors
will impact on the nature of the reactive products produced. The path taken by water
through soil is a major influence on the extent and distribution of mineral weathering.
Incoming acid deposition passing through soil and geology is likely to be buffered by the
processes discussed in earlier sections. However, water reaching streams may have
arrived as groundwater flow, throughflow, pipeflow, surface flow or peat drainage. For
example, lateral flow through organic surface horizons commonly occurs during storm
events. Each pathway will be associated with a particular set of reactions and, as such,
there is a need to know how water is likely to drain through the catchment before relating
surface water chemistry to soil, geology or land use type. The relative importance of the
influence of geology/soil/land cover is determined to a degree by the hydrological pathways.
In the first instance, it is recognised that the chemistry and mineralogy of bedrock primarily
influences base flow whereas the chemical processes at the soil/water interface determine
the composition of peak flow (Hornung, 1995a). Catchment hydrology may however, dictate
lateral flow pathways thus reducing the importance of reactions in the bedrock. Bedrock
weathering may not be increased significantly despite increasing acid deposition to the
catchment.
The response of a stream to rainwater in upland catchments depends on a number of
factors. These include;
• rainfall duration;
281
rainfall intensity;
antecedent moisture (Weyman, 1975; Whipkey and Kirkby, 1978);
infiltration capacity (Dunne, 1978);
hydraulic conductivity (Kirkby and Chorley, 1987);
thickness;
hillslope form (Anderson and Burt, 1978) and steepness.
The most important soil properties that influence the hydrological regime of a catchment are
hydraulic conductivity, soil moisture retention and pathways of water movement (Boorman
et al., 1995). Antecedent moisture, infiltration capacity and hydraulic conductivity are not
inherent catchment characteristics. They form part of a process-response feedback system
operating within the soil and do not remain constant in the short term. Hydraulic conductivity
depends on pore geometry and pore continuity. The influence of soil texture and antecedent
moisture is also important. Hydraulic conductivity is greater on lower slopes, nearer the
stream network. Values, particularly extreme ones, will depend on the inherent
characteristics of the soil and relief. Cresser et a!., (1986) examined the hydrology and
climate of British upland catchments. Rapid hydrological response to storm events is driven
by steep slopes with thin or no soil cover, limited vegetation and large drainage density
which reduces contact time with the soil. Upland catchments which presently or will in future
experience low water pH periods are generally characterised by;
• steep slopes;
• shallow mineral soils under organic rich surface horizons;
• rock outcrops;
• moderate/low soil and air temperatures (and thus modest evaporation).
282
Rainwater falling on these steep slopes initially wets up the soil and then penetrates to the
rock surface (or an impermeable horizon). Lateral flow occurs following saturation (pipe
structures increase the rate of lateral transportation). Water residence time in these
circumstances can be short.
Within the soil matrix, moisture movement is conveniently divided into matrix flow and pipe
flow. The former occurs through intergranular pores and smaller structural voids and
comprises downslope and vertical components. The water undergoes considerable
chemical modification. Pipe flow occurs through large voids (1-2 cm or more). Saturation of
soil around the pipe is a prerequisite. This is the most rapid mechanism of sub-surface
drainage water transmission. If the flow is restricted to peaty surface layers of upland
podsols the water may have little contact with the mineral soil. However, the flow of water
through the soil can be complicated by a number of factors including;
• areas of different hydraulic conductivity in the soil;
• relatively impermeable iron pans;
• indurated horizons;
• reduced permeability at depth due to overburden.
These latter three offer the potential for the bypassing of soil mineral horizons which can
increase the importance of storm flow from upper acidic organic soil horizons.
The key hydrological factors which determine catchment response to incoming deposition
are therefore a combination of dynamic processes (e.g. infiltration, hydraulic conductivity)
and micro- (pore geometry, soil texture) and macro-morphology (e.g. soil thickness,
hillslope form). In an ‘optimal’ dataset these would contribute the hydrological component.
Comprehensive data relating to these processes are likely only to be available for
experimental catchments (e.g. the Birkenes (Christopherson et al., 1982) and Allt a
283
Mharcaidh (Jenkins et ai, 1988) and as such, they would be of limited use in calibrating an
empirical model for national application.
The model in its' current formulation does not explicitly include the hydrological processes
and attributes discussed. Indirectly however, some of the key hydrological factors may be
represented by variables such as % bare rock, altitude and % peat which could provide an
indication as to whether the catchment is likely to be characterised by flashy flow regimes.
Beyond the current dataset, variables representing some of the important hydrological
parameters are available at the national scale. The soil profile databases introduced above
may be a useful starting point in terms of incorporating the morphological factors relating to
the nature of the soil. Data including porosity, bulk density and soil thickness are available
from these sources. Certain aspects of catchment morphology which influence hydrological
response can be derived from topographical maps. These would enable the use of variables
such as slope, and drainage density to be input into the model providing surrogates for the
likely flow pathways along which acidic deposition reaches the stream network.
The ‘pragmatic’ approach requires that data representing hydrological factors should have
national coverage and be able to reflect the fact that catchments vary with respect to their
ability to buffer short term changes in flow. Clay catchments tend to be flashy {i.e. they
exhibit a quick response to rainfall events) whereas permeable catchments (e.g. chalk)
show relatively little variation. The proportion of flow derived from groundwater sources is
termed the baseflow index (BSI). Areas of low BSI (<0.3) tend to be in low permeability,
hard rock catchments in upland areas. Here, the greatest temporal variations in stream
water chemistry might be found. Areas susceptible to acidification might exhibit high
correlations between BSI and the frequency of acid episodes. In the absence of detailed
284
catchment specific hydrological data representing the processes and attributes described
above the use of such an index may offer a more pragmatic approach.
Within this context, the use of Hydrology of Soil Types (HOST) database held by the
Institute of Hydrology (Boorman et al., 1995) may have provided the current model with a
hydrological component. HOST is a hydrologically based classification of the soils of the UK
based on data describing the soils and their distribution, together with the hydrological
response of catchments. The classification is underpinned by conceptual models of
processes occurring in the soil and substrates. Twenty nine classes were derived based on
11 response models. Soils were assigned, ultimately, to one of 29 classes on the basis of
the physical properties of the soil and also with reference to the hydrogeology of the
substrate. Because the classification is based on soil series it is independent of scale and,
as such, can be used with different soil datasets.
Initially the classes allow for different soil properties (e.g. a peaty soil layer) and wetness
regimes (e.g. as indicated by the presence of gleying). Further subdivisions based on other
properties or substrate geology define the 29 classes. The properties were derived from
databases of soil physical properties with input from catchment scale hydrological variables.
The end product is a base flow index (BFI) and a standard % runoff (SPR) for each soil
series. BFI is a dimensionless variable that expresses the volume of base flow as a fraction
of total flow volume. These parameters are well correlated (r = 0.75) and both are used to
calibrate and verify HOST. A high value for SPR indicates that rainfall passes through the
catchment quickly. BFI is the long term average proportion of flow occurring as base flow
and is derived using daily mean flow data. Values close to unity are observed in catchments
dominated by baseflow whereas values as low as 0.15 characterise the flashiest
catchments. BFI is the main hydrological variable used in the development of the HOST
285
classification. Soil series values for these variables can be applied to existing soil data or
digital catchment boundaries can be overlaid onto a digital HOST dataset at 1 km
resolution.
The key soil factors of hydraulic conductivity, soil moisture retention and pathways of water
movement are difficult and expensive to quantify. HOST uses data which could be used as
surrogates for direct measurement of soil hydraulic properties. Each soil series is
characterised according depth to gleying, depth to slowly permeable layer, integrated air
capacity and the presence of peaty surface layer using data derived from soil profile
descriptions. Similar criteria were used to infer and classify the hydrology of soils by other
authors (e.g. Robson and Thomasson, 1977; Bibby et ai, 1992).
Thus the HOST database represents a nationally available dataset which might be used to
incorporate the important hydrological aspects of catchment sensitivity into the model in
tandem with topographical data relating to slope and drainage density. These data were not
derived for the calibration catchments. This was an oversight during the model development
and as such, the role of hydrological pathways is not adequately represented. The use of
HOST, in tandem with more detailed relief and drainage density data, would go some
considerable way to rectify this.
8.5.5 Summary
For each of the key factors (geology, soil, land use, hydrology) optimal' variables are
suggested which might produce strong relationships between catchment and chemistry
datasets. These primarily relate to processes and attributes which would be determined for
individual catchments. As such, they would be more suitable for dynamic, process based
286
models and are not appropriate for national application. The data used in the current model
do incorporate, to varying degrees, some aspect of the processes and attributes discussed.
The soil classifications, for example, are based on combinations of soil pH, % base
saturation and mineralogy, important factors in terms of freshwater buffering. Nevertheless,
it is acknowledged that the Phase 2 data exhibit certain gaps in terms of representing key
processes governing catchment response to acid deposition. Other datasets which are also
available nationally are identified (e.g. soil physical/chemical attribute data, HOST) which
may improve the model performance statistically and model paramaterisation conceptually,
yet retaining the pragmatic ethos. This type of improvement in model paramaterisation
should form a substantial part of future development.
8.6 Model evaluation
The range of models which relate surface water chemistry to catchment attributes is
discussed in Chapters 2 and 3. These vary from catchment specific dynamic models
(Cosby etaL, 1985b; Jenkins etal., 1988) through approaches requiring progressively less
data including hybrid models (Jakeman etal., 1990), end-member modelling (Hooper at a!.,
1990; Christopherson at a!., 1990), and the PROFILE (Sverdrup at a!., 1990) and
BIRKENES models (Christopherson at a!., 1982). Other catchment studies have been
undertaken to predict alkalinity from geology (Lynch and Dise, 1985), calcium levels from
soil cation-exchange chemistry (Billett and Cresser, 1992) and buffering capacity from
runoff chemistry (Kirchener at a!., 1993). Regional approaches have also been used to
predict regional sensitivity to acidification in Scotland (Langan and Wilson, 1992), Wales
(Hornung at a!., 1990b) and throughout Great Britain (Hornung at a!., 1995a). Generally the
high resolution data used in catchment specific models (e.g. cation exchange capacity.
287
aluminium dissolution, etc) can accurately predict surface water chemistry, although
verification of long term dynamic models is difficult due to the lack of appropriate validation
data. Regional models such as those above, provide a synoptic national picture of
sensitivity. However, neither of these approaches can be used effectively for individual
catchments because of the scale at which they are applied. Process-based dynamic models
require large amounts of input data while the national sensitivity maps are regional
generalisations. Previous attempts to predict critical loads of individual catchments using
national data (e.g. Hall et al., 1995a; Kernan, 1995) have met with limited success. The
catchment model presented here uses nationally available, catchment specific data and as
such has the potential to facilitate the use of the critical loads approach as a catchment
management tool (see Section 8.7).
Throughout the discussion on methodology and data limitations the overriding theme has
been the use of the best available data to validate and calibrate the model. It would have
been possible to produce a model based on very high resolution catchment data derived
from intensive field studies in a limited number of catchments. However, the purpose of this
research was to produce a nationally applicable predictive model, and data at such high
resolution are available for very few catchments. The variables significant in terms of
predicting DCL are available nationally from hard copy maps and digital databases. In
Scotland, the soil, geology and land cover data used here is the highest resolution data
available nationally. In England and Wales, although geology and land cover data are
available at the same resolution, coverage of soil maps at 1:63,360 is not comprehensive.
Despite the reservations voiced in Section 8.4, it is clear that the model is using the best
available data {i.e. data that both exists and is 'freely' available). Therefore, with slight
modifications, it can be applied almost throughout the UK.
288
The flexibility of the model demonstrated in Chapter 7 is also a considerable strength. The
fact that a variety of predictors can be used with similar results enables decisions regarding
selection of input variables to be made on the basis of data availability. For example, if data
are available for geology and land cover but not for soil, Equation 7.3 can be used to
produce an adjusted coefficient of variation which is only slightly less (0.7499) than
Equation 7.4 which includes a soil variable (0.7829).
However, a major restriction on model applicability is imposed by the lack of sensitive sites
in the calibration dataset. This means that it is not possible to assess whether the model
can predict fine scale critical load variation at sensitive sites with the same accuracy as it
differentiates between sensitive and non-sensitive sites. An assessment of model efficacy
must conclude that, across broad gradients the model successfully differentiates between
sensitive and non-sensitive sites. However, it cannot be concluded that the model should
not be applied to sensitive sites, merely that current model performance is limited at the
sensitive end of the gradient. The model requires further modifications before applicability at
the sensitive end of the spectrum can be assessed. The next section examines how this
might be achieved together with other developments which might improve model
performance.
8.7 Further research possibilities
There are two main areas that might be considered as profitable in terms of further
research. Firstly, there is considerable scope for improvement of model prediction by
improving data input. Secondly, a comparison of model utility using data at different scales
would be useful. In addition to these improvements it would also be useful to validate the
289
model using independent data. This could be achieved, following modification, using other
catchments from the CLAG database.
8.7.1 Potential improvements in model paramaterisation
1. An increase In the scope of model application In geographical terms
The catchment model is currently calibrated using data from catchments in Scotland. The
reasons for focusing solely on Scotland are discussed in Chapter 4 and relate mainly to
data availability and the requirement to standardise data input. The analyses in Chapter 6
show that many of the catchment variables exhibit strong relationships with DCL. The
regression models in Chapter 7 demonstrate that this is reflected by the flexibility of the
predictive model in terms of its paramaterisation. Given that, to an extent, the model can be
adapted to suit data availability it is hoped that it can be applied in England and Wales to the
same effect. Although the soil classification system is different to that in Scotland (see
Chapter 4) the land cover data and geological legend are the same for England and Wales.
The regression models in Chapter 7 show that combinations of land cover and geology
have the same predictive power as combinations of land cover and soil. Additionally, the
soil sensitivity and soil critical load classifications have also been applied for Great Britain
as a whole and, although access to the classification systems was not gained in this
instance, future research might use them to widen the geographical scope of the model.
2. Calibration for sensitive sites
The uncertainty over whether a modification of the catchment model can be applied at the
sensitive end of the spectrum could be resolved by expanding the calibration database.
Using existing water chemistry datasets (e.g. Kreiser et al., 1993) it is a straightforward
matter to identify a number of sites where the critical load is less than 0.5 keq H* ha^ yr \
290
These sites can be appended to the existing calibration dataset to assess the predictive
power of the model for low critical loads. Additionally a separate analysis can be undertaken
on a screened dataset, similar to the sensitivity analysis in Chapter 6 but with a lower
threshold (e.g. Ca<200peq I ’) to assess whether the catchment data used here can predict
DCL at sensitive sites and, if not, to provide a pointer as to the type of data that might be
required to do so.
3. Modifications to the modei in the iight of modifications to the critical loads approach
The catchment model is calibrated only for the diatom model for sulphur. Other critical loads
models are discussed in Chapter 3. The diatom model has since been recalibrated to
incorporate N as an acidifying compound (Allott et al., 1995a). It is a simple matter to use
the modified DCL as a response to examine whether the relationships established between
DCL for sulphur and catchment attributes are the same if DCL for total acidity is the
response.
Contemporaneous with the diatom model is the steady state water chemistry model (HCL).
This is included as a passive variable in much of the previous analyses. Using HCL allows
critical loads to be set for individual species depending on the ANC used in the calculation
(see Chapter 3). DCL was selected as the response variable in these analyses so that an
attempt could be made to predict the baseline critical load. In the CLAG database ANC is
set to 0 which represents 50% probabiiity of damage to brown trout (CLAG Freshwaters,
1995). Analyses in Chapters 5 and 6 show a strong relationship between DCL and HCL.
HCL is also closely associated with the sensitivity gradient which characterises the Phase 2
data (Table 6.7) and as such, has strong relationships with the variables which were
eventually used to predict DCL. This is not surprising as over a broad gradient DCL and
HCL are strongly correlated (r=0.89). However, over the narrower gradient which
291
characterises more sensitive sites, although there is still a strong relationship, there is also
significant scatter (Battarbee etal., 1996). Thus the evidence suggests that, across a broad
chemical gradient, calibration for HCL (with ANC 0) should be straightforward. Future
research can assess the extent to which the model can be calibrated for HCL (ANC 0) along
the narrower gradient. Comparison could be made with a similar exercise for DCL. In
addition, it will be useful to examine the relationships between catchment attributes and
HCL with a variety of ANC values to assess whether these relationships diminish with
increasing ANC {i.e. as the response variable is altered by increasing amounts of artificial
buffering). It may be that an ANC coefficient can be built into the model to allow HCL to be
predicted for a range of ANC values.
More recent developments in the critical loads approach are discussed in Chapter 3,
including the First Order Acidity Balance (FAB) model (Downing et a!., 1993; Henriksen et
a!., 1993) which predicts maximum potential nitrate leaching for given N and S deposition
scenarios. Critical load values for these are calculated using both water chemistry data and
catchment characteristics. Although it requires much more data than the diatom and
Henri ksen models and the output terms are more complex, it is possible that values for a
simplified FAB model could be predicted using an empirical approach.
4. Improvements In predictive power with additional explanatory variables
The catchment attributes used in model development and calibration represent the most
important influences on the sensitivity of surface waters to acidification. These influences,
soil, geology and land cover, to a large part determine the buffering capacity of a catchment.
The single most important aspect of the catchment buffering system which has not been
incorporated into the model is the role of flow pathways (see Chapter 2) which determine
the length of time water is in contact with the cation exchange surfaces in the soil. Although
292
it may have been possible to incorporate some hydrological components into the predictive
model (e.g. the use of slope gradient and drainage density in the catchment to represent
catchment response to rainfall), hydrological pathways are based on a variety of spatial and
temporal factors which are difficult to represent in an empirically based model. The soil-
water response to precipitation events is likely to be spatially heterogeneous over the
catchment. The distribution of macropores within the soil, an important factor in terms of
preferred flow paths (Wheater et al., 1990), is dependent on soil structure and flora and
faunal activity which can vary throughout the catchment. Temporally, soil-water response is
governed by the magnitude of the precipitation event and the antecedent soil moisture
conditions (Bishop at a!., 1990). The Hydrology of Soil Types (HOST) database held at IH,
which classifies soil according to the dominant flow pathways (Boorman at a!., 1995),
provides data at a national level (although at 1 km resolution) and could potentially be used
to represent hydrological conditions in the catchment. Access to this database was not
available within the context of this study. However, it is hoped that any future development
of the model might incorporate a flow pathway component.
Other data exist for soils, at a national scale, relating to depth, bulk density, cation
exchange capacity and exchangeable bases. These data are available as point databases
(held at SSLRC for England and Wales and MLURI for Scotland) and are derived from a
grid network of soil pits. However these data, would not be specific to a particular catchment
but to the nearest soil pit and their use might not improve the performance of the model.
5. Davalopmant of the modal to identify critical load axcaadanca
The stated objective of this work is to predict surface water critical load. In terms of
catchment management this will enable the identification of sites which are vulnerable to
acidification and provide a guide to catchment sensitivity. The critical load for a freshwater
293
system can be said to have exceeded once the acid loading to that system is greater than
its critical load (Chapter 3). Although the aim here is not to predict where exceedance will
occur, the ability to do so should be of considerable value. Currently, critical load
exceedance maps in the UK utilise interpolated S and N deposition data presented at 20knf
resolution (Allott et al. 1995b). These maps are used in tandem with the grid based critical
loads maps to provide estimates for exceedance. However, if the requirement is to assess
whether surface waters within a specific catchment have been exceeded then the grid
based deposition data is of limited value (see Section 8.4.3). Estimates of the uncertainty of
total 8 input to a 20km square are as high as ±80% (Smith et ai, 1995). So although
predictions of critical load could be matched with the deposition data to produce
exceedance predictions, the accuracy of such predictions would be extremely unreliable.
Future refinements of deposition modelling (incorporating altitude, aspect, wind direction,
seeder feeding and cloud droplet deposition and land cover effects) may enable more
accurate estimates of deposition values that could then be used to predict exceedance for
individual catchments.
8.7.2 Analysis of catchment data at different spatial resolutions
The effect that using catchment data at different resolutions has on the explanatory power of
the model has, to an extent, been addressed during model development (Chapter 6 ).
Appendix 6 presents the results of using land use data at varying spatial resolutions. The
analysis is repeated using different hierarchical classifications for soil data {i.e. at the soil
association and soil map unit scales). It would be useful to examine the effects of using data
at different scales for a wider range of variables. For example, the model could be
calibrated across a range of data resolutions. Mapped soil data is available nationally (in
Scotland) at 1:250,000 and 1:50,000/1:63,360. Limited coverage is available at 1:25,000
294
and 1 :1 0 , 0 0 0 resolution while fine scale soil survey has been undertaken in individual
catchments (e.g. Billet and Cresser, 1992; MacPhee 1995;). Similarly geology maps have
been produced at various resolutions. The predictive power of the model using data at these
varying resolutions could be examined to assess whether it is substantially affected by low
resolution predictors. Clearly if the explanation at lower resolutions does not differ from that
at high resolution then the use of low resolution data would be preferable given its wider
availability and the logistical benefits inherent in using less detailed data. Applying the
model to catchments with fine resolution data would enable the importance of, for example,
detailed soil chemistry and flow pathway data, to be examined. The value of weighting
areas closer to stream networks could also be assessed. It is possible that the use of lower
resolution data would differentiate between sensitive and non-sensitive sites while higher
resolution data may be required to predict critical load for more sensitive sites. A two stage
model may therefore be possible which initially identifies sensitive sites and then, through
the use of different explanatory data, provides predictions of critical load below, for example
2 .0 keq ha^ yr ’
8.7.3 Potential for predicting other measures of sensitivity and acid -base status
The primary objective of this thesis is to present an empirically derived model which will
enable critical loads to be predicted from nationally available data representing catchment
processes and attributes. Analyses describing and quantifying the relationships between
diatom critical load (DCL) and catchment characteristics are presented in Chapters 5, 6 and
7. Prior to these, there are a series of exploratory analyses of the water chemistry data
together with an examination of the relationships between these and the catchment
variables described in Chapter 4. The Phase 1 chemistry data comprised variables in the
CLAG critical load database (CLAG, 1995) and the Phase 2 data were chosen to allow
295
comparison to be made between national data (Phase 1) and the calibration dataset (Phase
2 ). This section considers how the analyses of the full chemistry dataset fits into the overall
aim of the research and how the results of the full suite of determinands might be used to
broaden the scope of the model.
Principal components analysis (PCA) was used to summarise the major patte ms of
variation in water chemistry for both the national dataset (Phase 1 - Fig. 5.1) and the
calibration data (Phase 2 - Fig 6.2). This holistic approach was initially adopted so that the
structure of the chemical database could be examined and the distribution of sites along the
major chemical gradients assessed. These analyses also enabled the nature of the Phase 1
and Phase 2 datasets to be evaluated and compared. Collinearity between variables was
highlighted, showing which variables are most useful for characterising the chemical
composition of the dataset.
In both datasets the structure of the data was dominated by a gradient of ionic strength, the
primary axis (Axis 1) identified by PCA. Summary statistics of the Phase 1 data illustrated
the extended lengths of the chemical gradient along which the CLAG sites are spread
(Table 5.1). This gradient was considerably shorter for the Phase 2 data, a result of the
wider geographical spread of the Phase 1 sites. Highly dilute sites were poorly represented
in the Phase 2 dataset. The implications of this for model calibration are discussed in
Section 8.4.1. The diatom critical load is the variable most associated with the primary axis,
particularly in terms of the Phase 2 data, although many determinands are highly correlated
with both DCL and Axis 1 (Table 6.2). The dominance of this axis, which appears to reflect
variation in sensitivity (as defined by critical load) and/or acid -base status (pH, alkalinity)
raises the possibility, given the results of the regression analyses in Chapter 7, that other
important determinands can be predicted using catchment characteristics.
296
A second gradient was identified along PCA Axis 2 for both datasets. This was associated
with aluminium and organic acidity in Phase 1 but included these and sea-salts in Phase 2 .
PCA biplots of both datasets showed that, although sites range predominantly along Axis 1,
sites with lower Axis 1 scores {i.e., with low Ca " concentrations and DCL values) show
some variation along Axis 2. This secondary axis contrasts sites characterised by high
aluminium and Abs-250 values with low pH sites. The suggestion here is that acid-base
status of these sites is not being driven solely by the geochemistry of the catchment but also
by some combination of aluminium mobilisation, organic acidity and, potentially, the
influence of marine salts.
PCA undertaken on sensitive subsets of the data (Phase 1, Ca <200peq l \ Phase 1, Ca *
<400peq I"") resulted in a major alteration of the structure of the data so that the relative
importance of the two gradients was reversed and the sea-salt/organic acidity gradient
became dominant. This has implications for model performance at the more sensitive end
of the spectrum. Catchment variables strongly associated with the broader sensitivity
gradient may not be able to explain variation among streams for which sensitivity is not the
primary gradient of variation. This supposition is borne out by the redundancy analyses
(PDA) on the sensitive subsets (Sections 5.3.1 and 6.4.3.1). Model development was
geared primarily towards predicting chemistry based on geochemical variation within
catchments. The reversal of Axes 1 and 2 during the analyses of the more sensitive subsets
shows that, at this end of the spectrum, the complex processes involving organics,
aluminium and sea-salts are of crucial importance.
297
Dissolved organic matter in some upland catchments tends to increase during major
storms. The increased proportion of drainage water which passes only through the organic
surface horizons without entering the mineral layers can often lead to a temporary fall in pH.
These changes are a result of changing hydrological pathways (see Section 8.6.4). Water
draining through the more acidic surface organic horizons contributes disproportionately to
discharge levels. Flushes of acidity occurring in this way are often associated with elevated
total organic carbon (TOC). However, it is also possible that, as acidity increases, TOC in
water draining vertically from the organic horizon decreases as a result of organic anion
dissociation (Davis et al., 1985). The long term trend in TOC in response to freshwater
acidification may be a decrease (unless the depth of the organic horizon increases) but acid
flushes may still be associated with high TOC (unless the flush is due to direct snow input).
Observations have shown that, at sites in Sweden, acidity was more strongly correlated
with TOC than with sulphate (Kullberg at a!., 1993). At these sites therefore, it is possible
that TOC is more important in terms of sensitivity than DCL, which does not incorporate an
organic acidity component. Water dominated by organic acids is much less toxic than
mineral acids because soil waters retain the capacity to complex iron and aluminium in
solution and therefore depress the concentration of ionic species, particularly Al^. It may be
possible therefore, to distinguish sites acidified anthropogenically from those where natural
acidity dominates on the basis of labile aluminium concentration (Harriman et a!., 1988). In
terms of model development the choice of explanatory variables can be tailored to take this
into account with more emphasis being placed on variables correlated with Al^. The
secondary gradient associated with aluminium and Abs-250 observed during the Phase 2
analyses appeared to be linked to distance from sea and altitude rather than the soil,
geology and land use variables driving the response of Axis 1. In addition, the positive
correlation between % forestry and aluminium would appear to reflect the increased
scavenging of sulphate in forested catchments and the concomitant mobilisation of
298
aluminium (Harriman and Morrison, 1982). These relationships suggest there may be some
potential for predicting responses along the second axis although further data are likely to
be required.
The impact of major sea-salt incursions into freshwater systems can markedly reduce pH
levels during storm events (Harriman and Pugh, 1994). The increase in acidity results from
exchange processes on soil exchange sites which become saturated by excess Na and,
as a consequence, release an equivalent amount of other cations, including H (Harriman
and Wells, 1995). This effect is exacerbated in catchments where soils have already been
acidified by atmospheric deposition (Harriman and Pugh, 1994).
Following PCA of the chemistry data the key question was whether the primary gradient
could be explained by the variables representing catchment characteristics. To address this,
the chemistry data were constrained to be linear combinations of the catchment data using
redundancy analysis (RDA) (Sections 5.2.3.2 and 6.3.1). This enabled the variance in the
full chemistry data to be quantified prior to focusing on the response of DCL alone. It also
provided a useful summary of the catchment influence on chemical composition. The
structure of the data following RDA in Phase 2 shows that the chemistry determinands
retain the internal relationships identified after PCA of the chemistry data only (Fig. 6.2).
DCL remains highly correlated with the primary axis of variation along with a number of
other determinands including the base cations, conductivity, alkalinity, SO/' and NO/ (see
biplot scores. Table 6.7) which is to be expected given the high correlations between most
of these variables (Table A6.1). Although not the primary objective of the research, these
analyses are of some use in examining the role of other water quality parameters within an
acidification context. RDA of the wider chemical gradient enables the overall
catchment/chemistry relationships to be examined (see Section 6.3.1). This shows the
299
potential for predicting other indicators of sensitivity (e.g. Ca % ANC) which may be useful
where insufficient data exist to calculate critical load values or acid base status (e.g. pH,
alkalinity). This might, for exampie, be useful for lake classification exercises. In Phase 1,
sulphur and nitrogen deposition are shown to be important variables in terms of explaining
variation in overall chemical composition, but less so when DOL is the response. This
reflects how deposition inputs and catchment processes are equally important in controlling
the acid-base status of surface waters (Cosby etal., 1990; Nilsson, 1993). Consideration of
acid-base status rather than sensitivity should therefore encompass deposition regimes as
well as catchment characteristics.
The overview of relationships within the chemistry datasets and between these and the
environmental variables were initially undertaken so that general patte ms of variation could
be recognised. Within the context of model development it was observed that the variation
in DCL (explained substantively by the catchment variables in subsequent analyses) was a
reflection of the dominant chemical gradient exhibited by both the national and calibration
datasets. The collinearity between DCL and other indicators, both of sensitivity and acid-
base status, suggests that the catchment data used might also be used to predict other key
variables driven by the geochemical status of the catchment. However, where the Ca *
gradient is reduced and emphasis is placed on more sensitive sites, chemical response is
dominated by other determinands including, aiuminium, Abs-250 and marine salts. These
determinands are also influential in, or indicative of, the acid-base status of surface waters.
This gradient appears to be independent of sensitivity. Thus, in terms of model
development, it is important to be aware that freshwater systems are distributed not only
across a mineral acid-base gradient but that a significant organic acidity and aluminium
gradient exists independently. Thus, an examination of the overall structure of the data
might be of considerable use in identifying reasons why the model performs less well in
300
some areas (e.g. those where the geochemical signal from soil and geology is masked by
forestry or organic acidity) or explaining the presence of outliers.
Prior to undertaking analyses on individual variables, a core set of determinands should be
examined to characterise the chemistry data and to assess the influences of different
catchment attributes on overall chemical composition. The mineral acid-base and sensitivity
gradient should be represented by pH, Ca , alkalinity and DCL. To identify the influence of
organic acidity, TOC (or Abs-250 as a surrogate) should be included. Determinands which
can be used to infer the impact of deposition regimes (SO/' and, where there is no
agricultural source, NOg ) and forestry (Al^) on surface water acidity should also be included
as part of this preliminary analysis.
8. 7. 4. Model evaluation and further development using national data
During the Phase 1 analyses, grid based catchment data were employed with most of the
variables relating to the 1 km grid square in which the site is situated. During Phase 2,
catchment boundaries were used to define the extent and nature of the predictor variables
(see Chapter 4). Despite these differences in data resolution, redundancy analysis (RDA),
highlights the influence of certain variables common to both the preliminary (Phase 1) data
(Fig. 5.5) and the calibration (Phase 2) dataset (Fig. 6.5). These include soil critical load, %
arable land, % upland and altitude. The Phase 2 dataset comprises a relatively narrow
gradient of site types in terms of chemical composition (see Section 6.2.1.1). As such, it is
difficult to interpret fully some of the relationships identified between water chemistry (and
critical load) and catchment characteristics. Conversely, the broader geographical scope
and wider chemical gradients characterising the Phase 1 dataset suggests that there is
scope for analyses of subsets of these data to evaluate some of the conclusions and
301
interpretations drawn from the statistical analyses presented in Chapters 6 and 7. A number
of potential avenues of exploration are discussed here, initially in terms of general model
development and subsequently as a means of testing some of the hypotheses and
interpretations generated by the Phase 2 analyses.
The use of Phase 1 data to support Phase 2 analyses
1. To examine the response of streams and lakes to storms in upland catchments it might
be useful to examine sites above 300 or 400 m separately rather than using Ca<400peq I ’
as a cut-off (Section 6.4). SO/ is not strongly adsorbed on the exchange complex in most
upland UK soils (see section 8.6.1.2) as they are relatively young and not deeply
weathered. By looking at sites above and below, for example, 300 m it may be possible to
examine whether the relationships between S deposition, SO/ in freshwaters and
catchment characteristics differ given the nature of processes operating at different altitudes
(e.g. so/ adsorption and leaching - see Section 8.5.2).
2. It is argued that the model is being driven, to an extent, by catchments which have high
critical loads due to their arable nature. This may mask the influence of other variables.
Useful analyses might use the Phase 1 data to partial out the effects of land cover type on
critical load. Although a variance partitioning exercise was undertaken on the Phase 2
calibration dataset it would be useful to undertake this analysis on the larger dataset and
focus on catchments where there is no agricultural influence.
3. RDA with forward selection shows that forestry did not contribute significantly to the
variation in critical load (Table 6.10) or chemical composition in general (Table 6 .8 a). This is
counter to much of the work undertaken comparing forested with moorland catchments
(Harriman and Morrison. 1982; Stoner et ai, 1984). By focusing on catchments with no
302
arable land cover, the Phase 1 data might be used to compare water chemistry specifically
between moorland and forested catchments and examine whether forestry has a significant
impact on water chemistry once the heavily influential lowland catchments have been
omitted.
4. One area of model development neglected in the research presented here is validation
using an independent dataset. The calibration dataset is too small to enable such an
exercise. However, the much larger Phase 1 dataset could be divided into calibration and
validation datasets. Multiple regression results from analyses of the calibration set could be
tested on the validation dataset. Although this would not reveal information specifically
relating to the Phase 2 dataset, the similarities in the structure of the Phase 1 and 2 data are
such that the conclusions drawn from such an exercise on the former might allow useful
inferences to be drawn concerning the latter.
5. Multiple regression applied to the national dataset would be of considerable use in
identifying spatial pattems in the relationships between catchment characteristics and
surface water chemistry. Maps of regression residuals would enable regions/areas where
the model is over- or under-fit to be identified or systematic error to be detected. Similarly, if
the model was applied to sites in regional blocks it would be possible to assess the impact
of regional variations in the importance of various input parameters. Additionally, it would be
to possible hold certain parameters constant and assess how DCL variation in areas with,
for example, similar underlying geology, might be explained. Such an approach would
benefit enormously from the use of the national soils databases discussed in Section 8.5.
6 . Phase 1 data might usefully be employed to explore further the relationships between
catchment characteristics and diatom critical load (DCL) at sites where the latter is less
303
than 2.0 keq H* ha"" yr'\ The Phase 2 dataset is lacking in very sensitive sites with low
critical loads {i.e., <0.5 keq ha^ yr ’). Indeed, most sites have DCL values greater than
2.0 keq ha yr^ (See section 6.2.1.1). The Phase 1 analysis focusing on a reduced
gradient of more sensitive sites (Ca < 2 0 0 ^eq I"") highlights the differences between
catchment characteristics and water chemistry compared to analysis of the full dataset (see
Section 5.3). More detailed analyses of these data (e.g. partialling out the influence of
variables such as altitude and rainfall) may provide pointers as to the type of variable likely
to be most important at the more sensitive sites (see below).
The use of Phase 1 data to test / validate hypotheses and interpretations generated during
Phase 2
1. RDA with forward selection undertaken using the full suite of chemical determinands
indicated that sites with high rainfall values are characterised by more sensitive water
chemistry. It is suggested that this may be due to the elevated levels of rainfall in upland
areas which are, in turn, characterised by poorly buffered soils. The Phase 1 dataset could
be used to partition the variation in water chemistry so that the variance explained by
altitude, rainfall and soil critical load, independent of each other, can be assessed across a
broader gradient of site types. If the variation explained exclusively by rainfall was found to
be non-significant, it might be assumed that the relationship with water chemistry was a
function of altitude and underlying soils (which co-vary with rainfall). A high level of
explanation might suggest more causality with high rainfall levels leading to leaching out of
base cations from exchange sites in the soil faster than they can be replenished by mineral
weathering.
304
2. Phase 2 analyses show a strong positive relationship between soil critical load (SCL) and
DCL (Section 6.3.3). This relationship is based on the relatively narrow gradient of
sensitivity which characterises these data. The strength of this relationship might be tested
across the wider sensitivity gradient exhibited by the Phase 1 dataset using Analysis of
variance (ANOVA). ANOVA might also be employed using subsets of these data to assess
DOL/SCL relationships for progressively more sensitive sites.
3. The relationship between DCL and the amount of arable/managed grassland (LC2) in the
catchment is linear for non-sensitive sites but exhibits more of a scatter at more sensitive
sites (Section 6.3.3) suggesting that other factors might be more important. By running the
analyses on the Phase 1 data, eliminating sites where LC2 is the land cover type, it may be
possible to determine what these might be. The creation of similar subsets of the Phase 1
data (e.g. LC2 only, forested catchments only, upland moor only), even though much of the
data are not catchment specific, could provide some indication as to the variables driving
catchment/chemistry relationships under different land cover types.
4. An attempt has been made to examine the relationships between catchment attributes
using RDA (Section 6.4) and multiple regression (Section 7.3) for more sensitive sites
(defined by waters with Ca < 400 peq l' ). Principal components analysis was used to
examine the structure of the chemistry data for the ‘sensitive’ sites (Section 6.4.1) and
across the full sensitivity range (Section 6.2.1.2). Subsets of the Phase 1 data would allow
the interrelationships between chemical determinands to be examined under a variety of
conditions (in moorland, forested, arable catchments and under varying altitude bands).
This approach would show how the chemical gradient changes under different land cover
305
and altitude constraints and may suggest where the model requires more detailed
parameterisation.
5. The difficulties inherent in using spot samples to calibrate a model seeking to predict DCL
are discussed in Sections 4.3 and 8.4.1. It has been noted that seasonal variation of stream
critical loads are greater than those exhibited by lakes (Harriman et ai, 1995b). Given that
the Phase 2 dataset comprised streams only, it would be useful to examine relationships
between catchment attributes and water chemistry for streams and lakes separately.
Although the Phase 1 data predominantly consist of lake sites there are sufficient numbers
of stream sites included (CLAG Freshwaters, 1995) to enable a valid comparison to be
made. This would be particularly valuable in areas with similar underlying geology and soil
where differences in DCL values for streams and lakes may simply be due to the greater
stability of lake systems and the nature of flow / chemistry relationships (e.g. Creasey at
ai, 1986; Kreiser at ai, 1995).
This section has shown where the Phase 1 data might be used to test the assumptions and
interpretations generated by the Phase 2 analyses. It would be difficult to use the Phase 1
data comprehensively this way because of the differences in the nature of the catchment
variables. Although the ranges of the response data (/.e., water chemistry/DCL) are wider,
the catchment data are more limited. Nevertheless, while Phase 1 constitutes a preliminary
analysis its’ use as a validation dataset might provide some useful pointers, particularly in
terms of future model development.
8.8 Implications of model improvement for catchment management
306
A wide variety of organisations in Britain are concerned with surface water acidification.
These include conservation bodies (e.g. English Nature, Scottish Natural Heritage, Royal
Society for the Protection of Birds), land use planners (e.g. the Forestry Authority,
environmental regulators (e.g. the Environment Agency) and environmental policy makers
(e.g. Department of Environment (DoE), Welsh Office, Scottish Office). Environmental
policy makers such as DoE use the critical loads maps as a regional guide to acidification
problems. However, many of these organisations are concerned with acidification at a
catchment scale and attempts have been made to use the national data for individual
catchments (Forestry Authority, 1993). Assessments of critical loads at a catchment scale
currently requires costly fieldwork and laboratory analysis.
The model presented here offers scope for extensive, catchment based, critical load
assessments to be made. At present this would appear to be confined to identifying those
sites which may, or may not, be sensitive to acidification. This approach has foundations in
work on national sensitivity maps (Langan and Wilson, 1992; Hornung etal., 1990b, 1995a)
and previous attempts to predict critical loads (Hall at a!., 1995a; Kernan. 1995). However,
these approaches use data which do not relate to specific catchments. The sensitivity maps
provide a regional guide to sensitivity whereas the approach utilised by Hall et a!., (1995a)
is similar in principle to that adopted during the Phase 1 analysis (Chapter 5). A
comparison of the results obtained during Phase 1 (which used surrogates for catchment
data) and Phase 2 (using catchment specific data) shows that when the latter are employed,
the level of explanation is considerably higher.
However, the limitations of model for more sensitive sites is a major concern. There is a
requirement to know whether deposition in these sensitive catchments will be in excess of
the catchments ability to absorb the impact. Sulphur deposition levels in sensitive areas
307
generally vary across a fairly narrow range (CLAG Freshwaters, 1995). Therefore minor
variations in critical loads values at more sensitive sites can have a substantial bearing on
whether the site is exceeded. There is a requirement to know whether relatively low
deposition levels in sensitive areas are in excess of the capability of catchments in these
areas to absorb the impact. In its present state the predictive model cannot provide these
answers.
Even within these constraints the model can still be used for a variety of purposes. For
example, sampling programmes could be targeted to sensitive sites by use of a desk study
rather than undertaking extensive fieldwork. However, if the modifications described in the
above section are successfully implemented, particularly the two stage approach, then it is
to be hoped that the model can then be used successfully across the complete range of
catchment sensitivities. National datasets are already in existence, both digital and hard
copy. These may be purchased enabling organisations involved in catchment scale
management to adopt the critical loads approach at a national scale particularly if the use of
lower scale digital data is seen to maintain the levels of explanation achieved at
1:50,000/1:63,360 scales. The flexibility of the model, in terms of predictor inputs, means
that the model can be applied using the data most readily available to individual users.
In addition to individual catchment assessment, the approach could also facilitate stock-at-
risk assessments both nationally and in nested catchment systems. In the former case, for
example, conservation bodies may wish to know how many SSSIs are at risk from
acidification. In the latter instance the areas within a single catchment that are vulnerable
could be identified by defining a series of sub-catchments for input into the model.
8.9 Conclusions
308
The discussion has focused on a range of issues. However, several key conclusions can be
drawn from the study as a whole. Clear statistical relationships between catchment
variables and water chemistry were identified. In terms of (diatom) critical load the key
variables were i) % arable land in the catchment, ii) weighted soil critical load, ill) % granite
and gneisses in the catchment, iv) % limestones and chalk in the catchment and v)
weighted soil water concentration. Prediction models were developed using these
variables to predict diatom critical load varied between 0.7 and 0 .8 ). It was possible to
use different combinations of predictors with negligible loss of predictive power. Therefore
the model can be altered depending on data availability. The approach has the potential to
allow discrimination between sensitive and non-sensitive sites at a catchment scale
providing a management tool to facilitate, for example, impact assessment and stock-at-risk
studies. There are a number of limitations to model application. The most important of these
are related to the nature of the predictor variables and the uncertainty of model utility at
more sensitive sites. Recalibration using different data may improve the model in these
respects.
309
REFERENCES
Adams, W.A. and Evans, G.M. (1989) Effects of lime applications to parts of an upland catchment on soil properties and the chemistry of drainage waters. Journal of Soil Science 40, 585-597.
Adamson, J.K. and Hornung, M. (1990) The Effect of Clearfelling a Sitka Spruce {Picea sitchensis) Plantation on Solute Concentrations in Drainage Water. Journal of Hydrology 116, 287-298.
Aitken, M.A. (1974) Simultaneous inference and choice of variable subsets in multiple regression. Technometrics, 16, 221-228.
Albert, O.T. and Bergstad, O.A. (1993) Temporal and spatial variation in the species composition of trawl samples from a demersal fish community. Journal of Fish Biology 43, 209- 222.
Ailott, T.H., Battarbee, R,W., Curtis C., Kreiser, A.M., Juggins, S. and Harriman, R. (1995a) An Empirical Model of Critical Acidity Loads for Surface Waters based on Palaeolimnological Data. In Hornung, M., Sutton, M. and Wilson, R.B. (Eds.) Mapping and modelling of critical loads - a workshop report. Report of a workshop held at Grange-over-Sands, Cumbria. UN-ECE Convention on Long Range Transboundary Air Pollution, Working Group for Effects, 24-26 October, 1994. ITE, Edinburgh, pp. 50-54.
Ailott, T.E.H., Battarbee, R.W, Curtis, C., Harriman, R., Hail, J., Bull, K. and Metcalfe, S.E.(1995b) Sulphur deposition, present and projected critical load exceedances. In CLAG Freshwaters: Critical Loads of Acid Deposition for UK Freshwaters. A Reporl to the Department of the Environment from the Critical Loads Advisory Group, Freshwaters sub-group. Environmental Change Research Centre, London, pp. 25-34.
Ailott, T.E.H., Curtis, C., Hall, J., Harriman, R. and Battarbee, R.W. (1995c) The Impact of Nitrogen Deposition on Upland Surface Waters in Great Britain: A Regional Assessment of Nitrate Leaching. Water, Air and Soil Pollution 85, 297-302.
Ailott, T.E.H., Golding, P.N.E. and Harriman, R. (1995d) A Palaeolimnological Assessment of the Impact of Acid Deposition on Surface Waters in North-West Scotland, a Region of High Sea- Salt Inputs. Water, Air and Soil Pollution 85, 2425-2430.
Appleberg, M., Lingdell, P-A. and Andrén. C. (1995) Integrated Studies of the Effects of Liming Acidified Waters (ISELEW-Programme). Water, Air and Soil Pollution 85, 883-888.
ApSimon, H.M., Kruse, M. and Bell, J.N.B. (1987) Ammonia emissions and their role in acid deposition. Atmospheric Environment 2^, 1939-1946.
Avery, B.W. (1987) Soil Survey Methods: A Review. Soil Survey Technical Monograph No. 18, Silsoe.
Avery, B.W. (1980) Soil classification for England and Wales. (Higher categories). Soil Survey Technical Monograph No. 14. Harpenden, UK: Soil Survey of England and Wales.
Bache, B.W. (1984) Soil water interactions. Philosophical Transactions of the Royal Society. B305, 393-407.
Bache, B.W. (1985) Soil acidification and aluminium mobility. Soil Use and Management 10- 14.
310
Bache, B.W. (1986) Aluminium mobilization in soils and waters. Journal of Geological Society, London 143, 699-706.
Bader, M.J. and Roach , W.T. (1977) Orographic rainfall in warm sectors of depression. Quarterly Journal of the Royal Meteorological Society ^03, 269-280.
Bajpai, A.C., Calus, I.M. and Fairley, J.A. (1992) Descriptive statistical techniques. In Hewitt, C.N. (Ed), Methods of environmental data analysis. Elsevier Appiied Science; London pp. 1-35.
Barnett, V. and Lewis, T. (1994) Outliers In statistical data (3rd edition). Wiley, Chichester.
Barrett, O.P. and Irwin, J.G. (Eds.) (1983) Acid Deposition In the United Kingdom. Warren Spring Laboratory Report, December, 1983.
Battarbee, R. W. (1990) The causes of lake deposition with special reference to the role of acid deposition. Philosophical Transactions of the Royal Society, London B 327, 339-347.
Battarbee, R.W., Flower, R.J., Stevenson, A C. and Rippey, B. (1985) Lake acidification in Galloway: a palaeoecological test of competing hypotheses. Nature 314, 350-352.
Battarbee, R.W., Anderson, N.J., Appleby, P.B., Flower, R.J., Fritz, S.C., Haworth, E.Y., Higgitt, S., Jones, V.J., Kreiser, A., Munro, M.A.R., Natkanski, K., Oidfield, F., Patrick, S.T., Richardson, N.G., Rippey, B. and Stevenson, A.C. (1988) Lake Acidification In the United Kingdom 1800-1986; Evidence from Analysis of Lake Sediments Report prepared for the DoE under contracts PECD 7/7/139 (UGL) and PECD 7/7/142 (FBA), Palaeoecology Research Unit, University College London, ENSIS Publishing, London.
Battarbee, R.W., Ailott, T.E.H., Birks, H.J.B., Bull, K., Flower, R.J, Jenkins, A., Hail, J., Harriman, R., Juggins, S., Kreiser, A., Ormerod, S.J., Patrick, S.T., Rose, N., and Weatherley, N. (1992a) Critical loads and acid deposition for UK freshwaters. An interim report to the Department of the Environment (Research Paper no. 5) London: Environmental Change Research Centre, University College London.
Battarbee, R.W., Kreiser, A.M., Harriman, R., Bull, K., Jenkins A. and Ormerod, S.J. (1992b) The distribution of nitrate In UK surface waters and Its Implication for calculating critical loads: a preliminary assessment. Environmental Change Research Centre, Research Paper No. 2. University College London.
Battarbee, R.W., Aliott, T.E.H., Kreiser, A.M. and Juggins, S. (1993) Setting critical loads for UK surface waters: the diatom model. In Hornung, M. and Skeffington R.A. Critical loads: concepts and applications. ITE Symposium No.28, HMSO: London, pp. 99-102.
Battarbee, R.W., Aliott, T.E.H., Harriman, R., Juggins, S., Kreiser, A.M., Ormerod, S.J. and Patrick S. (1994) Estimating critical acidity loads for UK freshwaters. In Battarbee, R.W. (Ed.) Acid rain and Its Impact: the critical loads debate. Proceedings of a conference held at the Environmental Change Research Centre, University College London. ENSIS Publishing, London, pp. 53-60.
Battarbee, R.W., Aiiott, T.E.H., Juggins, S. and Kreiser, A.M. (1995) Estimating the Base Critical Load: The Diatom Modei. In CLAG Freshwaters, Critical Loads of Acid Deposition for United Kingdom Freshwaters. Department of Environment.
Battarbee, R.W., Aliott, T.E.H., Juggins, S., Kreiser, A.M., Curtis, C. and Harriman, R., (1996) Critical loads of acidity to surface waters - an empirical diatom-based paiaeoiimnoiogical model. Amblo 25, 366-369.
Bauer, S. and Feger, K.H. (1992) Importance of natural soil processes relative to atmospheric deposition in the mobility of aluminium in forested watersheds of the Biack Forest. Environmental
311
Pollution 77, 99-105.
Beamish R.J. and Harvey, H.H. (1972) Acidification of the La Cloche Mountain lakes, Ontario and resulting fish mortalities. Journal Fisheries Board of Canada 29, 1131-1143.
Beckett, P.H.T. and Burrough, P.A. (1971) The Relation Between Cost and Utility in Soil Survey IV. Comparison of the Utilities of Soil Maps Produced by Different Survey Procedures, and to Different Scales. Journal of Soil Science 22, 466-480.
Bergeron, T. (1965) On the low-level redistribution of atmospheric water caused by orography. Supplementary Proceedings International Conference on Cloud Physics, Tokyo, May 1965, pp. 96-100.
Bie, S.W. and Beckett, P.H.T. (1971) Quality Control in Soil Survey. Introduction: I. The Choice of Mapping Unit. Journal of Soil Science 22, 32-49.
Billett, M.F. and Cresser, M.S. (1992) Predicting stream-water quality using catchment characteristics and soil chemical characteristics. Environmental Pollution 77, 263-268.
Billet, M.F., Parker-Jervis, F., Fitzpatrick, E.A. and Cresser, M.S. (1990) Forest soil chemical changes between 1949/50 and 1987. Journal of Soil Science 41, 133-145.
Bird, S.C., Brown, S.J. and Vaughan, E. (1990a) The influence of land management on stream water chemistry. In Edwards, R.W., Gee, A.S. and Stoner, J.H. (Eds.) Acid Waters in Wales. Klewer Academic Publishers, The Netherlands, pp. 241-254.
Bird, S.C., Walsh, R.P.D. and Littlewood, I.G. (1990b) Catchment characteristics and basin hydrology: their effects on stream water acidity. In Edwards, R.W., Gee, A.S. and Stoner, J.H. (Eds.) Acid Waters in Wales. Klewer Academic Publishers, The Netherlands, pp. 203-222.
Birks, H.J.B. (1993) Is the hypothesis of survival on glacial nunataks necessary to explain the present-day distributions of Norwegian mountain plants. Phytocoenogla 23, 399-426.
Bishop, K.H., Grip, H. and Piggott, E.H. (1990a) Spate-specific flow pathways in an episodically acid stream. In Mason, B.A. (Ed.) The Surface Waters Acidification Programme, Cambridge University Press, Cambridge, pp. 107-120.
Bishop, K.H., Grip, H. and O’Neill, A. (1990b) The origins of acid runoff in a hillslope during storm events. Joumal of Hydrology 36-61.
Boon, R. and Kay, D. (1990) Recent land use change. In Edwards, R.W., Gee, A.S. and Stoner, J.H. (Eds.) Acid Waters in Wales. Klewer Academic Publishers, The Netherlands pp. 67-80.
Boorman, D.B., Hollis, J.M. and Lilly, A. (1995) Report No. 26, Hydrology of soil types: a hydrologlcally-based classification of soils of the United Kingdom. Institute of Hydrology, Wallingford.
Borcard, D. and Legendre, P. (1994) Environmental control and spatial structure in ecological communities: an example using oribatid mites (Acari Orlbates). Environmental and Ecological Studies 1, 37-61.
Borcard, D., Legendre, P. and Drapeau, P. (1992) Partialling out the spatial component of ecological variation. Ecology 73, 1045-1055.
Boyle, J.F., Smayda, T. and Birks, H.J.B. (1989) The influence of catchment characteristics on the chemical composition of chronically acidified lakes in southern Norway. Botanisk Institutt, Universitetet i Bergen, Rapport 48.
312
Brady, N.C. (1990) The nature and properties of soils (10th edition) Macmillan Publishing Company, New York.
Brakke, D.F., Henriksen, A. and Norton, S.A. (1990) A variable F-factor to explain changes in base cation concentrations as a function of strong acid deposition. Verheissungen der Internationalen Vereinigung fuer gasammten Limnologie 24, 146-149.
Brakke, D.F. and Henriksen, A. (1989) Uncertainties in Using Empirical, Steady State Models to Estimate Critical Loads of Strong Acids to Lakes. In Kamari, J., Brakke, D.F., Jenkins, A., Norton, S.A. and Wright, R.F. (Eds.) Regional Acidification Models: Geographic Extent and Time Development, Springer-Verlag, Berlin pp. 45-54.
Bricker, O.P. (1986) Geochemical investigations of selected Eastern United States watersheds affected by acid deposition. Journal of the Geological Society, London 143, 621-626.
Brodin, Y-W. and Kuylenstierna, J.C.I. (1992) Acidification and Critical Loads in Nordic Countries: A background. Ambio2^, 332-338.
Buijsman, E., Maas, J.F.M. and Asman, W.A.H. (1987) Anthropogenic NH3 emissions in Europe. Atmospheric Environment 2^, 1009-22.
Bull, K.R. (1991) The critical loads/levels approach to gaseous pollutant emission control. Environmental Pollution 69, 105-123.
Bull, K.R. (1992) An Introduction to critical loads Environmental Pollution 77, 173-176.
Bull, K.R. (1995) Critical Loads - Possibilities and Constraints. Water, Air and Soil Pollution 85, 201- 212 .
Bull, K.R., Brown, M.J., Dyke, H., Eversham, B.C., Fuller, R.M., Hornung, M., Howard, P.O., Rodwell, J. and Roy, D.B. (1995) Critical loads for nitrogen deposition for Great Britain. Water, Air and Soil Pollution 85, 2527-2532.
Bunce, R.G.H. and Heal, O.W. (1984) Landscape evaluation and the impact of changing land- use on the rural environment: the problem and an approach. In Roberts, R.D. and Roberts, T.M. (Eds.) Planning and Ecology. Chapman and Hall, London, pp. 164-88.
Bunce, R.G.H. and Howard, P.O. (1992) Aggregation of ITE Land Classes for Great Britain into broad groups (Merlewood research development paper). Grange-over-Sands: Institute of Terrestrial Ecology.
Bunce, R.G.H., Barr, C.J. and Whittaker, H.A. (1981) An integrated system of land classification. Annual Report of the Institute of Terrestrial Ecology 1980, 28-33.
Bunce, R.G.H., Barr, C.J. and Whittaker, H.A. (1982) Land classes in Great Britain: preliminary description for users of the method of land classification (Merlewood research development paper no.8 6 ). Grange-over-Sands: Institute of Terrestrial Ecology.
Carruthers, P.J. and Choularton, T.W. (1983) A model of the seeder-feeder mechanism orographic rain including stratification and wind drift effects. Quarterly Journal of the Royal Meteorological Society 109, 575-588.
Catt, J.A. (1985) Natural soil acidity. Soil use and management 8-10.
Campbell, G.W., Cocksedge, J.L., Coster, S.M., Dennis, A.L., Pevenish, M., Heyes, C.J., Perrin, P.A., Stone, B.H. and Irwin J.G. (1987) Acid Rain in the United Kingdom: Spatial Distributions and Seasonal Variations in 1986. Report LR691(AP)M. Warren Spring Laboratory, Stevenage.
313
Cape, J.N., Fowler, D., Kinnaird, J.W., Nicholson, I.A. and Paterson, I.S. (1987) Modification of rainfall chemistry by a forest canopy. Pollutant transport and fate in ecosystems. Special Publication No.6 of the British Ecological Society.
Charleson, R.J and Rodhe, H. (1982) Factors controlling the acidity of natural rainwater. Nature 295, 683-5.
Christopherson, N., Seip, H.M. and Wright, R.F. (1982) A model of streamwater chemistry at Birkenes, Norway. Water Resources Research 18, 997-996.
Christopherson, N., Neal, C., Hooper, R.P., Vogt, R.D. and Anderson, S. (1990) Modelling streamwater chemistry as a mixture of soilwater and end-members - a step towards second generation acidification models. Journal of Hydrology 307-320.
CLAG (1994) Critical Loads of Acidity in the United Kingdom. Critical Loads Advisory Group Summary Report prepared at the request of the Department of the Environment, February 1994. ITE, Edinburgh.
CLAG Freshwaters (1995) Critical loads and acid deposition for UK freshwaters. A Report to the Department of the Environment from the Critical Loads Advisory Group, Freshwaters subgroup. Environmental Change Research Centre, London.
Clarke, P.A., Fletcher, I.S., Kallend, A.S., McElroy, W.J., Marsh, A.R.W. and Webb A.M.(1984) Observations of cloud chemistry during long-range transport of power plant plumes. Environmental Pollution 18, 1849-1858.
Cosby, B.J., Hornberger, G.M., Galloway, J.N. and Wright, R.F. (1985a) Time scales of catchment acidification. Environmental Science and Technology 19, 1144-1149.
Cosby, B.J., Wright, R.F., Hornberger, G.M. and Galloway, J.N. (1985b) Modelling the effects of acid deposition: assessment of a lumped parameter model of soil water and stream water chemistry. Water Resources Research 2^, 51-63.
Cosby, B.J., Jenkins, A., Miller, J.D., Ferrier, R.C., and Walker, T.A.B. (1990) Modelling stream acidification in forested catchments: long term reconstruction at two sites in Central Scotland. Journal of Hydrology “[20, 143-162.
Coscio, M.R, Pratt, G.C. and Krupa, S.V. (1982) An automatic, refrigerated, sequential precipitation sampler. Atmospheric Environment 1939-44.
Cottrill, S.M., Heyes, C.J. and Irwin, J.G. (1986) Spatial distribution of acidity and other ions in United Kingdom precipitation-1986 In Proc. EMEP workshop on data analysis and precipitation, Koln. Lillestrom, NILU.
Cowling, E.B. (1982) Acid precipitation in historical perspective. Environmental Science and Technology 16,11 OA-123A.
Craig, G.Y. (Ed.) (1983) Geology of Scotland, Scottish Academic Press, Edinburgh.
Crawshaw, D.H. and Diamond M. (1988) Effects of agricultural liming on surface water. In Barth, H. (Ed.) Effects of Land Use in Catchments on the Acidity and Ecology of Natural Surface Waters. Air Pollution Research Report 13, Proceedings of a Workshop at Cardiff, 11-13 April 1988, pp. 137-151.
Creasey, J., Edwards, A.C., Reid, J.M., MacLeod, D.A. and Cresser, M.S. (1986) The Use ofCatchment Studies for Assessing Chemical Weathering Rates in Two Contrasting Upland Areas in Northeast Scotland. In Colman, S.M. and Dethier, D.P. (Eds.) Rates of Chemical Weathering of Rocks and Minerals Academic Press Ltd, London, pp. 467-502.
314
Cresser, M.S., Edwards, A.C., Ingram, S., Skiba, U. and Pierson-Smith, T. (1986) Soil-acid déposition interactions and their possible effects on geochemical weathering rates in British uplands. Journal of the Geological Society, London 143, 649-658.
Cresser, M.S., Smith, C., and Sanger, L. (1993) Critical loads for peats. In Hornung, M. and Skeffington, R.A. (Eds.) Critical loads concepts and applications. ITE symposium No.28, London: HMSO, pp. 34-39.
Crossley, A., Wilson, D.B. and Milne R. (1992) Pollution in the upland environment Environmental Pollution 75, 81-87.
Cullis, C.F. and Hirschler, M.M. (1980) Atmospheric sulphur : natural and man-made sources. Atmospheric Environment 1263-1278.
Curtis, C.J., Ailott, T.E.H., Battarbee, R.W. and Harriman, R. (1995) Validation of the UK Critical Loads for Freshwaters: Site Selection and Sensitivity. Water, Air and Soil Pollution 85, 2467-2472.
Davis, J.C. (1986) Statistics and data analysis in Geology (2nd edition). John Wiley & Sons, Inc., New York
Davison, B. and Hewitt, C.N. (1992) Natural Sulfur Species from the North Atlantic and their Contribution to the United Kingdom Sulfur Budget. Journal of Geophysical Research - Atmospheres 97, 2475-2488.
Derwent, R.G. (1987) Modelling the long-range transport of ammonia and ammonia compounds. In Asman, W.A.H. and Diederen, H.S.M.A, Ammonia and acidification, Bilthoven, RIVM, pp. 223- 38.
Derwent, R.G and Wilson, R.B. (1992) Acidification research; evaluation and policy applications. In Hornung, M. and Skeffington, R.A (Eds.) Critical loads: concepts and applications. ITE Symposium No. 28, London: HMSO, pp. 5-7.
Derwent, R.G., Dollard, G.J. and Metcalfe, S.E. (1988) On the nitrogen budget for the United Kingdom and north-west Europe. Journal of the Royal Meteorological Society A, 1127-1152.
de Vries, W. and Latour, J.B. (1995) Methods to derive Critical Loads for Nitrogen for Terrestrial Ecosystems. In Hornung, M., Sutton, M. and Wilson, R.B. Mapping and modelling of criticalioads for nitrogen - a workshop report. Report of a workshop held at Grange-over-Sands, Cumbria, UK under the auspices of the UN-ECE Convention on Long Range Transboundary Air Pollution, Working Group for Effects, 24-26 October, 1994. ITE Edinburgh, pp.20-33.
Dollard, G.J., Unsworth, M.H. and Harve, M.J. (1983) Pollutant transfer in upland regions by occult precipitation. Nature 302, 241-243.
Donald, A.P., Stoner, J.H., Reynolds, B., Oldfield, P., Rippey, B. and Natkanski, J. (1990) Atmospheric deposition. In Edwards, R.W., Gee, A.S. and Stoner, J.H. (Eds.) Acid Waters in Wales. Klewer Academic Publishers, The Netherlands, pp. 39-54.
Dore, A.J., Choularton, T.W., Brown, R. and Blackall, R.M. (1992) Orographic rainfall enhancement in the mountains of the Lake District and Snowdonia. Atmospheric Environment 20, 357-371.
Downing, R.J., Hettelingh, J.P. and de Smet, P.A.M. (1993) Calculation and mapping of critical loads In Europe: Status Report 1993. CCE, RIVM Report 259101003. Bilthoven, The Netherlands.
Draper N.R. and Smith, H, (1981) Applied Regression Analysis (2nd edition) John Wiley & Sons, New York.
315
Duarte, C.M. and Kalff, J. (1989) The influence of catchment geology and lake depth on phytoplankton biomass. Archives of Hydrobiology A15, pp. 27-40.
Ebdon, D. (1987) Statistics in Geography. Basil Blackwell Ltd, Oxford.
Edmunds, W.M. and Kinniburgh, D.G. (1986) The susceptibility of UK groundwaters to acidic deposition. Journal of the Geological Society, London. 143, 707-720.
Eriksson, E. (1988) Critical Deposition Limits for Forest Soils. In Nilsson, J. and Grennfelt, P. (Eds) (1988) Critical loads for sulphur and nitrogen. UNECE/Nordic Council workshop report, Nordic Council of Ministers; Copenhagen, pp. 151-184
Eriksson, E., Karltun, E. and Lundmark J-E. (1992) Acidification of forest soils in Sweden. Ambio2^, 150-154.
Erisman, J.W., Potma, C., Draaijers, G., van Leeuwen, E. and van Pul, A. (1995) A generalised description of the deposition of acidifying pollutants on a small scale in Europe. Water, Air and Soil Pollution 85, 2101-2106.
Ferrier, R.C., Miller, J.D., Bruce-Walker, T.A. and Anderson, H.A. (1990) Hydrochemical changes associated with vegetation and soils. In Mason, B.A. (Ed.) The Surface Waters Acidification Programme, Cambridge University Press, Cambridge, 57-68.
Ferrier, P.O., Jenkins, A., Cosby, B.J., Helliwell, P.O., Wright, P.F. and Bulger, A.J. (1995) Effects of Future Deposition Scenarios on the Galloway Region of SW Scotland Using a Coupled Sulphur and Nitrogen Model {MAGIC-WAND). Water, Air and Soil Pollution 85, 707-712.
Forestry Authority (1993) Forests and Water Guidelines (Third Edition). Forestry Commission, London: HMSC.
Forsius, M., Kamari, J. and Posch, M. (1992) Critical loads for Finnish lakes: comparison of three steady-state models. Environmental Pollution 77, 185-193.
Foth, H.D. (1990) Fundamentals of soil science (8th edition). John Wiley & Sons, New York.
Fowler, D. and Unsworth, M.H. (1979) Turbulent transfer of sulphur dioxide to a wheat crop. Quarterly Journal of the Royal Meteorological Society “[OS, 767-783.
Fowler, D. and Cape, J.N. (1985) How is acid-rain formed and where does it fall? In Report of the Acid Rain Inquiry held in Edinburgh on 27th-29th September organised by the Scottish Wildlife Trust pp. 30-45.
Fowler, D., Cape, J.N., Leith, I.D., Patterson J.S., Kinnaird, J.W. and Nicholson I.A. (1982) Rainfall acidity in Northern Britain Nature 279, 383-3866.
Fowler, D., Cape, J.N., Leith, I.D., Choularton, T.W., Gay, M.J. and Jones, A. (1988) The influence of altitude on rainfall composition at Great Dun Fell. Atmospheric Environment22,1355- 1381.
Fowler, D., Cape, J.N., and Unsworth, M.H. (1989) Deposition of atmospheric pollutants on forests. Philosophical Transactions of the Royal Society. B324: 1355-1362.
Fowler, D., Leith, I.D., Smith, I.D., Choularton, T.W., Inglis, D. and Campbell, G. (1994) Atmospheric Inputs of Acidity, Sulphur and Nitrogen in the UK. In Battarbee, R.W. (Ed.) Acid Rain and its Impact: the Critical Loads Debate. Proceedings of a conference held at the Environmental Change Research Centre, University College London. ENSIS Publishing, London, pp. 17-26.
Fowler, D., Leith, I.D., Binnie, J., Crossley, A., Inglis, D.W.F., Choularton, T.W., Gay, M.,
316
Longhurst, J.W.S. and Conland, D.E. (1995) Orographic enhancement of wet deposition in the United Kingdom: Continuous monitoring. \Na\er, Air and Soil Pollution S5, 2107-2112.
Fritz, S.C., Kreiser, A.M., Appleby, P.G. and Battarbee, R.W. (1990) Recent acidification of upland lakes in North Wales: Palaeolimnological evidence. In Edwards, R.W., Gee, A.S. and Stoner, J.H. (Eds.) Acid Waters in Wales. Klewer Academic Publishers, The Netherlands, pp. 27- 38.
Fuller, R.D., Driscoll, C.T., Lawrence, G.B. and Nodvin, S.C. (1987) Processes Regulating Sulphate Flux After Whole Tree Harvesting. Nature 325, 707-710
Fuller, R.M. and Groom, G.B. (1993a) The Land Cover Map of Great Britain. GIS Europe 2, 25- 28.
Fuller, R.M. and Groom, G.B. (1993b) The Land Cover Map of Great Britain. Mapping Awareness 7, 18-20.
Fuller, R.M., Groom, G.B. and Jones, A.R. (1994) The Land Cover Map of Great Britain: An Automated Classification of Landsat Thematic Mapper Data. Photogrammetric Engineering and Remote Sensing 60, 553-562.
Galloway, J.N. (1995) Acid deposition: perspectives in time and space. Water, Air and Soil Pollution 85, 15-24.
Galloway, J.M. and Likens, G.E. (1981) Acid precipitation: The importance of nitric acid. Atmospheric Environment 15, 1081-1085.
Galloway, J.M., Likens, G.E., Keene, W.C. and Miller, J.M. (1982) The composition of precipitation in remote areas of the world. Journal of Geophysical Research 87, 8771-8786.
Garland, J.A. (1978) Dry and wet removal from sulphur in the atmosphere. Atmospheric Environment “\2, 349-362.
Gasser, J.K.R. (1985) Processes causing loss of calcium ions from agricultural soils. Soil Use and Management 14-17.
Gjessing, E.T., Henriksen, A., Johannesen, M. and Wright, R.F. (1976) Effects of acid precipitation on freshwater chemistry. In Braake, F. (Ed). Impact of acid precipitation on forestand freshwater ecosystems in Norway. SNSF FR 6/76, Oslo, Norway, pp. 64-85.
Golterman, H.L., Clymo, R.S. and Ohnstad, M.A.M. (1978) IBP Handbook No. 8. Methods for Physical and Chemical Analysis of Freshwater (2nd Edition). Blackwell Scientific Publications, Oxford
Gorham E. (1957) The influence and importance of daily weather conditions in the supply of chloride, sulphate and other ions to freshwater from atmospheric precipitation. Philosophical Transactions of the Royal Society, London 6 241, 147-178 .
Gorham, E. (1992) Atmospheric deposition to lakes and its ecological effects: A retrospective and prospective view of research. Japanese Journal of Limnology 53, 231-248.
Gunderson, P. and Bashkin, V.N. Nitrogen Cycling. In Moldan, B. and Cerny, J. Biogeochemistry of Small Catchments: A Tool for Environmental Research. John Wiley & Sons Ltd, Chichester, England.
Hald, A. (1952) Statistical Theory with Engineering Applications. Wiley, New York.
Hall, D. and Titterlngton, P.M. (1989) The effect of simulation on level accuracy and power of
317
Monte Carlo tests. Journal of the Royal Statistical Society B. 51 459-467
Hall, J.R., Wright, S.M., Sparks, T.H., Ullyet, J., Allott, T.E.H. and Hornung, M. (1995a) Predicting Freshwater Critical Loads from National Data on Geology, Soils and Land Use. Water, Air and Soil Pollution 85, 2443-2448.
Hall, J.R., Bull, K., Brown, M., Dyke, H., Ullyet, J. and Hornung, M. (1995b) The effects of scale and resolution in developing percentile maps of critical loads for the UK. Water, Air and Soil Pollution 85, 2521-2526.
Hallberg, R.O., BÂgander, L.E. and Engvall, A.-G. (1976) Dynamics of phosphorus, sulphur and nitrogen at the sediment-water interface. In Nriagu, J.O. (Ed.) Environmental Biogeochemistry Ann Arbor Science, Ann Arbor, ML, pp. 295-308
Hamaker, H.C. (1962) On multiple regression analysis. Statistica Neeriandica 16, 31-56.
Hammer, C.U., Clausen, H.B. and Dansgaard, W. (1980) Greenland ice sheet evidence of postglacial volcanism and its climatic effect. Nature 288, 230-235.
Harriman, R. and Welis, D.E. (1985) Causes and Effects of Surface Water Acidification in Scotland. Journal of Water Pollution Control 84, 215-224.
Harriman, R. and Morrison, B.R.S. (1982) Ecology of Streams Draining Forested and Nonforested Catchments in an Area of Central Scotland, UK Subject to Acid Precipitation. Hydrobiologla 88, 251-264.
Harriman, R. and Christie, A.E.G. (1992) Evaluation of the steady-state water chemistry method for surface waters. In Hornung, M. and Skeffington, R.A (Eds.) Critical loads: concepts and applications. ITE Symposium No. 28, London: HMSO, pp. 103-106.
Harriman, R. and Pugh, K.B. (1994) Water Chemistry. In Maitland, P.S., Boon, P.J. and McLusky, D.S. (Eds.) The Fresh Waters of Scotland: A National Resource of International Significance. John Wiley & Sons, Chichester, pp. 89-112.
Harriman, R., Gillespie, E., King, D., Watt A.W., Christie, A.E.G., Cowan, A.A. and Edwards,T. (1990) Short-term ionic responses as indictors of hydrochemical processes in the AI It A’Mharcaidh catchment. Western Cairngorms, Scotland. Journal of Hydrology "1 6, 267-285.
Harriman, R., Anderson, H. and Miller, J.D. (1995a) The Role of Sea-salts in Enhancing and Mitigating Surface Water Acidity. Water, Air and Soil Pollution 85, 553-558.
Harriman, R., Aliott, T.E.H., Battarbee, R.W., Curtis, C., Hall, J. and Bull, K. (1995b) Critical Load Maps for UK Freshwaters. In CLAG Freshwaters, Critical Loads of Acid Deposition for United Kingdom Freshwaters. Critical Loads Advisory Group, Sub-group report on Freshwaters, ITE, Edinburgh, pp. 19-24.
Harriman, R., Bridcut, E.E. and Anderson, H. (1995c) The Relationship Between Salmonid Fish Densities and Critical ANC at Exceeded and Non-Exceeded Stream Sites in Scotland. Water, Air and Soil Pollution 85, 2455-2460.
Harriman, R., Ailott, T.E.H., Battarbee, R.W., Curtis, C., Jenkins, A. and Hali, J.R. (1995d) Critical Loads of Nitrogen and their Exceedance in UK Freshwaters. In Hornung, M., Sutton, M. and Wilson, R.B. Mapping and modelling of critical loads for nitrogen - a workshop report. Report of a workshop held at Grange-over-Sands, Cumbria, UK under the auspices of the UN-ECE Convention on Long Range Transboundary Air Pollution, Working Group for Effects, 24-26 October, 1994. ITE Edinburgh, pp.39-49.
Havas, M. and Rosseiand, B.C. (1995) Response of zooplankton, benthos, and fish to
318
acidification: An overview. Water, Air and Soil Pollution 85, 51-62.
Henriksen, A. (1979) A simple approach for identifying and measuring acidification of freshwater. Nature 278, 542-545.
Henriksen, A. (1982) Changes in base cation concentrations due to freshwater acidification. Verheissungen der internationalen Vereinigung fuer gasammten Limnologie 22, 692-698.
Henriksen, A. (1995) Critical Loads of Acidity to Surface Waters - How Important is the F-Factor in the SSWC-Model? Water, Air and Soil Pollution 85, 2437-2441.
Henriksen, A. and Posch, M. (1995) Critical Loads for Nitrogen: Surface Waters. In Hornung, M., Sutton, M. and Wilson, R.B. (Eds.) Mapping and modelling of critical loads - a workshop report. Report of a workshop held at Grange-over-Sands, Cumbria. UN-ECE Convention on Long Range Transboundary Air Pollution, Working Group for Effects, 24-26 October, 1994. ITE, Edinburgh, pp. 55-62.
Henriksen, A., Dickson, W. and Brakke, D.F. (1986) Estimates of critical loads to surface waters. In Nilsson, J. (Ed.) Critical loads for sulphur and nitrogen. Nordic Council of Ministers, Copenhagen, pp. 87-120.
Henriksen, A., Lien, L., Traaen, T.S., Rosseiand, R.G. and Sevairud, I.S. (1988) The 1000-lake survey in Norway in 1986. In Mason, B.J. (Ed.) The Surface Waters Acidification Programme. Cambridge University Press, Cambridge, pp. 199-212.
Henriksen, A., Kamari, J. Posch, M., Lovbiad, G., Forsius, M. and Wiiander, A. (1990) Critical Loads to Surface Waters in Fennoscandia : Intra- and Inter- Grid Variability of Critical Loads and their Exceedance. Miljorapport 7990:77 Nordic council of Ministers, Copenhagen.
Henriksen, A., Kamari, J. Posch, M. and Wiiander, A. (1992) Critical loads of acidity: Nordic surface waters. Ambio 21, 356-363.
Henriksen, A., Forsius, M., Kamari, J. Posch, M. and Wiiander, A. (1993) Exceedance of Critical Loads for Lakes in Finland, Norway and Sweden: Reduction Requirements for Nitrogen and Sulfur Deposition. Report 32/1993, Norwegian Institute for Water Research.
Henriksen, L. and Brodin, Y.-W. (Eds.) (1995) Liming of acidified surface waters: A Swedish synthesis. Springer Verlag, Berlin.
Henriksen, L., Hindar, A. and Thornelof, E. (1995) Freshwater Liming. Water, Air and Soil Pollution 85, 131-142.
Hettelingh, J-P., Downing, R.J. and de Smet, P.A.M. (Eds) (1991) Mapping critical loads for Europe. CCE Technical Report No. 1. RIVM report 259101001. RIVM: Bilthoven, Netherlands
Hettelingh, J.P., Posch, M., de Vries, W., Bull, K. and Sverdrup, H.U. (1992a) Guidelines for the Computation and Mapping of Nitrogen Critical Loads and Exceedances in Europe. Workshop on Critical Loads for Nitrogen, Lokeberg, Sweden, April 1992.
Hettelingh, J.P., Gardner, R.H. and Hordijk, L. (1992b) A statistical approach to the regional use of critical loads. Environmental Pollution 77, 177-183.
Hill, M.O. (1979) TWINSPAN - A FORTRAN program for arranging multivariate data in an ordered two-way table by classification of the individuals and attributes. Cornell University, Ithica, New York.
Hill, M.O., Bunce, R.G.H. and Shaw M.W. (1975) Indicator species analysis, a divisive polythetic method of classification, and its application to a survey of native pinewoods in Scotland. Journal
319
of Ecology. 63, 597-613.
HMSO (1990) This Common Inheritance. London: HMSO.
Hooper, R.P., Christopherson, N. and Peters, N.E. (1990) Modelling streamwater chemistry as a mixture of soilwater end-members - an application to the Panola Mountain catchment, Georgia, U.S.A. Joumal of Hydrology 321-343.
Hornbeck, J.W. (1992) Comparative impacts of forest harvest and acid precipitation on soil and streamwater acidity. Environmental Pollution 77, 151-155.
Hornung, M. (1985) Acidification of soils by trees and forests. Soil Use and Management 1, 24- 28.
Hornung, M. (1993) The provisional map of critical loads for acidity for soils of Great Britain. In Hornung, M. and Skeffington, R.A. (Eds.) Critical loads concepts and applications. ITE symposium No.28, London: HMSO, pp. 31-33.
Hornung, M., Stevens, P.A. and Reynolds, B. (1986) The impact of pasture improvement on the soil solution chemistry of some stagnopdzols in mid-Wales. Soil Use and Management 2 ,18- 26
Hornung, M., Roda, P., and Langan, S. (1990a) A Review of Small Catchment Studies in Western Europe Producing Hydrochemical Budgets. Commission of the European Communities Directorate-General for Science, Research and Development Environment Research Programme. Air Pollution Research Report 28. E.Guyot SA, Brussels, Belgium.
Hornung, M., Le-Grlce, S., Brown, N. and Norris, D. (1990b) The role of geology an soils in controlling surface water acidity in Wales. In Edwards, R.W., Gee, A.S. and Stoner, J.H. (Eds.) Acid Waters in Wales. Klewer Academic Publishers, The Netherlands, pp. 55-66.
Hornung, M., Brown, S.J. and Ranson, A. (1990c) Amelioration of surface water acidity by catchment management. In Edwards, R.W., Gee, A.S. and Stoner, J.H. (Eds.) Acid Waters in Wales. Klewer Academic Publishers, The Netherlands, pp. 311-328.
Hornung, M., Bull, K.R., Cresser., Hall, J., Loveland, P., Langan, S.J., Reynolds, B. and Robertson W.H. (1994) Mapping Critical Loads for the Soils of Great Britain, in Battarbee, R.W. (Ed.) Acid Rain and its Impact: The Critical Loads Debate, Proceedings of a conference held at the Environmental Change Research Centre, UCL, ENSIS Ltd, London, pp. 43-52.
Hornung, M., Bull, K., Cresser, M., Ullyett. J., Hall. J.R., Langan, S., Loveland, P.J. and Wilson, M.J. (1995a) The sensitivity of surface waters of Great Britain to acidification. Environmental pollution 87, 204-217.
Hornung, M., Sutton, M. and Wilson, R.B. (Eds.) (1995b) Mapping and modelling of critical loads - a workshop report. Report of a workshop held at Grange-over-Sands, Cumbria. UN-ECE Convention on Long Range Transboundary Air Pollution, Working Group for Effects, 24-26 October, 1994. ITE, Edinburgh.
Hultberg, H., ApSimon, H., Church, R.M., Grennfelt, P., Mitchell, M.J., Moldan, F. and Ross, H.B. (1994) Sulphur. In Moldan, B. and Cerny, J. Biogeochemistry of Small Catchments: A Tool for Environmental Research. John Wiley & Sons Ltd, Chichester, England, pp. 229-254.
Hunsaker, C.T., Levine, D.A., Timmins, S.P., Jackson, B.L. and O’Neill, R.V (1991) Landscape characterisation for assessing regional water quality. Report to the U.S Government. Contract No. DE-ACOS-840R21400 pp. 997-1006.
Ikuta, K. and Kitamura, S. (1985) Effects of Low pH Exposure of Adult Salmonids on
320
Gametogenesis and Embryo Development. Water, Air and Soil Pollution 85, 327-332.
Inglis, D.W.F., Choularton, T.W., Wicks, A.J., Fowler, P., Leith, I.D., Werkman, B. and Binnnie, J. (1995) Orographic Enhancement of Wet Deposition in the United Kingdom: Case Studies and Modelling. Water, Air and Soil Pollution 85, 2119-2124.
Irwin, J. (1989) Acid Rain: Emissions and Deposition. Archives of Environmental Contamination and Toxicology 95-107.
Irwin, J.G. and Williams, M.L. (1988) Acid Rain: Chemistry and Transport. Environmental Pollution 50, 29-59.
Israëls, A. (1992) Redundancy Analysis for Various Types of Variables. Statistica Applicata 4, 531-542.
Iverson, T., Halvorsen, N., Mylona, S. and Sandnes, H. (1991) Calculated Budgets for Airborne Acidifying Components in Europe 1985, 1987, 1988 and 1990. Meteorological Synthesizing Centre -West, the Norwegian Meteorological Institute, Oslo, Norway.
Jager, J.C. and Looman, C.W.M. (1987) Data collection. In Jongman, R.H.G., ter Braak, C.J.F. and van Tongeren, O.F.R. (Eds.) Data analysis in community and landscape ecology. Pudoc, Wageninge, pp. 10-28.
Jakeman, A.J., Whitehead, P.O., Robson. A., Jenkins, A. and Taylor, A.J. (1990) A method for predicting the extremes of stream acidity and other water quality variables. Journal of Hydrology '\^6, 375-390.
Jenkins, A., Ferrier, B.C., Walker, T.A.B. and Whitehead, P.G. (1988) A modelling study of long-term acidification in an upland Scottish catchment. Water, Air and Soil Pollution AO, 275-291.
Jenkins, A., Cosby, B.J., Ferrier, B.C., Walker, T.A.B. and Miller, J.D. (1990a) Modelling stream acidification in forested catchments an assessment of the relative effects of acid deposition and afforestation. Journal of Hydrology A20, 163-181.
Jenkins, A., Harriman, B. and Tuck, S.J. (1990b) Integrated hydrochemical responses on the catchment scale. In Mason, B.A. (Ed.) The Surface Waters Acidification Programme, Cambridge University Press, Cambridge, pp. 47-56.
Jenkins, A., Wright, B.F. and Cosby, B.J. (1994) Modelling long-term hydrochemical response at ENCORE catchments In the UK and Norway. Acid Rain Research Report 35/1994, Norwegian Institute for Water Research, Osio, Norway.
Johnson, P.L. and Swank, W.T. (1973) Studies of cation budgets in the southern Appalachians on four experimental watersheds with contrasting vegetation. Ecology 56, 70-80.
Johnston, B.J. (1991) Multivariate statistical analysis in geography. Longman Scientific and Technical, John Wiley & Sons, Inc., New York.
Juggins, S. and ter Braak, C.J.F. (1993) CALIBRATE v 0.52 (Beta Test) - a computer program for species-environment calibration by [weighted averaging] partial least squares regression ECRC, University College London.
Juggins, S., Ormerod, S.J. and Harriman, B. (1995) Relating Critical Loads to Aquatic Biota. In CLAG Freshwaters: Critical Loads of Acid Deposition for UK Freshwaters. A Report to the Department of the Environment from the Critical Loads Advisory Group, Freshwaters sub-group. Environmental Change Research Centre, London, pp. 9-14.
Kalkhoff, S.J. (1993) Using a Geographical Information System to Determine the Relation
321
Between Stream Quality and Geology in the Roberts Creek Watershed, Clayton County, Iowa. Water Resources Bulletin 29, 989-997.
Kamari, J., Amman, M., Brodin. Y-W., Chadwick, M.J., Henriksen A., Hettelingh, J-P., Kuylenstierna, J., Posch, M. and Sverdrup, H. (1992a) The Use of Critical Loads for the Assessment of Future Alternatives for Acidification. Ambio 21, 377-386.
Kamari, J., Jeffries, J., Hessen, P.O., Henriksen A., Posch, M., and Forsius, M. (1992b) Nitrogen critical loads and their exceedance for surface waters. In Grennfelt, P. and Thornelof, E. (Eds.) Critical Loads for Nitrogen - a Workshop Report. Nord 1992:41, pp. 161-200.
Kellog, W.W., Cadle, R.D., Allen, E.R., Lazrus, A.L. and Martell, E.A. (1972) The sulphur cycle. Science 175, 587-596.
Kennedy, I.R. (1992) Acid Soil and Acid Rain. Research Studies Press Ltd, Taunton.
Kent, M. and Coker, P. (1992) Vegetation Description and Analysis : A Practical Approach. Bellhaven Press, London.
Kernan, M. (1995) The use of catchment attributes to predict surface water critical loads: a preliminary analysis. Water, Air and Soil Pollution 85, 2479-2484.
Kinniburgh, D.G. and Edmunds, W.M. (1984) The susceptibility of UK groundwaters to acid deposition. Hydrogeological Report, British Geological Survey. No. 86/3.
Kirchner, J.W., Dillon, P.J. and LaZerte, B.D. (1993) Predictability of Geochemical Buffering and Runoff Acidification from Spatially Heterogeneous Catchments. Water Resources Research 29, 3891-3901.
Kramer, J. (1976) Geochemical and Lithological Factors in Acid Precipitation. Proceedings of the 1st International Symposium on Acid Precipitation and the Forest Ecosystem, US Department of Agriculture Forest Service.
Kreiser, A.M., Appleby, P.O., Natkanski, J., Rippey, B. and Battarbee R.W. (1990) Afforestation and acidification: a comparison of four sites in Scotland. Philosophical Transactions of the Royal Society of London B 327, 377-383.
Kreiser, A.M., Patrick, S.T. and Battarbee, R.W. (1993) Critical loads for UK freshwaters - introduction, sampling strategy and use of maps. In Hornung, M. and Skeffington, R.A. (Eds.) Critical loads: concepts and applications. ITE Symposium No. 28, London: HMSO, pp. 94-98.
Kreiser, A.M., Patrick, S.T. and Battarbee, R.W., Hall, J. and Harriman R. (1995) Mapping water chemistry. In CLAG Freshwaters: Critical Loads of Acid Deposition for UK Freshwaters. A Report to the Department of the Environment from the Critical Loads Advisory Group, Freshwaters sub-group. Environmental Change Research Centre, London, pp. 15-18.
Kreutzer, K. (1988) The impact of forest management practices on the soil acidification in established forests. In Barth, H. (Ed.) Effects of Land Use in Catchments on the Acidity and Ecology of Natural Surface Waters. Air Pollution Research Report 13, Proceedings of a Workshop at Cardiff, 11-13 April 1988, pp. 75-90.
Krug, E.G. and Frink, C.R. (1983) Acid Rain on Acid Soil: A new perspective. Science 22^, 520- 525.
Langan, S.J. (1989) Sea-salt Induced Streamwater Acidification. Hydrological Processes 3, 25- 41.
Langan, S.J. and Wilson, M. (1991) Predicting the occurrence of surface water acidification in
322
Scotland using an approach based on geology, soils and land use. Final Report (Part 3) for the Scottish Development Department Critical Load Contract, MLURI, Aberdeen.
Langan, S.J. and Wilson, M.J. (1992) Predicting the regional occurrence of acid surface waters in Scotland using an approach based on geology, soils and land use. Journal of Hydrology ^3S, 515-528.
Langan, S.J. and Harriman, R. (1993) Critical loads as a tool for examining the impact of acid deposition at Loch Dee, S.W. Scotland. In Acidification, forestry and fisheries management In upland Galloway - Proceedings of the Loch Dee Symposium. Gatehouse of Fleet, December 1992 Foundation for Water Research, pp. 137-146.
Langan, S.J., Bain, D. and Wilson, J. (1994) The calculation of weathering rates and their role in the determination of critical loads. In Battarbee, R.W. (Ed.) Acid rain and its impact: the critical loads debate. Proceedings of a conference held at the Environmental Change Research Centre, University College London. ENSIS Publishing, London, pp. 124-127.
Langan, S.J., Sverdrup, H.U. and Coull, M. (1995) The calculation of base cation release from the chemical weathering of Scottish soils using the PROFILE model. Water, Air and Soil Pollution 85, 2497-2502.
Lapointe, F-J. and Legendre, P. (1994) A Classification of Pure Malt Scotch Whiskies. Applied Statistics 43, 237-257.
Larsson, S. (1985) The effects of liming on aquatic flora. In Henriksen, L. and Brodin, Y.-W. (Eds.) Liming of acidified surface waters: A Swedish synthesis. Springer Verlag, Berlin, pp. 193- 220.
Lau, W.M and Mainwairing, S.J. (1985) The determination of soil sensitivity to acid deposition. Water, Air and Soil Pollution 25, 451 -464.
Lawrence, G.B. (1988) Whole tree harvesting of an acid sensitive watershed. In Barth, H. (Ed.) Effects of Land Use in Catchments on the Acidity and Ecology of Natural Surface Waters. Air Pollution Research Report 13, Proceedings of a Workshop at Cardiff, 11-13 April 1988, pp. 26-36.
Lawrence, G.B., Driscoll, C.T. and Fuller, R.D. (1988) Hydrologie control of aluminium chemistry in an acidic headwater stream. Water Resources Research 24, 650-669.
Lein, L., Raddum, G.G. and Fjellheim, A. (1992) Critical Loads of Acidity to Freshwater -Fish and Invertebrates. Norwegian Institute for Water Research Report 0-89185.
Lelong, F., Dupraz, C., Durand, P. and Didon-Lescot, J.F. (1990) Effects of vegetation type on the biogeochemistry of small catchments (Mont Lozere, France). Journal of Hydrology 125-145.
Le Maitre, R.W. (1968) Chemical variation within and between volcanic rock series - a statistical approach. Journal of Petrology 9, 220-252.
Le Maitre, R.W. (1982) Numerical Petrology. Elsevier Publishing Company, Amsterdam.
Likens, G.E., Wright, R.F., Galloway, J.N. and Butler, T.J. (1979) Acid Rain. Scientific American 241, 42-51.
Liss, P.S., Malin, G., Turner, S.M. and Holligan, P.M. (1994) Dimethyl Sulphide and Phaeocystis - A Review. Journal of Marine Systems 5, 41-53.
Lotter, A.F. and Birks, H.J.B. (1993) The Impact of the Laacher See tephra on Terrestrial and Aquatic Ecosystems in the Black Forest, Southern Germany. Journal of Quaternary Science 8 ,
323
263-276.
Lôvblad, G., Amman. M., Anderson, B., Hovmand, M., Joffre, S. and Pederson, U. (1992) Deposition of Sulphur and Nitrogen in Nordic Countries: Present and Future. Ambio 2^, 339-347.
Lynch, D.D. and Dise, N.B. (1985) Sensitivity of stream basins in Shenandoah National Park to Acid deposition. U.S. Geological Survey Water Resources Investigations Report 85-4115
Macauley Institute for Soil Research (1981) Soil survey maps of Scotland 1:250 000 scale. Aberdeen: Macauley Institute for Soil Research
Macauley Institute for Soil Research (1984) Organization and methods of the 1:250,000 Soil Survey of Scotland. Aberdeen: Macauley Institute for Soil Research.
Mackenzie, S. (1989) Catchment liming: effects on plants and invertebrates in a Welsh blanket bog. In Ormerod, S.J. (Ed.) Ecology Sub-group Report of the Llyn Brianne Acid Waters Project November, 1989, pp. 38-40.
Mackenzie, S. and Shore, R.J. (1989) Possible effects on small mammals of liming upland areas. In Ormerod, S.J. (Ed.) Ecology Sub-group Report of the Llyn Brianne Acid Waters Project November, 1989, pp. 41-49.
MacPhee, K.P., Langan, S.J. and Billett, M.F. (1995) Critical Loads for Soils and Waters in a Selected Scottish Catchment. Water, Air and Soil Pollution 85, 2473-2478.
Madera, V., Allen, H.E. and Minear, R.A. (1982) Non-metallic Constituents. In Suess, M.J. (Eds.), Examination of Water for Pollution Control. A Reference Handbook. Volume 2, Physical, chemical and radiological examination. Pergammon Press Ltd, Oxford.
Manly, B.J.F. (1992) The design and analysis of research studies. Cambridge University Press, Cambridge.
Marchetto, A., Mosello, R., Psenner, R., Bendetta, G., Boggero, A., Tait, D. and Tartai, G.A.(1995) Factors affecting water chemistry of alpine lakes. Aquatic Sciences 57, 81-89.
Matschullat, J., Andraea, H., Lessmann, D., Malessa V. and Siewers, U. (1992) Catchment acidification - from the top down. Environmental Pollution 77, 143-150.
Meisch, A T. (1980) Scaling Variables and Interpretation of Eigenvalues in Principal Components Analysis of Geologic Data. Mathematical Geology ^2, 523-528.
Meiwes, K.J., (1995) Application of lime and wood ash to decrease acidification of forest soils. Water, Air and Soil Pollution 85, 143-152.
Metcalfe, S.E. and Whyatt, J.D. (1994) Modelling future acid deposition with H.A.R.M. In Battarbee, R.W. (Ed.) Acid rain and its impact: the critical loads debate. Proceedings of a conference held at the Environmental Change Research Centre, University College London. ENSIS Publishing, London, pp. 27-36.
Metcalfe, S.E., Atkins, D.H. and Derwent, R.G. (1989) Acid deposition monitoring and the interpretation of the United Kingdom secondary precipitation network data. Atmospheric Environment 23, 2033-2052.
Miller, H.G. (1985) The possible role of forests in streamwater acidification. Soil Use and Management 1, 28-29.
Miller, H.G., Miller, J.D. and Cooper, J.M. (1987) Transformations in rainwater chemistry on passing through forested ecosystems. Pollutant transport and fate in ecosystems. Special
324
Publication No.6 of the British Ecological Society.
Miller, R. and Drever, J.l. (1977) Chemical weathering and related controls on surface water chemistry in the Absaroka Mountains, Wyoming. Geochimica et Cosmochimica Acta 41, 1693- 1702.
Milner, N.J. and Varallo, P.V. (1990) Effects of acidification on fish and fisheries in Wales. In Edwards, R.W., Gee, A.S. and Stoner, J.H. (Eds.) Acid Waters in Wales. Klewer Academic Publishers, The Netherlands, pp. 311-329.
Montgomery, D.C. and Peck, E.A. (1982) Introduction to linear regression analysis. Wiley, New York.
Mulder, J and Cresser, M.S. (1994) Soil and Soil Solution Chemistry. In Moldan, B. and Cerny, J. Biogeochemistry of Small Catchments: A Tool for Environmental Research. John Wiley & Sons Ltd, Chichester, England, pp. 107-132.
Mulder, J., Christopherson, N., Haughs, M., Vogt, R.D., Anderson, S. and Anderson, D.O.(1990) Water flow paths and hydrochemical controls in the Birkenes catchment as inferred from a rainstorm high in sea-salts. Water Resources Research 26, 611-622.
Muniz, I.P. (1991) Freshwater acidification: its effects on species and communities of freshwater microbes, plants and animals. Proceedings of the Royal Society of Edinburgh, 978, 227-254.
Murlis, J. (1994) Critical Loads: the Policy Background. In Battarbee, R.W. (Ed.) Acid rain and its impact: the critical loads debate. Proceedings of a conference held at the Environmental Change Research Centre, University College London. ENSIS Publishing, London, pp. 7-16.
Muscatt, A.D., Wheater, H.S. and Reynolds, B. (1990) Stormflow Hydrochemistry of a Small Welsh Upland Catchment. Journal of Hydrology 239-249.
Myers, R.H. (1986) Classical and Modern Regression with Applications. Duxbury Press, Boston, MA.
Neal, 0., Smith, C.J., Walls, J. and Dunn, 0.8. (1986) Major, minor and trace element mobility in the acidic upland forested catchment of the upper River Severn, Mid Wales. Journal of the Geological Society, London 143, 635-648.
Neal, 0., Fisher, R., Smith, C.J., Hiil, S., Neal, M., Conway, T., Ryland, G.P. and Jeffrey, H.A.(1992) The Effects of Tree Harvesting on Stream-Water Quality at an Acidic and Acid-Sensitive Spruce Forest. Journal of Hydrology ^35, 305-319.
Newbold, P. (1985) Improvement of native grassland in the uplands. Soil Use and Management 1, 43-49.
Newson, M.D. (1991) Catchment control and planning. Land use policy 8, 9-15.
Nilsson, S.l. (1993) Acidification of Swedish oligotrophic lakes - interaction between deposition, forest growth, and effects on lake-water quality. Ambio 22, 272-276.
Nilsson, J. and Grennfelt, P. (Eds.) (1988) Critical loads for sulphur and nitrogen. UNECE/Nordic Council workshop report, Skokloster, Sweden. March 1988. Nordic Council of Ministers, Copenhagen.
Norcliffe, G.B. (1982) Inferential Statistics for Geographers (2nd Edition). Century Hutchinson Ltd., London.
Norton, S.A. (1980) Geologic factors controlling the sensitivity of aquatic ecosystems to acidic
325
precipitation. In Shriner, D.S. Atmospheric Sulfur Deposition. Ann Arbor Science, Ann Arbor, Ml., pp. 521-531.
Norton, S.A., Kahl, J.S., Henriksen, A. and Wright, R.F. (1990) Buffering of pH depressions by sediments in streams and lakes. In Norton, S.A., Lindberg, S.E. and Page, A.L. (Eds.) Acidic Precipitation Volume 4 : Soils, Aquatic Processes and Lake Acidification. Springer-Verlag, New York, pp. 133-157.
Noy-Meir, I., Walker, D. and Williams, W.T. (1975) Data transformation in ecological ordination II. On the meaning of data standardization. Journal of Ecology 63, 779-800.
Odeh, I.O.A., Chittleborough, D.J. and McBratney, A.B. (1991) Elucidation of soil-landform interrelationships by canonical ordination analysis. Geoderma 49, 1-32.
Oden, S. (1968) The acidification of air precipitation and its consequences in the natural environment. Energy Committee Bulletin 1 . Swedish Natural Sciences Research Council, Stockholm.
Okland, R.H. and Eilertsen, O. (1994) Canonical correspondence analysis with variation partitioning; some comments and an application. Journal of Vegetation Science 5, 117-126.
Olea, R.A. (1975) Optimum mapping techniques using regionalized variable theory. Series on Spatial Analysis, No. 2. Kansas Geological Survey, Lawrence.
Ormerod, S.J. and Wade, K.R. (1990) The role of acidity in the ecology of Welsh lakes and streams. In Edwards, R.W., Gee, A.S. and Stoner, J.H. (Eds.) Acid Waters in Wales. Klewer Academic Publishers, The Netherlands.
Osborne, L.L. and Wiley, M.J. (1988) Empirical relationships between land use/cover and stream water quality in an agricultural watershed. Journal of Environmental Management 26,9-27.
Ott, W.R. (1990) Physical Explanation of the Lognormality of Pollutant Concentrations. Journal of the Air and Waste Management Association 40, 1378-1383.
Paces, T. (1986) Weathering rate of gneiss and depletion of exchangeable cations in soils under environmental acidification. Journal of the Geological Society, London 143, 673-677.
Patrick, S.T. and Stevenson, A.C. (1990) Acidified Welsh lakes: the significance of land use and management. In Edwards, R.W., Gee, A.S. and Stoner, J.H. (Eds.) Acid Waters in Wales. Klewer Academic Publishers, The Netherlands, pp. 81-92.
Patrick, S., Juggins, S., Waters, D. and Jenkins, A. (1991) The United Kingdom Acid Waters Monitoring Network. Site Description and Methodoiogy Report. Report to the DoE and DoE (Nl), ENSIS Ltd.
Penkett, S.A., Jones, B.M.R., Brice, K.A. and Eggleton, A.E.J. (1979) The importance of atmospheric ozone and hydrogen peroxide in oxidising sulphur dioxide in cloud and rainwater. Atmospheric Environment ^3, 123-37.
Porcelia, D.B., Schofield, C.L., Pinto, J.V., Driscoll, C.T., Bukaveckas, P.A., Gloss, S.P. and Young, T.C. (1989) Mitigation of Acidic Conditions in Lakes and Streams. In Norton, S.A., Lindberg, S.E. and Page, A.L. (Eds.) Acidic Precipitation Volume 4 : Soils, Aquatic Processes and Lake Acidification. Springer-Verlag, New York, pp. 155-186.
Posch, M. (1995) Critical Loads for Aquatic Ecosystems. In Hornung, M., Sutton, M. and Wilson, R.B. (Eds.) Mapping and modelling of critical loads - a workshop report. Report of a workshop held at Grange-over-Sands, Cumbria. UN-ECE Convention on Long Range Transboundary Air Pollution, Working Group for Effects, 24-26 October, 1994. ITE, Edinburgh, pp. 63-68.
326
Posch, M., Hettelingh, J.-P., Sverdrup, H., Bull, K.R. and de Vries, W. (1993) Guidelines for the computation and mapping of critical loads and exceedances of S and N in Europe. In Downing, R.J., Hettelingh, J.-P. and de Smet, P.A.M. Calculation and mapping of critical loads in Europe. RIVM Report 259101003. RIVM, Bilthoven.
Posch, M., de Smet, P.A.M., Hettelingh, J.-P. and Downing, R.J. (1995) Calculation and mapping of critical thresholds in Europe: Status Report 1995. CCE , RIVM Report 259101004, RIVM, Bilthoven, The Netherlands.
Rampino, M.R. and Self, S. (1982) Historic eruptions of Tambora (1815), Krakatau (1883), and Agung (1963), their stratospheric aerosols, and climatic impact. Quaternary Research 18, 127- 143.
Rassmussen, R.A. (1974) Emission of biogenic hydrogen sulphide. Tellus 28, 355-372.
Rawlings, R.O. (1988) Applied Regression Analysis: A Research Tool. Wadsworth and Brooks/Cole Advanced Books and Software, Pacific Grove, California.
Reckhow, K.H., Black, R.W., Stockton, T.B. Jr., Vogt, J.D. and Wood, J.G. (1987) Empirical models of fish response to lake acidification. Canadian Journal of Fish and Aquatic Science 44, 1432-1442.
Rees, R.M., Parker-Jervis, F. and Cresser, M. (1989) Soil effects on water chemistry in three adjacent upland streams at Glendye in Northeast Scotland. Water Resources 23, 511-517.
Reid, J.M., MacLeod, D.A. and Cresser, M.S. (1981) Factors affecting the chemistry of precipitation and river water in an upland catchment. Journal of Hydrology 50, 129-145.
Reuss, J. and Christopherson, N. (1986) A critique of models for freshwater and soil acidification. Water, Air and Soil Pollution 30, 909-930.
Reuss, J.O. and Johnson, D.W. (1986) Acid Deposition and the Acidification of Soils and Waters. Springer-Verlag Inc., New York.
Reuss, J.O. and Walthall, P.M. (1990) Soil reaction and acidic deposition. In Norton, S.A., Lindberg, S.E. and Page, A.L. Acidic Precipitation Volume 4: Soils, Aquatic Processes and Lake Acidification. Springer Verlag, New York, pp. 1-33.
Reuss, J.O., Cosby, B.J. and Wright, R.F. (1987) Chemical processes governing soil and water acidification. Nature 329, 27-31.
Reynolds, B. (1987) Precipitation acidity at a rural site in North Wales. Weather 42, 236-240.
Reynolds, B. and Ormerod, S.J. (1993) A review of current and future acid deposition trends in Wales. Report to the Welsh Office, Institute of Terrestrial Ecology.
Reynolds, B. and Edwards, A. (1995) Factors influencing dissolved nitrogen concentrations and loadings in upland streams of the UK. Agricultural Water Management 27, 181-202.
Reynolds, B., Neal, C. Hornung, M. and Stevens, P.A. (1986) Baseflow buffering of streamwater acidity in five mid-Wales catchments. Journal of Hydrology 87, 167-185.
Reynolds, B., Stevens, P.A., Adamson, J.K., Hughes, S. and Roberts, J.D. (1992) Effects of Clearfelling on Streams and Soil Water Aluminium Chemistry in 3 UK Forests. Environmental Pollution 77, 157-165.
Reynolds, B., Stevens, P.A., Hughes, S., Parkinson, J.A. and Weatherley, M.S. (1995) Stream chemistry impacts of conifer harvesting in Welsh catchments. Water, Air and Soil Pollution 79,
327
147-170.
Rodhe, H. and Rood, M.J. (1986) Temporal Evolution of Nitrogen Compounds in Swedish Precipitation since 1955. Nature 32^, 762-764.
Rodhe, H., Langner, J., Gallardo, L. and Kjellstrom, E. (1995) Global scale transport of acidifying pollutants. Water, Air and Soil Poiiution 85, 37-50.
Rosenqvist, I.Th. (1978) Alternative sources for acidification of river water in Norway. Science of the Total Environment 10, 39-49.
Rosenqvist, I.Th. (1990) From rain to Lake: Water Pathways and Chemical Change. Journal of Hydrology 116, 3-10.
Ross, H.B. and Lindberg, S.E. (1994) Atmospheric Chemical Input to Small Catchments. In Moldan, B. and Cerny, J. Biogeochemistry of Small Catchments: A Tool for Environmental Research. John Wiley & Sons Ltd, Chichester, England, pp. 55-84.
SAS Institute Inc., (1993) SAS/INSIGHT® User’s Guide, Version 6, Second Edition, Cary, NC: SAS Institute Inc., 1993.
Schnoor, J.L. and Stumm, W. (1985) Acidification of aquatic and terrestrial systems. In Stumm, W. (Ed) Chemical Processes in Lakes. John Wiley & Sons, New York, pp. 311-338.
Scire, J.A. and Venkatram, A. (1985) The contribution of in-cloud oxidation of SOg to wet scavenging of wet sulphur in convective clouds. Atmospheric Environment 19, 637-50.
Seip, H.M., Blaker, I.A., Christopherson, N., Grip, H. and Vogt, R.D. (1990) Hydrochemical Studies in Scandinavian Catchments. In Mason, B.A. (Ed.) The Surface Waters Acidification Programme, Cambridge University Press, Cambridge, pp. 19-30.
Semkin, R.G., Jeffries, D.S. and Clair, T. (1994) Hydrochemical Methods and Relationships for Study of Stream Output from Small Catchments. In Moldan, B. and Cerny, J. Biogeochemistry of Small Catchments: A Tool for Environmental Research. John Wiley & Sons Ltd, Chichester, England, pp. 163-188.
Skeffington, R.A. and Wilson, E.J. (1988) Excess Nitrogen Deposition : Issues for Consideration. Environmental Pollution 54, 159-184.
Smith, R.A. (1852) On the air and rain of Manchester. Memoirs and Proceedings of the Manchester Literary and Philosophical Society Series 2, 207-217.
Smith, C.M.S., Cresser, M.S. and Mitchell, R.D.J. (1993) Sensitivity to acid deposition of dystrophic peat in Great Britain. Ambio 22, 22-26.
Smith, R.I., Hall, J.R. and Howard, D.C. (1995) Estimating the uncertainty in the current critical loads exceedance models. Water, Air and Soil Poiiution 85, 2503-2508
Soil Survey of England and Wales (1983) Soil Survey of England and Wales 1:250 000 maps. Soil Survey of England and Wales
Stevenson, A.C., Patrick, S.T., Kreiser, A. and Battarbee, R.W. (1987) Palaeoecological evaluation of the recent acidification of susceptible lakes: methods utilised under DoE contract PECD 7/7/139 and the Royal Society SWAP Project. Paiaeoecoiogy Research Unit, Research Paper No.26 Dept, of Geography, University College London.
Stoddart, J.L. (1995) Episodic Acidification During Snowmelt of High Elevation Lakes in the Sierra Nevada Mountains of California. Water, Air and Soil Poiiution 85, 353-358.
328
Stoner, J.H.and Gee, A.S. (1985) Effects of Forestry on Water Quality and Fish in Welsh Rivers and Lakes. Journal of the Institute of Water Engineers and Scientists 39, 27-45.
Stoner, J.H., Gee, A.S. and Wade, K.R. (1984) The Effects of Acidification on the Ecology of Streams in the Upper Tywi Catchment in West Wales, UK. Environmental Pollution (Series A) 35, 125-158.
Sutton, M.A. and Fowler, D. (1992) Estimating the relative contribution SO , NO and NH inputs to effects of atmospheric deposition. In Hornung, M. and Skeffington, R.A (Eds.) Critical loads: concepts and applications. ITE Symposium No. 28, London, HMSG, pp. 119-126.
Sullivan, T.J., Christopherson, N., Muniz, I.P., Seip, H.M. and Sullivan, P.O. (1986) Aqueous aluminium chemistry response to episodic increases in discharge. Nature 323, 324-327.
Svenson, T., Dickson, W., Hellberg, J., Moberg, G. and Munthie, N. (1985) The Swedish Liming Programme. Water, Air and soil Pollution 85, 1003-1995.
Sverdrup, H. and Warfvinge, P. (1988) Weathering of primary silicate minerals in the natural soil environment in relation to a chemical weathering model. Water Air and Soil Pollution 38, 387-408.
Sverdrup, H. and Warfvinge, P. (1990) The role of weathering and forestry in determining the acidity of lakes in Sweden. Water, Air and Soil Pollution 52, 71-78.
Sverdrup, H., de Vries, W. and Henriksen, A. (1990) Mapping critical loads and examples for mapping critical loads. Nordic Council of Ministers Report 1990.14.
Sverdrup, H., de Vries, W. and Henriksen, A. (1990) Mapping critical loads - Guidance to Criteria Methods and Examples for Mapping Critical Loads. Nordic Council of Ministers Report 1990:14.
Sverdrup, H., Warfvinge, P., Rabenhorst, M., Janicki, A., Morgan R. and Bowman M. (1992) Critical loads and steady-state chemistry for streams in the state of Maryland. Environmental Pollution 77, 195-203.
Tamm, 0.0 and Hallbacken, L. (1988) Changes in soil acidity in two forest areas with different acid deposition: 1920s to 1980s. Ambio 17, 56-61.
Tarrasson, L., Turner, S. and Floisand, I. (1995) Estimation of dimethyl sulphide fluxes over the north Atlantic ocean to European pollution levels. Journal of Geophysical Research - Atmospheres 100, 11623-11639.
Taylor, P.J. (1993) A Pedagogic Application of Multiple Regression Analysis. Geography75,203- 212.
ter Braak, C.J.F. (1986) Canonical correspondence analysis : a new eigenvector technique for multivariate direct gradient analysis. Ecology 67 91-173.
ter Braak, C.F.J. (1987a) Ordination. In Jongman, R.H.G., ter Braak, C.J.F. and van Tongeren, O.F.R. (Eds.) Data Analysis in community and iandscape ecology. Pudoc: Wageningen, pp 91- 173.
ter Braak, C.F.J. (1987b) CANOCO - a FORTRAN Program for CANOnical Community Ordination by [Partial] [Detrended] [Canonical] Correspondence Analysis, Principle Components Analysis and Redundancy Analysis (Version 3.1). Agriculture Mathematics Group, Wageningen.
ter Braak, C.F.J. (1990) Update notes: CANOCO Version 3.10. Agricultural Mathematics Group, Wageningen.
329
ter Braak, C.F.J. (1995) Canonical community ordination. Part I: Basic theory and linear methods. Ecoscience 1, 127-140.
ter Braak, C.F.J. and Prentice, I.C. (1988) A Theory of Gradient Analysis. Advances in Ecological Research 18,(271-317)
Thompson, J.M. (1992) Visual Representation of Data Including Graphical Exploratory Data Analysis. In Hewitt, C.N. (Ed.) Methods of environmental data analysis. Elsevier Applied Science: London, pp. 213-238.
Thompson, M., Elder, F., Davis, A. and Whitlow, S. (1980) Evidence of acidification of rivers of eastern Canada. In Drablos D. and Tollen, A. (Eds.) International impact of acid precipitation. International Conference, Sandefjord, Norway, Proceedings: Oslo, Nonvay, SNSF project, pp. 244-245.
Townsend, G.S., Bishop, K.H. and Bache, B.W. (1990) Aluminium spéciation during episodes. In Mason, B.A. (Ed.) The Surface Waters Acidification Programme, Cambridge University Press, Cambridge, pp. 275-278.
UKAWRG (1986) Acidity in United Kingdom Fresh Waters, The United Kingdom Acid waters Review Group, First Report, Department of Environment.
UKAWRG (1988) Acidity in United Kingdom Fresh Waters, The United Kingdom Acid Waters Review Group, Second Report, Department of the Environment, December 1988.
UKRGAR (1990) Acid Deposition in the United Kingdom 1986-88. Third Report of the United Kingdom Review Group on Acid Rain, Department of the Environment, 1990.
Ullyet, J., Hall, J. and Bull, K. (1994) Mapping the Potential Sensitivity of Fresh waters to Acidification using Catchment Characteristics. In Battarbee, R.W. (Ed.) Acid rain and its impact: the critical loads debate. Proceedings of a conference held at the Environmental Change Research Centre, University College London. ENSIS Publishing, London, pp. 103-107.
UN ECE (1993) Manual on Methodologies and Criteria for Mapping Critical Levels/Loads and the Geographical Areas where they are Exceeded. Prepared by the Task Force on Mapping, Federal Environmental Agency, Berlin.
UN ECE (1994) The critical loads and levels approach and its application in preparation of the new S protocol. Report by the Bureau of the Working Group on Effects, EB.AIR/R.85.
Van Breeman, N., Burrough, P.A., Velthorst, E.J., van Dobben, H.F., de Wit, T., Bidder, T.B. and Reinjders, H.F.R. (1982) Soil acidification from atmospheric ammonium sulphate in forest canopy throughfall. Science 299, 548-550.
Van den Wolienberg, A.L. (1977) Redundancy analysis. An alternative for canonical correlation analysis. Psychometrika 42 207-219.
Van Leeuwen, E.P., Draaijers, G.P.J., Potma, C.J.M., Van Pul, W.A.J. and Erisman. J.W.(1995) The Compilation of Measurement Based European Wet Deposition Maps of Acidifying Components and Base Cations. Water, Air and Soil Pollution 85, 2173-2178.
Verdonschot, P.F.M. and Ter Braak, C.J.F. (1994) An experimental manipulation of oligochaete communities in mesocosms treated with chlorpyrifos or nutrient additions: multivariate analyses with Monte Carlo permutation tests. Hydrobiologia 278, 251-266.
Waters, D. and Jenkins, A. (1992) Impacts of afforestation on water quality trends in two catchments in Mid-Wales. Environmental Pollution 77, 167-172.
330
Waters, D., Jenkins, A., Staples, T. and Donald, A.P. (1991) The importance of hydrological source areas in terrestrial liming. Journal of the Institute Water and Environmental Management 5, 336-341.
Webster, R. (1977) Canonical correlation in Pedology: how useful? Journal of Soil Science 28, 196-221.
Webster, R. (1989) Is regression what you really want? Soil Use and Management 5, 47-53.
Welsberg, S. (1985) Applied Linear Regression, Second Edition. John Wiley & Sons, New York.
Werrltty, A. and Maucotel, S. (1993) Hydrochemistry of lake sediments and the impact of liming Loch Dee. In Acidification, forestry and fisheries management in upland Galloway - Proceedings of the Loch Dee Symposium. Gatehouse of Fleet, December 1992. Foundation for Water Research, pp. 49-60.
Wheater, H.S., Langan, S.J., Miller, J.D., Perrier, R.C., Jenkins, A., Tuck, S. and Beck, M.B.(1990) Hydrological processes on the plot and hillslope scale. In Mason, B.A. (Ed.) The Surface Waters Acidification Programme. Cambridge University Press, Cambridge, pp. 121-136.
Whitehead, P.O., Musgrove, T.J. and Cosby, B.J. (1990) Hydrochemical modelling of acidification in Wales. In Edwards, R.W., Gee, A.S. and Stoner, J.H. (Eds.) Acid Waters in Wales. Klewer Academic Publishers, The Netheriands, pp. 255-278.
Whittaker, J. (1984) Model interpretation from the additative elements of the likelihood function. Applied Statistics 33, 52-64 .
Williams, M.L. (1987) The impact of motor vehicles on air pollution and air quality in the UK - an overview. The Science of the Total Environment 59, 47-61.
Wilson, E.J., Skeffington, R.A., Maltby, E., Immlrz, P., Swanson, 0. and Proctor, M. (1995) Setting critical loads for acidity for ombrotrophic peat -a new approach. Water, Air and Soil Poiiution 85, 2491-2496.
Wilson, G.V., Jardine, P.M., Luxmoore, R.J. and Jones, J.R. (1990) Hydrology of a forested hillslope during storm events. Geoderma 46, 119-138.
Wolock, D.M., Hornberger, G.M. and Musgrove, T.J. (1990) Topographic effects on flow path and surface water chemistry of the Llyn Brianne catchments in Wales. Journal of Hydrology ^5, 243-259.
Wright, R., Conry, N., Dickson, W., Harrlman, R., Henriksen, A. and Schofield, 0. (1980) Acidified lake districts of the world: a comparison of water chemistry in lakes in southern Norway, southern Sweden, southwestern Scotland, the Airondack Mountains of New York, and southeastern Ontario. In Drablos D. and Tollen, A. (Eds.) international impact of acid precipitation. International Conference, Sandefjord, Nonway, Proceedings: Oslo, Nonway, SNSF project, pp. 377-378.
331
Appendix 4.1 : Sampling strategy originally adopted during Phase 2
Appendix 4.1 describes the sampling strategy and subsequent analytical chemistry for the
first Phase 2 fieldwork season. It is shown that these data not ultimately suitable for
inclusion in the Phase 2 analysis. The collection and analysis of these samples is included
to illustrate some of the potential pitfalls of water sampling programmes.
Site selection
A random stratified sampling strategy was employed to target sites across a critical loads
gradient. Each of the 976 sites in the Phase 1 dataset has a critical load value calculated
on the basis of measured water chemistry. For mapping critical loads on a national basis the
CLAG critical load values were arranged into class intervals (CLAG, 1995). Within this study,
the number of squares in each class was determined prior to the Phase 1 analysis. Critical
loads greater than S.Okeq ha' yr^ were omitted from the sampling framework because
the inclusion of these would result in the stratified sample being heavily weighted towards
high critical loads squares. A 2.5% sample was decided upon to restrict the fieldwork to a
manageable level while maintaining the integrity of the random approach. The number of
squares in each class required to make up a 2.5% stratified sample was calculated this
being proportional to the class structure of the national data. Table A.1 shows the
breakdown of this stratified sample.
Table A.1 Derivation of a random stratified sample from the CLAG database
Diatom critical load (keq H* ha y r ')
No. of squares
% of total
No. of squares in 2.5% stratified sample
<0.2 95 10.2 3>0.2 <0.5 216 27.5 9>0.5 <1.0 192 25.8 9>1.0 <2.0 173 23.3 8>2.0 <3.0 98 13.2 5
Total 744 100.0 34
332
The CLAG database was subsequently manipulated so that smaller datasets were created,
each comprising sites belonging to individual critical load classes. Sites within each subset
of the data were selected randomly so that the 2.5% sample criterion was met for each
class.
Most sites from the CLAG database are standing water bodies rather than streams. This
ameliorates some of the difficulties related to seasonal variability, an issue discussed more
fully in Chapter 4. It was initially intended to sample all standing water bodies in each of the
selected squares, thus eliciting information on variation in water chemistry within grid
squares as well as providing sufficient sites for calibration. Squares which contained fewer
than an arbitrarily defined number (five) of standing water bodies were avoided to ensure
the latter criterion was fulfilled. However once reference was made to the appropriate
1:50,000 Ordnance Survey (O.S) maps this approach was deemed unsuitable. A substantial
number of squares are characterised by a paucity of standing water bodies of an appropriate
size {i.e. larger than a small pond and small enough so that the entire catchment is
contained within a single 10km grid square). In addition many lowland sites and high altitude
headwater lakes are characterised by very small contributing catchments. This would
preclude the analysis of water chemistry/catchment attribute relationships given the spatial
scale of the catchment data used (see Chapter 4). Consequently, the sampling programme
was expanded to include streams and these ultimately provide the bulk of sites. This
necessarily requires a more subjective approach to individual site selection than if a finite
number of standing water bodies are to be sampled. With the inclusion of streams, the
number of potential sampling points increases beyond the point where it would be possible
to sample them all. It was therefore decided that five sites should be sampled within each
square. Each catchment is required to be wholly contained within the 10km grid square.
Where possible a range of catchment sizes was selected. Sites were selected immediately
adjacent to confluences or recognisable mapped features (e.g. bridges) to facilitate
333
catchment delineation.
Initially, the first three sites (ranked by random number) were chosen from the most sensitive
class (<0.2 keq ha' yr'^), the first nine from the next sensitive (>0.2 <0.5) and so on,
according to the sampling breakdown described in Table A.1. The O.S maps relating to each
of these squares were examined to determine suitability for sampling. A number of squares
were rejected because they contained too few catchments or the land area was insufficient
(e.g. coastal grid squares). Other squares were characterised by poorly defined catchments
(e.g. Benbecula in the Outer Hebrides, an area with many lochs but negligible relief) and
were not therefore selected. If squares were rejected, the next randomly ranked grid square
was examined. The final selection comprises 34 10km grid squares, 5 in Wales, 3 in
England and 26 in Scotland.
However during the course of the fieldwork the sampling strategy was refined. After
sampling 5 squares in Wales and 9 in Scotland the selection criteria for sites was amended
and the modified strategy is described in Chapter 4.
Sampling strategy
Ultimately 45 sites were sampled as part of this stratified random sample. Three water
samples were taken at each site. An acid washed 250ml high density polyethylene (HOPE)
was filled and delivered to Fresh waters Fisheries Laboratory, Pitlochry (FFL) within a week
of sampling, to determine alkalinity, conductivity and pH. During sampling, bottles were
rinsed in the sample lake or stream and the lid was tightened undenwater to ensure that the
amount of trapped air was minimal. Samples were taken at a depth of approximately 40cm.
Measurements of pH were also made in s/fu with a Jenway 3010 pH meter with a standard
combination electrode. Further samples were also collected in two acid washed 60ml HOPE
334
bottles, one for cation analysis and one for anions. To reduce levels of particulate matter
and the possibility of contamination by micro-organisms (Creasey et al., 1986) these were
filtered through 0.45 micron Whatman cellulose nitrate filters using a plastic syringe and a
Swinex filter holder, both rinsed at the site. The sample for cation determination was
acidified using 3ml of ARISTAR nitric blank to prevent precipitation of iron and other metals
(Golterman at a!., 1978). The other sample was later analysed for anion concentration. Both
these samples were analysed at the end of the fieldwork which was spread over a five week
period. Where lakes or reservoirs were sampled the sample was taken at the outflow.
Additionally, a sample was taken from the headwaters at each site so that the water
chemistry could be compared with that at the sample site. This analysis was discontinued
following the modification of the sampling framework as there was no meaningful relationship
between differences in water chemistry at the site and the headwater stream and the
distance downstream (Figure A4.1).
Analytical chemistry
Water chemistry samples collected during the first Phase 2 fieldwork period were analysed
at three laboratories to determine water chemistry. Alkalinity, pH and conductivity were
analysed at FFL. Alkalinity and pH were measured using a Radiometer TTT85 titration
system with a remote reference electrode and 40cm KCI head. Alkalinity values were
produced using dual endpoint titration, titrated to endpoint of pH 4.5. Conductivity was
analysed using a Philips PW9256 digital conductivity meter. Major anions (SO/', NOg' and
Cl ) were determined at the Department of Geology, UCL, using a Dionex ion chromatograph
system, with autosampler, after the samples had equilibrated to room temperature. Samples
were diluted (x10 or x25 where high concentrations necessitated) and standards were run
after every sixth sample. Standards comprised Cl' at 2 parts per million (ppm), NOg' at 3ppm
and s o / ' at 5ppm. Base cations (Ca '', Na"", and Mg "") were determined by inductively
335
Figure A4.1: Scatterplot of differences in Ca '' concentration between sampling site and headwater stream vs distance between the two
6000
5000-•
t - ' iII1JZ
4000"
02 3000"
1oc 2000"
Q
1000"
-1500 -1000 -500 0Site Ca minus headwater Ca (microeq/1)
-2000 500 1000
coupled plasma mass spectrometry (ICP/MS) using the NERC facility based at the
Department of Geology, Royal Holloway, University of London (RHUL). Standards comprised
Ca "" and Na at SOppm and and Mg "" at 10ppm. Samples were not diluted and standards
were measured every six samples to facilitate drift correction. Samples were initially
determined as ppm and subsequently converted to microequivalents per litre (peq 1' ).
Once the data for the initial field season had been collated a number of problems became
apparent. Three control samples were also collected during this fieldwork and these were
analysed fully at FFL. A comparison of the results between the FFL values and those from
UCL Geology (anions) and RHUL (cations) is presented in Table A4.2
336
Table A4.2 : Comparison of chemistry values (ueq determined by FFL with other laboratories used in during the initial Phase 2 fieldwork
Variable Site FFL UCL RHUC Diff.* % diff
cr SNB11 375 461 - 8 6 18.6
S121 223 441 - 218 49.4
S170 267 304 - 37 1 2 . 1
NOg- SNB11 2 27 - 25 92.5
S121 1 32 - 31 96.9
S170 15 25 - 1 0 39.6
804 /- SNB11 67 51 - -16 -31.8
S121 46 61 - 15 24.6
S170 187 160 - -27 -16.7
Ca^" SNB11 43 - 55 1 2 21.7
S121 161 - 159 - 2 - 1 . 2
SI 70 117 - 115 - 2 -1.9
Na+ SNB11 322 - 309 -13 -4.3
S121 208 - 198 - 1 0 -5.3
S170 288 - 275 -13 -4.7
K" SNB11 1 0 - 26 16 60.9
S121 9 - 18 9 49.4
S170 17 - 18 1 4.9
Mg2 " SNB11 78 - 89 1 1 1 2 . 2
S121 6 6 - 76 1 0 13.0
S170 106 - 109 3 2.7
NB. SNB11 is a CLAG mapping site.* Difference between FFL and UCL or RHUL value.
It is apparent that there are significant differences between the two sets of results. NOg* is
substantially higher for the samples analysed at UCL compared to those from FFL. Similarly,
values for K"" are substantially higher in samples analysed at FFL. Additionally, closer
337
examination of the chemistry suggests that the UCL Cl values are also suspect as there is
insufficient Na"" to give a reasonable sea-salt ratio. The problems with NO ' are likely to have
arisen from two error sources. The use of cellulose nitrate filters is likely to have led to
nitrate contamination. Additionally, sample dilution of the water used for anion analysis at
UCL means that, for many sites, a value close to the detection limit is being multiplied by
a factor of 10. It is probable that K* values differ for the same reason. The correct
procedure requires that straight samples are run (as was the case for cation analysis at
RHUL) with a modification of the calibration range rather than sample dilution (R.Harriman,
pers comm). Golterman et al., (1978) noted that contamination by compounds of nitrogen
may occur with the use of cellulose filters and suggested that a correction might be provided
by using a blank of equal volume of HgO. Given the additional problems encountered as a
result of sample dilution, no attempt at correction was made. Given the uncertainties with
this chemistry data it was decided that only sites sampled during the second Phase 2 field
season would be used to calibrate the predictive model. However, the omission of the sites
sampled in the first season means that there are no catchments characterised by the Ettrick
and Thurso soil associations. These cover 9.3% and 1.35% of the land area of Scotland,
respectively (Macauley Institute for Soil Research, 1994). Although these are major soil
associations, the range of soil sensitivity covered by the sites sampled during the second
field season is such that it was not felt necessary to resample in these areas.
338
U)U)VO
Appendix 4.2: Phase 2 site locations •
Site Grid Grid
code Easting Northing Name Catchment type Other observations
mkOl 2900 6847 Muirdyke Burn Arable land, stubble V. muddy, shallow, slow moving
mk02 2880 6875 Bum Grassland Shallow, gravelly substrate
mk03 2868 6867 Sauchinford Burn Pasture Fast flowing, gravel substrate
nrik04 2855 6858 Tor Bum Pasture & rough grassland 3m deep
mk05 2828 6818 Bum Rough pasture Gravel substrate, .5m deep
mk06 3078 7138 Binzian Bum Pasture & woodland
mk07 3020 7145 Dunning Bum Flowing through town approx 4m wide
mk08 3037 7150 Garrock Bum Improved grassland Gravel substrate, 2m wide, 2m deep
mk09 3050 7157 Broomhill Bum Rough pasture & woodland .2m deep & 1.5m wide
mklO 3077 7158 Culteuchar Bum Pasture Shallow, 1 5m wide
mkl 1 3084 7301 Bum Improved grassland Silty bottom, 3m deep, l-2m wide
ink 12 3052 7344 Ordic Burn Mixed pasture/arable Shallow, many macrophytes, Im wide
mkl3 3048 7357 Garry Bum Arable land 2m wide, .5m deep
mkl 4 3055 7364 Glenshauch Burn Pasture & woodland Adjacent to cattle farm, shallow, v.firie gravel substrate, 1.5m wide
mkl 5 3095 7391 Geliy Bum Rough pasture/deciduous wood 3m wide
mkl6 2795 7229 Balmenoch Bum Deciduous woodland/pasture Shallow, gravel substrate, < lm wide
mkl7 2743 7220 Allt an Tamhaisg Upland & deciduous woodland 3-4m wide, shallow, fast flowing, v.clean, pebbly substrate
ê
mkl 8 2768 7258 Lurg Bum Heather, bracken, rough pasture 4m wide, fast flowing, gravel substrate
mkl9 2734 7278 Allt na Faing Coniferous woodland l-2m wide, shallow, gravelly substrate
mk20 2753 7290 Bum Bracken, heather upland 2m wide, fast flowing, gravel substrate
mk21 3188 7722 Glen Brighty Moorland & coniferous forest 4-5m wide, gravel & boulder substrate, shallow & fast flowing
mk22 3195 7756 Caenlochan Bum Upland, grasses & peat 3m wide, fast flowing, gravel substrate
mk23 3146 7703 Allt an Daimh Upland , heather Im wide, shallow
mk24 3126 7738 Allt Coolah Rough pasture/rocky outcrops 3-4m wide, gravel.... pool & riffle sequence
mk25 3134 7756 Allt Choire Dhirich Rough pasture/rocky outcrops 3-4m wide, gravel.... pool & riffle sequence
mk26 3572 8114 Sheil Bum Pasture/improved grassland l-2m wide, gravel substrate
mk27 3538 8125 Ininteer Bum Improved grassland Cattle grazing, I-2m wide, fast flowing & shallow
tnk28 3549 8153 Strow Bum Pasture, woodland l-2m wide, gravel substrate
mk29 3564 8183 Bum Pasture l-5m wide, slow flowing, shallow, v.muddy - no stones
mk30 3527 8174 Boggerie Bum Improved pasture & arable Grazing, muddy, shallow & fast flowing
mk31 3461 8212 Murchte Burn Improved pasture/grassland V. fine gravel & mud substrate, l-2m wide
mk32 3465 8230 Bum of Corinthian Improved grassland & woodland 2-3m wide, gravel substrate, shallow & fast flowing
mk33 3433 8253 Bum of Buck Moorland Peaty water colour, fast flowing, l-2m wide
mk34 3433 8273 Bum of Longley Coniferous forest Coarse gravel substrate, ferrous deposits, l-2m wide
mk35 3478 8273 Bum of Tonbum Arable & pasture l-2m wide, gravel & silt substrate
mk36 3260 8500 Back Bum Coniferous wood & pasture V. fast flowing, 3-5m wide, shallow, downcut to bedrock .no gravel
mk37 3292 8510 Bum of Sourden Upland catchment
mk38 3296 8573 Red Bum Upland catchment
mk39 3213 8544 Gedloch Bum Coniferous woodland & rough pasture Reddy brown water, 2-3m wide, .5m deep
mk40 3247 8542 Glen Bum Pasture & conifers l-2m wide, quite deep
U)4
ink42 2473 8807 Allt Coire nam Meanh Moorland, heather 4-5m wide, boulder strewn, shallow & fast flowing
mk43 2468 8854 Abhaina Coire a Mhalagain Moorland, heather 6-7m wide, boulder strewn, shallow & fast flowing
mk45 2458 8888 Allt Coire nan Con Moorland, heather l-2m wide, ,5m deep
mk46 2833 8447 Allt Coire nan Poite Upland heath, peat l-2m wide, coarse gravel
mk47 2856 8448 Allt na Cliche Upland 2-3m wide, coarse gravel
mk48 2842 8442 Allt Meallin Gobhar Moorland & some conifers 2-3m wide, small-medium pebbles
mk49 2839 8422 Allt Mor Upland 2-3m wide, coarse gravel
mkSO 2867 8406 Abhainn Cumhang a Ghiinne
Upland, bracken, heather 3-4m wide, coarse gravel
mk5I 1777 8274 Loch Skalpaidh outflow Upland l-2m wide, fast flowing
mk52 1736 8262 Allt Anavig Coniferous forest 3m wide, coarse gravel & boulders
mk53 1714 8253 Allt an Dariach Coniferous forest, bracken 3m wide, shallow, coarse gravel & boulders
mk54 1731 8216 Allt Mor Upland, heather 2-3m wide, coarse gravel
mk55 1788 8233 Allt a Choire Bhuidhe Moorland, heather/coniferous woodland l-2m wide, flowing through gorge
mk56 1475 8376 Easa na Coille Moorland 3-4m wide, coarse gravel
mk57 1476 8368 Lon Chaorach Moorland 2-3m wide, mixed gravel
mk58 1453 8320 Alt Mor Moorland, pasture 2-3m wide, bedrock, some gravel
mk59 1413 8321 Allt na Guile Moorland, bracken, heather/rough pasture 3-4m wide, coarse gravel
ink60 1405 8323 Allt coir a Ghobhainn Moorland
ink6! 1974 7781 Fionn Alltan Mixed deciduous woodland 2-3m wide, coarse gravel substrate
mk62 1925 7794 Allt Dubhaidh Coniferous forest 2-3m wide, fast flowing, reddish water
ink53 1924 7783 Allt an Tairbeart Fhaing Upland, bracken, heather 3-4m wide, shallow, coarse gravel substrate
mk64 1993 7748 Allt Laith Coniferous woodland 2-3m wide, coarse gravel & boulder strewn
U)6
mk65 1980 7776 North Garvan River Grassland, moorland & mixed forest lOm wide, coarse gravel substrate
mk66 2482 7873 Coille a Choire Upland Broad floodplain, 10-12m wide, coarse & fine garvels, boulders
mk67 2479 7871 Allt a Choire Choniharsain Upland, bracken Broad floodplain, 10m wide, coarse to fine gravel
mk68 2445 7842 Allt Choire Choilte-rais Woodland & coniferous forest 10m wide, coarse gravel & boulders
mk69 2421 7828 Moy Burn Upland 3-4m wide, v. fast flowing, bedrock with some coarse gravel
mk70 2410 7808 Bum Coniferous forest & moorland Im wide, fast flowing, shallow, bedrock & coarse gravel substrate
mk7I 2094 7261 Allt Lorgaidh Deciduous woodland l-3m wide, pool riffle sequence, coarse gravel
mk72 2017 7279 Allt Mhit O’ Haragain Deciduous woodland 3m wide, fast flowing
mk73 2071 7236 Allt an Dunain Mixed woodland (conifer plantations) 5m wide, coarse gravel substrate
mk74 2036 7224 Allt na h-Airigh Pasture & deciduous woodland 3-4m wide, fast flowing & deep
mk75 2024 7211 Allt Ath nan Each Moorland & deciduous wood l-2m wide, fine gravel substrate
mk76 1402 6624 Barr River Moorland, bracken & grass 2-3m wide, fast flowing
mk77 1445 6680 Abhainn Mor Moorland, bracken & grass Reddish colour, 3m wide, gravel substrate
mk78 1445 6674 Abhainn an Daimh-sgeir Peat moorland, bracken, heather 3-4m wide
mk79 1464 6642 Allt am Torrain Bhuidhe Heather, bracken l-2m wide, variable depth, coarse gravel
mkSO 1474 6634 Struthan Achadh an datha Dhuibh
Heather, grassland l-2m wide, shallow, gravel substrate
Appendix 4.3: Scatterplots of calcium concentration (peq 1 ) against flow (cumecs) for selected Acid Waters Monitoring Network sites
Allt a’Mharcaidh
60-
55-
#50-
I 4 5 --
40-----
35-#
30-
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9Flow (cumecs)
0.1
Allt na Coire nan Con125-
115--
105--
95--
85--
I 7 5 -
u65--
55--
45--
35--
0.2 0.3 0.4 0.5Flow (cumecs)
0.6 0.7 0.9
343
River Ethrow220-
210- -
200- -
190- -
E 180- -
U
170- -
160-"
150-"
140-0.5 1 1.5
Flow (cumecs)2.5
100-
Dargall Lane
9 0 -"
80 -"
70 -"
E 60 -"
U
50- -
40 -"
30 -"
20-1-----------1----------- 1-----------1----------- 1-----------1----------- 1-----------r0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Flow (cumecs)
—I---------- 1-------0.8 0.9 1
344
Old Lodge360
320-
280-'
240-
^ 200-"^
160-;
120-1
80
% *
%
t .
0 002 004 006 008 OJ 012 014 016 018 0.2Flow (cumecs)
Narrator Brook
70-
60-c ro
Ô50-
40-
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9Flow (cumecs)
345
I K)-
lOO
90- -•
80- -- ! t
70
Afon Gwy
60- ■
40- •
30--
20- -
10- -
0 —T“0.2
—r-0.4
— T -0.6 ~L2 Tô"0 0.8 1
Flow (cumecs)1.4 1.8
Afron Hafren110
100-
90-
80-
70-
E
a 50-i
40-«
30-
20-
10-
0.2 0.4 0.6 0.8 1.4 1.6Flow (cumecs)
346
350"n»
300-
250-
200
1 0 0
50
0
«
Beaghs Burn
0 005 OJ 015 02 025 03 035 0.4Flow (cumecs)
Bencrom River180
160-
140-
1 2 0 -
I 100-
80-
60—
0.1 0.3Flow (cumccs)
0.2 0.4 0.5 0.6 0.7
347
Appendix 4.4: Water chemistry for Phase 2 sites (including critical loads values)
site pH Aik Cond Na NH4 K Mg Ca 0 1 N03 S04 AI-TM Ai-NL Al-L ABS-250 HGL DOL
mk0 1 7.94 2116 476 1240 0 87 1270 3597 1269 238 1722 9 4 5 0.272 44.7 37.8mk0 2 7.76 3264 444 1115 0 138 1462 3847 661 271 1306 19 7 1 2 0.404 54.5 40.7
mk03 7.55 1483 304 593 0 83 672 2328 638 470 664 5 2 3 0.124 27.7 24.5mk04 7.55 2402 341 829 0 106 1049 2351 548 195 747 14 1 0 4 0.158 34.8 24.8mk05 7.4 6 8 6 145 347 0 33 284 1109 344 109 322 8 5 3 0.162 12.5 1 1 . 1
mk06 7.67 1 2 1 1 186 425 0 25 506 1369 411 82 225 6 2 4 0.168 18.3 15.8mk07 7.78 788 130 300 0 2 1 388 974 263 76 193 9 2 7 0.199 13.4 1 1 . 0
mk08 8.03 1607 233 422 0 38 648 1728 426 181 239 4 1 3 0.161 23.5 20.3mk09 7.91 1199 194 412 0 39 502 1541 390 154 291 7 1 6 0.259 19.7 17.2mk1 0 8.15 1627 205 370 0 15 482 1723 274 79 245 0 0 0 0 . 1 22.4 2 0 . 0
mk1 1 7.71 1037 284 445 0 43 660 1936 547 612 500 1 0 1 0.086 19.9 2 0 . 1
mk1 2 7.53 814 2 2 1 357 0 56 481 1549 474 553 278 2 1 1 0.089 16.6 17.4mk13 7.27 396 8 6 193 0 1 1 2 2 2 498 193 46 173 6 2 4 0.188 5.5 5.0mk14 7.56 743 188 272 0 50 476 1368 374 387 309 7 2 5 0.149 14.7 14.8mk15 7.54 744 191 470 0 60 374 1300 504 166 394 63 14 49 0.487 12.7 13.0mk16 7.34 384 81 230 0 28 246 380 213 28 148 1 0 3 7 0.184 9.5 3.9mk17 7.03 148 54 183 0 6 72 303 179 13 143 7 3 4 0.063 4.6 2.7mk18 6.93 1 2 1 40 155 0 1 0 104 185 142 3 8 6 23 16 7 0.19 4.2 1.9mk19 7.01 140 43 158 0 8 103 207 165 1 75 14 1 1 3 0.169 4.3 2 . 2
mk2 0 6.89 1 2 2 35 1 2 2 0 1 1 84 175 104 2 76 14 13 1 0.13 3.9 1 . 8
mk2 1 7.44 293 45 119 0 9 92 313 84 5 6 6 17 7 1 0 0.118 5.6 3.6mk2 2 7.29 303 50 1 0 2 0 7 1 1 2 370 74 8 125 0 0 0 0.037 5.9 3.7mk23 6.64 113 40 141 0 15 74 225 146 2 1 1 2 1 1 8 3 0.141 3.1 2 . 1
U)U\o
site pH Aik Cond Na NH4 K Mg Ca Cl N03 S04 AI-TM AI-NL Al-L ABS-250 HCL DCL
mk24 7.53 397 54 116 0 8 108 432 85 2 75 2 0 15 5 0,158 7.4 5.0
mk25 7,55 459 64 103 0 1 0 131 517 8 6 4 116 29 7 2 2 0.143 8 . 2 5,6
mk26 7,73 8 6 6 209 619 0 74 548 1251 584 236 372 2 1 1 0.088 13.4 13.1
mk27 7,74 859 230 645 0 67 552 1357 760 264 330 5 2 3 0.106 13.4 14,7
mk28 7,35 407 116 478 0 43 274 434 403 1 1 1 198 5 3 2 0.146 5.5 4.3
mk29 6,81 640 231 731 27 47 590 1233 615 582 364 0 0 0 0,037 14,1 13.1
mk30 7,35 438 164 529 0 72 413 865 524 341 259 2 0 2 0.068 9.5 9.2
mk31 7.01 384 93 376 0 23 217 384 295 95 130 2 2 0 0.053 5.5 4.2
mk32 7,6 744 118 320 0 25 736 410 307 32 134 1 0 4 6 0.191 1 0 . 0 5.3
mk33 7,24 431 8 6 285 0 18 485 305 292 0 1 1 0 38 2 0 18 0.515 6 . 6 3.7
mk34 7.29 485 117 346 0 40 326 547 447 2 0 180 18 17 1 0.195 6.5 5.7
mk35 7,45 774 154 415 0 44 512 933 451 148 193 3 3 0 0.05 1 2 . 0 10.7
mk36 5.27 16 65 382 0 14 157 170 339 2 169 144 107 37 1.314 2 . 0 1.3mk37 7,43 509 139 551 0 49 195 897 553 1 249 2 2 1 0 1 2 0.252 7,9 9,3
mk38 7,21 337 1 1 2 430 0 42 165 545 458 29 2 2 0 43 28 15 0.424 4.5 5,1
mk39 5,9 64 71 380 0 2 1 181 216 367 7 129 139 131 8 1.481 2.7 2 , 0
mk40 5,85 58 105 472 0 45 238 319 461 5 419 350 258 92 0.821 2 . 1 0 . 6
mk42 6,57 70 41 256 0 1 1 87 127 236 3 53 1 0 0 75 25 0.486 2.7 1.4mk43 6,81 89 43 253 0 9. 84 141 227 2 51 69 57 1 2 0.376 2,9 1.5mk45 6,76 99 48 288 0 9 8 8 164 261 2 55 49 44 5 0.482 3.2 1 . 8
mk46 6,51 44 25 160 0 7 58 71 134 4 38 9 9 0 0.062 3.3 0,7
mk47 8 , 2 1 1692 169 276 0 8 937 1163 214 6 57 14 1 13 0.06 56,1 15.7mk48 6,81 1 2 0 39 226 0 8 1 0 1 130 2 1 0 3 36 8 6 2 0,093 5,7 1,5
mk49 7,98 880 114 307 0 1 2 431 792 264 5 58 1 0 2 8 0.074 32.4 10,3mk50 7,42 602 1 0 2 372 0 19 194 564 360 2 77 6 3 3 0.076 19.0 6.9
mk51 7,17 186 6 8 393 0 1 0 1 2 0 235 386 2 48 23 2 1 2 0.361 6.4 2 , 8
mk52 6,76 95 48 302 0 8 1 1 0 136 289 1 43 48 47 1 0,465 4,4 1,5mk53 6 . 6 6 104 62 402 0 7 185 136 383 1 48 75 6 8 7 0,917 5,9 1 , 6
mk54 7,46 334 6 6 281 0 7 104 349 264 3 42 18 8 1 0 0.148 8.7 4.4mk55 7,25 206 48 236 0 8 •■94 2 2 0 209 3 45 13 8 5 0,123 6 , 1 2 , 6
mk56 7.73 689 1 0 1 435 0 7 416 387 310 0 29 1 0 8 2 0.308 24,4 5.7mk57 7,64 477 84 369 0 8 304 308 314 3 48 1 0 5 5 0 . 2 0 1 17.0 4,1mk58 7,81 657 99 418 0 9 405 371 313 " 2 39 1 1 5 6 0 . 2 2 1 23,0 5,3mk59 7,55 484 8 6 406 0 9 250 326 325 2 35 15 8 7 0.283 17.1 4,5
w
site pH Aik Cond Na NH4 K Mg Ca Cl N03 S04 AI-TM AI-NL Al-L ABS-250 HCL DCL
mk60 7.6 548 90 430 0 9 238 377 327 0 28 26 19 7 0.379 18.7 5.2mk61 5.16 2 38 233 0 7 63 72 195 0 59 116 78 38 0.648 4.0 0.7mk62 5.14 2 39 234 0 9 61 74 203 2 52 80 49 31 0.641 3.9 0.7mk53 5.98 15 27 158 0 13 56 60 155 0 50 28 18 1 0 0.174 2 . 8 0 . 6
mk64 6.41 40 26 153 0 9 55 94 131 1 47 38 31 7 0 . 2 0 1 4.3 1 . 0
mk65 6 . 1 35 32 186 0 1 0 62 96 171 2 50 43 35 8 0.294 4.3 1 . 0
mk6 6 6.59 60 2 1 1 0 1 0 8 42 99 75 3 41 2 2 19 3 0.131 2.5 1 . 1
mk67 6 . 2 1 29 2 0 118 0 8 39 78 90 1 41 43 36 7 0 . 2 1 2 . 1 0 . 8
mk6 8 6.69 57 2 0 98 0 8 40 1 0 1 73 1 41 18 17 1 0.094 2.5 1 . 1
mk69 6 . 6 55 2 1 107 0 1 0 42 1 0 2 81 1 42 2 0 2 0 0 0.152 2 . 6 1 . 1
mk70 6.95 157 46 216 0 8 65 231 181 1 69 49 36 13 0 . 2 0 1 5.0 2.5mk71 6.94 103 42 2 0 1 0 15 76 175 194 2 85 15 1 0 5 0.087 5.7 1.7mk72 7.07 209 57 270 0 1 1 107 273 248 4 64 30 23 7 0.343 1 0 . 0 3.2mk73 7.15 275 6 8 255 0 15 94 394 256 1 0 75 33 27 6 0.569 12.3 4.5mk74 7.25 290 70 273 0 7 82 418 266 5 72 25 24 1 0.56 12.7 4.9mk75 7.01 2 1 1 59 267 0 1 1 85 293 252 3 61 25 23 2 0.432 9.9 3.4mk76 7.69 1031 160 428 0 13' 144 1380 468 5 85 27 2 2 5 0.617 21.4 16.7mk77 4.87 - 1 2 6 8 393 0 2 0 1 2 2 82 420 2 8 6 94 81 13 0.828 1 . 8 0.7mk78 5.11 2 69 408 0 19 130 81 432 1 79 119 83 36 0.977 2 . 0 0.7mk79 5.95 47 82 553 0 2 2 187 137 540 1 84 87 74 13 1.03 4.2 1.4mk80 6 . 8 6 192 1 0 1 643 0 2 1 233 250 623 2 92 41 38 3 0.727 6 . 6 2.9
Appendix 4.5a : Summary of ITE Land classification system (from Bunco et a/.,1982).
Land class 1
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
8 Wales, S.W. England, 8. England.Alluvial plains, low ridges or plateaus with little surface drainage. Gently rolling country or almost flat country at medium/ low altitude. Varied lowland landscapes with hedges, trees and farm buildings. Cereals, good grasslands and limited native vegetation.Mainly brown earths but also gleys.Limited but grassland where present.
Land class 2
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
8. England, S.W. Midlands.Downland summits and scarps, low ridges or occasionally alluvial plains. Sweeping curves or smooth slopes with land at medium/ low altitude.Mainly open or wooded downland with few hedges and scattered farm houses. Mainly good grasslands but extensive cereals and built up land.Brown earths or calcareous brown earths.Rough grassland or bracken where present.
Land class 3
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
E.Anglia, S.E. England.Alluvial plains or shallow river valleys with low broad ridges Flat or almost flat with virtually all land at low altitude.Prairie type lowlands with intensive agriculture and declining hedges. Cereals and other crops and short term grassland.Gleys, calcareous brown earths and brown earths.Virtually absent.
Land class 4
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
E.Anglia margins, S. England, S. Midlands Fenland or flood plains with intricate drainage patterns.Flat or virtually flat, almost entirely at low altitude.Intensive farmed lowlands often under urban pressure.Arable with cereals and other crops, good grassland and urban. Gleys with some calcareous brown earths.Virtually absent.
Land class 5
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
S. England, S.W. England, S.W. Midlands, S.Wales.Variable from scarpland to downland and valley floors.Uniform gentle slopes or smooth outlines mostly at low altitude. Varied lowlands with many natural features.Mixed farmland although predominantly good grass: much urban. Gleys and brown earths.Limited but varied where present from bracken to rushes.
Land class 6
Geography: Land form: Topography: Landscape: Land use: Soils:
S.W. England, S.W. Midlands, S.Wales.Dissected tablelands and plateaus with many small rivers.Complex with many broad even slopes and the majority of land at medium/low altitude. Intricate with small fields enclosed by hedges on banks with small woodlands.Mainly good grassland but with some barley.Gleys and brown earths predominate.
352
Vegetation: Limited to small areas.
Land class 7
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
S. England, S.W. England and Wales coasts.Variable coastal morphology, mainly cliffs cut into tablelands. Usually coastal cliffs, rarely estuarine, most land at low altitude. Varied coasts backed by lowland farmland with farm houses. Mainly pasture with some arable and good grass.Brown earths but also other types.Limited but varied, particularly moorland and grassland types.
Land class 8
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
E. Anglia, S. England, Wales and N.W. England coasts.Marine alluvial plains bordering estuaries or, rarely, rocky coasts.Mainly flat but with some steeper coasts, mostly at low altitude.Usually flat coasts backed by good farmland affected by urban development. Mainly pasture but some arable, extensive mudflats and urban development. Gleys and brown earths.Limited, but rough grassland where present.
Land class 9
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
N. Midlands, N.E. England, S.E. Scotland.Mainly valley floors and flood plains of large rivers together with bluffs. Almost flat or gently rolling, most land medium/ low altitude.Open lowland country often with declining hedges, intensive agriculture. Mixture of good grass and arable with many urban areas.Brown earths, gleyed brown earths and gleys.Very limited, bracken or rough grassland where present.
Land class 10
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
N. Midlands, N.E. England, S.E. Scotland.Mainly valley floors or alluvial plains often with moderate scarps on margins.Gentle slopes, often long with the majority of land medium/ iow but also low altitude. Well farmed lowland country with many hedgerows and small woods.Mainly arable but with good grassiand and pasture also widespread.Gleys with some brown earths.Very restricted.
Land class 11
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
E. and C. Midlands.Alluvial plains or low broad ridges drained by small streams.Gentle slopes, or flat with almost all land at low altitude.Open landscape with large fields and declining hedgerows.Arable predominates, particularly wheat with good grassland and urban. Gleys and brown earths.Very restricted.
Land class 12
Geography: Land form: Topography: Landscape: Land use: Soils:
E. Midlands and Fens.Mainly fens or flood plains and large rivers, otherwise graded ridges. Flat or almost flat, entirely at low altitude.Prairie landscapes with derelict hedges and urban development. Arable, mainly wheat with limited good grassland and urban.Gleys and brown earths.
353
Vegetation: Virtually absent.
Land class 13
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
N. Wales, N.W. England, S.W. Scotland.Heterogeneous from low ridges in alluvial plains to scarps and river valleys. Smooth slopes, often steeper almost entirely at low altitudes.Varies lowland landscapes with hedged small fields often affected by urban. Usually mixtures of arable and good grassland but also a variety of other uses. Gleys and brown earths predominate but other types often present.Bracken or rough grassland, but also some moorland.
Land class 14
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
N.W. and N.E. England, S.W. Scotland.Mainly marine or alluvial flood plains bordering estuaries; rarely rocky coasts.Flat or gently sloping with the majority of land at low altitude.Prairie landscapes with fences or neglected hedges much affected by urban development. Mainly arable but also good grassland and much urban.Gleys, gleyed brown earths and brown earths.Very little present.
Land class 15
Geography: Land form: Topography:
Landscape: Land use: Soils:Vegetation:
Wales, N. England.Variable from dissected plateaus to valley floors bordered by escarpments.Complex with shallow or occasionally steep slopes, flat land almost entirely medium/low altitude.Intricate lowland landscapes with many natural features.Mainly pasture mixed with arable and good grassland.Gleys, brown earths and some brown podsols.Restricted but mainly rough grassland and some bracken.
Land class 16
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
N.England, S.W. Scotland.Flood plains or valley floors with escarpments or gently folded. Mainly undulating land with some flat areas mainly at low altitude. Varied lowland, well farmed landscapes with many hedges.Varied with mixtures of arable pasture and good grassland.Brown earths and gleys.Varied but with grassland types predominating and some moorland.
Land class 17
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
S.W. England, Wales, N.England.Plateaus or tablelands with scarps often dissected by small rivers.Some gentle slopes but mainly quite steep hillsides at medium/high altitude.Open or enclosed marginal uplands with walls, fences and occasional farmhouses. Mainly pasture with some good grassland.Brown earths and brown podsolics but a range of other soils.Mainly rough grassland types but also some moorland.
Land class 18
Geography: Land form: Topography:
Landscape:
Wales, N. England, W. Scotland.Glaciated river valleys with steep scarps often dissected by small rivers.Steep hillsides predominate with some more moderate slopes mainly at medium high altitude.Mainly open rugged uplands but with some areas transitional to enclosed land.
354
Land use: Soils:Vegetation:
Predominantly rough grazing with some limited pasture land. Brown podsolics, brown rankers peats and other upland types. Mainly moorland with extensive peatland and montane grassland.
Land class 22
Geography: Land form: Topography:
Landscape: Land use: Soils:Vegetation:
N. England, S. C. and N. Scotland.Dip slopes of plateaus or broad glacial valleys leading to rounded summits.Slopes of variable gradient from steep to moderate and almost entirely at medium high altitude.Mainly high moors but sometimes enclosed or afforested.Mainly rough grazing but also woodland and occasional crops.Peaty gleys, peaty podsols and peats but also other upland soils.Mainly moorland types.
Land class 23
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
N. England, C. and N. Scotland.Ridges scarps and corries leading to mountain summits or rarely glaciated valleys. Extremely steep hillsides, sometimes less so, with the land at high altitudes.Open mountainous landscapes with wide vistas.Limited open range grazing.Peats, peaty podsols, podsols and brown rankers.Mainly moorland types but also mountain grassland and peatland types.
Land class 24
Geography: Land form:
Topography: Landscape: Land use: Soils:Vegetation:
C. and W. Scotland.Glaciated valley sides often reaching from base to rocky summits sometimes with peak emergent from peneplains.Precipitous and extremely steep slopes with land at high altitude.Rugged mountain scenery often rocky with fast flowing streams.Limited open range grazing.Brown rankers, peats or peaty podsols and some peaty gleys.Mainly peatland types but also mountain grassland and moorland.
Land class 25
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
N.E England, S.E., C. and N.E. Scotland.Alluvial, flood plains and moraines of glacial origin.Virtually flat or gently rolling land mainly at low altitude.Intensively farmed lowland with fences and scattered farmhouses. Mainly barley but with much good grassland.Brown earths, gleys and gleyed brown earths.Restricted to a few grassland types.
Land class 26
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
N.E. England, C. and E. Scotland.Valley floors and coastal plains of glacial origin sometimes with emergent outcrops. Undulating or smooth slopes mainly at low altitude.Rather mixed lowland landscapes often affected by urban development.Mainly good grassland but also much barley and pasture.Brown earths and gleys.Limited but mainly moorland types where present.
Land class 27
Geography: Land form: Topography:
N. England, C. E. and N.E. Scotland.Varied but mainly valley floors and bluffs occasionally with ridges and scarps.Variable from mixtures of gentle and steep slopes to uniform moderate gradient mainly at
355
Landscape: Land use: Soils:Vegetation:
medium low or low altitude.Mainly well fenced lowlands, often mixed with woodland.Arable, particularly barley but also much pasture and good grassland. Brown earths and gleys.Restricted but some grassland and moorland types.
Land class 28
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
N. England, S. and N.E. Scotland.Heterogeneous from meandering topography to peneplains or alluvial plains. Mainly virtually flat but some gentle gradients at medium/low altitude. Heterogeneous from enclosed farmed landscapes to open moorland.Pasture or rough grazing predominate but some good grassland also. Variable but mainly brown earths, gleys or peats.Mainly peatland types where present but also grassland and moorland.
Land class 29
Geography: Land form: Topography:
Landscape: Land use: Soils:Vegetation:
W. Scotland.Indented coastlines with more cut platforms and raised beaches.Uneven topography, usually with easy slopes but some steeper areas at low or medium/low altitude.Complex scenery containing many contrasting elements.Mainly open range grazing but also some crofting.Mainly peats but also rankers and brown earths.Mainly peatland and moorland types but some bracken.
Land class 30
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
Extreme W. Scotland.Mainly peneplains with meandering streams, sometimes with low hills.Variable from complex to almost flat at medium/low extending to medium/high altitude. Open moorlands near to the sea with rocky outcrops and lochs.Open range grazing and crofting.Mainly peats with some peaty podsols.Mainly peatland with some moorland types.
Land class 31
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
N. Scotland and Isles.Drowned coastlines indented with some coastal plains backed by low hills.Mainly broad gentle curved outlines and some steeper areas mainly at low/medium altitude. Windswept exposed coasts with the enclosed land divided into small fields.Mainly rough grazing but with some good grassland and pasture with crofting.Brown earths, peats and some podsols.Mainly moorland but also some peatland grassland types.
Land class 32
Geography: Land form: Topography: Landscape: Land use: Soils:Vegetation:
N.W. Scotland and Isles.Peneplain surfaces or low ridges, sometimes coastal.Variable from complex to even rounded slopes mainly at medium/low altitude. Bleak moorlands often with scattered lochs and eroding peat hags.Mainly open range grazing but some pasture.Mainly peats but some rankers.Predominantly peatland types but also some moorland.
356
Appendix 4.5b : Hierarchical aggregations of ITE Land classification system (from Bunce et a/.,1982), aggregated using TWINSPAN (Hill, 1979)
LAND CLASS (DIVISION 1) DIVISION 2 DIVISION 3 DIVISION 4
2 A AARABLE
3 BB
4
9
11C
12
14 D
25 EC
26
1PASTORAL
5F D
6
7 G
8
10 H
13 1E
15
16 J
27
17 KMARGINAL UPLAND
18F
19 L
20
28 M G
31 N H
21 0UPLAND
221
23 P
24
290
30J
31
357
Appendix 4.6: Classes from the Land Cover Map of Great Britain. Correspondence between the 25 ’target’ cover types and 17 ’key’ cover types (from Fuller and Groom, 1993a)
LAND C (17 clas
Î0VER CATEGORY ,s system)
TARGEl(25 clas!
CLASSES 5 system)
1 Sea / Estuary 1 Sea / Estuary
2 Inland Water 2 Inland Water
3 Beach / Mudflat / Cliffs 3 Beach and Coastal Bare
4 Saltmarsh 4 Saltmarsh
5 Rough Pasture / Dune 5 Grass HeathGrass / Grass Moor
9 Moorland Grass
6 Pasture / Meadow / 6 Mown / Grazed TurfAmenity Grass
7 Meadow / Verge / Semi-natural
7 Marsh / Rough Grass 19 Ruderal Weed
23 Felled Forest
8 Rough / Marsh Grass
8 Grass Shrub Heath 25 Open Shrub Heath
10 Open Shrub Moor
9 Shrub Heath 13 Dense Shrub Heath
11 Dense Shrub Moor
10 Bracken 12 Bracken
11 Deciduous / Mixed Wood 14 Shrub / Orchard
15 Deciduous Woodland
12 Coniferous / Evergreen 16 Coniferous Woodlandwoodland
13 Bog (Herbaceous) 24 Lowland Bog
17 Upland Bog
14 Tilled (Arable crops) 18 Tilled Land
15 Suburban/Rural Devel. 20 Suburban / Rural Development
16 Urban Development 21 Continuous Urban
17 Inland Bare Ground 22 Inland Bare Ground
0 Unclassified
359
Appendix 4.7: Percentage of each land cover class at each site (25m resolution, 6 class aggregation)
Site LC1 LC2 LC3 LC4 LC5 LC6
mk01 18.85 72.96 2.47 2.09 3.55 0.08mk02 6.00 60.55 14.71 2.37 16.34 0.03mk03 18.63 71.14 2.99 1.20 5.92 0.13mk04 3.61 75.42 6.94 7.51 6.47 0.05mkOS 26.21 48.17 6.77 8.54 10.31 0.00mk06 2.55 61.57 0.31 15.70 11.48 8.39mk07 4.61 54.38 1.09 4.80 18.66 16.45mk08 2.69 89.70 0.35 1.48 5.45 0.34mk09 3.24 56.34 1.34 2.40 16.84 19.85mklO 6.91 55.06 0.08 3.69 32.58 1.67mk11 2.86 78.64 0.00 14.33 4.16 0.02mk12 9.17 63.31 0.77 0.31 20.95 5.49mk13 5.05 16.03 1.16 0.17 41.60 35.99mk14 7.50 51.34 1.85 1.92 32.10 5.30mk15 1.42 45.85 14.03 13.88 24.81 0.00mk16 0.00 0.00 0.00 0.00 50.28 49.72mk17 0.06 2.23 1.61 17.86 34.20 44.03mk18 0.03 1.54 0.25 0.00 47.28 50.89mk19 0.48 0.00 0.00 0.32 66.77 32.42mk20 0.00 0.00 5.24 0.00 54.04 40.72mk21 0.00 0.00 0.00 0.00 100.00 0.00mk22 12.79 0.00 0.42 0.00 63.46 23.34mk23 1.18 9.20 3.56 . 0.00 60.30 25.75mk24 3.90 0.59 0.68 0.00 51.36 43.46mk25 0.43 0.16 0.00 0.00 41.47 57.95mk26 4.31 62.38 4.83 7.63 20.81 0.04mk27 3.51 64.21 4.19 10.05 18.03 0.00mk28 1.02 41.42 5.13 5.67 34.37 12.39mk29 0.00 31.65 20.23 0.00 48.12 0.00mk30 6.58 50.51 12.10 0.76 19.91 10.14mk31 2.78 27.69 10.71 4.22 27.56 27.04mk32 2.71 18.53 5.94 0.00 33.41 39.42mk33 0.04 0.58 6.19 0.14 42.52 50.54mk34 1.55 0.42 5.97 35.00 27.26 29.80mk35 0.65 41.07 3.75 2.63 44.69 7.22mk36 0.28 1.82 1.16 4.36 45.62 46.77mk37 2.03 0.35 1.76 74.23 11.66 9.97
361
COü
in 0 0 in 1-N CD
COo
i - T - c O ' ^ c O ' ^ i n o o c O ' r - c o c o e o o c oc v j o o o T - o c o o h - ' ^ m N T f c j ) h > . L ni n r ^ n J ' ^ ’ c o T - ' c D C D c o r ^ r ^ c D r ^ c o o JC O C O i n C \ J C O C V J ( M C O C O C \ J C \ J C \ J ( M C D C D
oodo
O O C O C M C V J 'c f C D C O O J C O O ) C\J CJ) CO ^ C\JCO CM in CM 00 d CO
CD CM CM ^ CO T— CM 1— T-
in00 CO 00 CO CM h*' K CM h-
^ ' ! t i -C3)T--CONOOCJ)CD^ ^ t C D O O i n O C D ^ K - t ^d in T-' in CO cm* co cd cdC M C O C D C M i n C O C O C O C O C M
Ü
r^ ' »- inc j5CMCDr^Ti -cMcocDc: )T- r^ ' ^c j - (3)oo>t - c D c o c n c M p h ' . o q ' ^ L n c q o q c D ' c t o q c j j o o c D r ^0 0 * d T-‘ d 0 0 oc) in c \i N M cd in cm* M o ) CM cd in c o T f - T - c D i n c o i n c D i n c D C D i n i n c M C D C D c o c o
O 0 ) i n N C D T - C D i n C M T - 0 0 M 0 0 Op p c M p p p p c q c M p p i n i - ^ o q d c d c d c d i n o ) T - : c d c d o d ( d c M c d M i n c M c o L n c D i n i n T f i D i n i n T f N
i n C M ' ^ O C D T “ C0 CJ)0 )h - ' M ' C M O p r ^ p p pc dc M NC M dT tC M CM Tfi n C M C O ^ t C D C D C D C D C D
Ü
i n c D r ^ o c o i n ' ^ o i n o c o o c D c M O O i n c D O o o c D i — o o o c o o o o o c DC 3 ) 0 ) C 3 ) O i n T - C D O T - O O O C D O O O O O O O O O T t O O O r ^ l ^ O O O O C Dc d o d d d d d d d d d d d K c d d d d d d d d c d c d d c d ' M - d d d d c M T— CO . T- 1— CM CM
O O C O ^ O O O O O O0 0 0 ) 0 0 0 0 0 0 0 0O O O t- C M O O O O O
T- CO
Ü
OCM T“ CM Tf
' t - O i n O ) C D K i n C D T j - C J ) C J ) 0 ) C M O O C O O C D O i n T f C M C M ' ^ i n O O C O O O O C J ) C O O O C M C D O O ON O C O C D O O O T f N C O T - T f c O T f O O C M O ^ O O O O m C O C M O O O O O O O i - O C M O O O ON d d d T - ^ d d T - ^ i - ^ c d i - ^ i n d d d d d d d N c d d d d d d d d d d d d c M d c d d d d d
T- CM
Ü
CDcdCM
CO o i n CM 1- o i n 00 CO
o o o
CD h*. CO CO CMp P T-; i n CM
<D CD cS cS <D 'r
' ^ o o h - c D i n c o m O ' ^ 1— O O ' M - C D O O O C J ) i n d d c M c d c d d d r :
0 ) 0 0 O O O C 0 C D ( D 0 0 i - O O 0 ) N C D C D O C M ' ^ o mT - 0 ) O O O T - O C O T - C O O O C M O O T - r ^ O O O O ' ^c d d d d d d d d d d d d cm* t- ' cm* o* o* o* d d i n
Ü
C M i n C D C D N C O C O C DO C M C O O ) C M C M C M C D
COO
O O CO CM O 00 CM O 1- CM
P:CM cd d d
i n m 1- 00
i n CO m s CM O
TT OO CM O
CD N CD CM
CO CD i n O m Tf
CM* cd T-*
1— O 1— P P N00 cd 00
CO T-
CM CD N 'M- p p MCM CM CM
CD CM NM m O T-' i n cd
Lop p p pT— d ^ d
CD CM O 1 - N
1— 1— cd
OOC3)OCMCOinCDr^OOO)OT-CMCO’M- i nCDh^OOC3)0- r -CMCO'^LnCDN.OOCJ)OT-CMCO'^ inCDh'OOCDO
m E E E E E E E E E E E E Ê E E E E E E E E E E E E E E E E E E E E E E Ê Ê E E E E
Appendix 4.8: Percentage of each sensitivity class for geology in each catchment
site G1 G2 G3 G4
mk0 1 0 . 0 99.9 0 . 0 0 . 0
mk0 2 61.9 9.4 0 . 0 28.7mk03 19.3 0 . 0 0 . 0 80.7mk04 29.3 0 . 0 0 . 0 70.7mk05 14.2 0 . 0 0 . 0 85.8mk06 20.9 79.1 0 . 0 0 . 0
mk07 28.7 65.2 6 . 1 0 . 0
mkOB 10.3 42.0 47.7 0 . 0
mk09 1.4 8 6 . 0 1 2 . 6 0 . 0
mk1 0 0 . 0 1 0 0 . 0 0 . 0 0 . 0
mk1 1 0 . 0 0 . 0 1 0 0 . 0 0 . 0
mk1 2 5.7 0 . 0 94.3 0 . 0
mk13 59.8 15.5 24.7 0 . 0
mk14 28.6 4.6 6 6 . 8 0 . 0
mk15 6.3 0 . 0 93.7 0 . 0
mk16 50.7 49.3 0 . 0 0 . 0
mk17 32.2 67.8 0 . 0 0 . 0
mk18 62.7 37.3 0 . 0 0 . 0
mk19 70.7 29.3 0 . 0 0 . 0
mk2 0 98.8 1 . 2 0 . 0 0 . 0
mk2 1 56.4 43.6 0 . 0 0 . 0
mk2 2 53.2 42.9 0 . 0 4.0mk23 31.1 68.9 0 . 0 0 . 0
mk24 84.4 0 . 0 0 . 0 15.6mk25 75.6 5.1 ^ 0 . 0 19.3mk26 0 . 0 1 0 0 . 0 “ o.o 0 . 0
mk27 0 . 0 1 0 0 . 0 0 . 0 0 . 0
mk28 2.9 97.1 0 . 0 0 . 0
mk29 2 2 . 2 77.8 0 . 0 0 . 0
mk30 0 . 0 1 0 0 . 0 0 . 0 0 . 0
mk31 72.6 8.5 18.9 0 . 0
mk32 34.1 35.7 30.1 0 . 2
mk33 8 . 6 87.6 3.8 0 . 0
mk34 15.6 0 . 0 84.4 0 . 0
mk35 90.4 0 . 0 9.6 0 . 0
mk36 1 0 0 . 0 0 . 0 0 . 0 0 . 0
mk37 1 0 0 . 0 0 . 0 0 . 0 0 . 0
mk38 2 . 0 0 . 0 98.0 0 . 0
mk39 84.7 15.3 0 . 0 0 . 0
mk40 78.9 19.5 1 . 6 0 . 0
363
site G1 G2 G3 G4
mk42 99.4 0 . 6 0 . 0 0 . 0
mk43 89.8 1 0 . 2 0 . 0 0 . 0
mk45 92.9 7.1 0 . 0 0 . 0
mk46 97.3 0 . 0 2.7 0 . 0
mk47 54.3 0 . 0 0 . 0 45.7mk48 96.6 0 . 0 1 . 1 0 . 0
mk49 89.5 2 . 0 0 . 0 8.4mkSO 80.8 19.2 0 . 0 0 . 0
mk51 95.4 0 . 0 4.6 0 . 0
mk52 99.3 0 . 0 0.7 0 . 0
mk53 94.6 0 . 0 5.4 0 . 0
mk54 97.9 0 . 0 2 . 1 0 . 0
mk55 1 0 0 . 0 0 . 0 0 . 0 0 . 0
mk56 0 . 0 0 . 0 1 0 0 . 0 0 . 0
mk57 0 . 0 0 . 0 1 0 0 . 0 0 . 0
mk58 0 . 0 0 . 0 1 0 0 . 0 0 . 0
mk59 0 . 0 0 . 0 1 0 0 . 0 0 . 0
mk60 0 . 0 0 . 0 1 0 0 . 0 0 . 0
mk61 82.3 17.7 0 . 0 0 . 0
mk62 91.9 8 . 1 0 . 0 0 . 0
mk63 51.6 48.4 0 . 0 0 . 0
mk64 1 0 0 . 0 0 . 0 0 . 0 0 . 0
mk65 97.7 1 . 8 0.5 0 . 0
mk6 6 1 0 0 . 0 0 . 0 0 . 0 0 . 0
mk67 1 0 0 . 0 0 . 0 0 . 0 0 . 0
mk6 8 1 0 0 . 0 0 . 0 0 . 0 0 . 0
mk69 1 0 0 . 0 0 . 0 * 0 . 0 0 . 0
mk70 1 0 0 . 0 0 . 0 0 . 0 0 . 0
mk71 67.0 33.0 0 . 0 0 . 0
mk72 19.6 80.4 0 . 0 0 . 0
mk73 58.9 33.6 0 . 0 7.5mk74 57.6 13.6 5.7 23.1mk75 30.5 69.5 0 . 0 0 . 0
mk76 3.7 64.0 4.6 27.7mk77 99.4 0 . 0 0 . 6 0 . 0
mk78 1 0 0 . 0 0 . 0 0 . 0 0 . 0
mk79 98.7 1.3 0 . 0 0 . 0
mk80 1 0 0 . 0 0 . 0 0 . 0 0 . 0
See Table 4.7 for derivation of geology classes
364
Appendix 4.9: Percentage of each drift type in each catchmentsite D1 D2 D3 D4 D5 D6
mkOI 0.0 0.0 0.0 99.9 0.0 0.0mk02 30.2 0.0 0.9 46.1 14.0 0.0mk03 43.0 1.0 7.1 37.9 0.4 6.7mk04 70.9 0.0 2.1 19.7 0.0 0.0mk05 0.0 19.1 0.0 0.0 0.0 32.2mk06 18.9 0.0 0.0 0.0 0.0 0.0mk07 31.7 0.1 0.0 0.0 0.0 0.0mk08 65.3 0.0 0.0 0.0 0.0 0.0mk09 54.0 0.0 0.0 0.0 0.0 0.0mklO 35.6 0.0 0.0 0.0 0.0 0.0mk11 97.6 0.0 2.4 0.0 0.0 0.0mk12 80.6 5.2 0.9 0.0 0.0 0.0mk13 23.3 3.9 0.4 0.0 0.0 0.0mk14 70.8 1.8 11.7 0.0 0.0 0.0mk15 88.9 0.0 9.7 0.0 1.0 0.0mk21 40.7 0.0 0.0 0.0 7.1 0.0mk22 25.9 0.0 0.0 0.0 0.0 0.0mk23 16.5 0.0 0.0 0.0 0.0 0.0mk24 0.3 0.0 0.4 0.0 0.0 0.0mk25 51.8 0.0 0.1 0.0 0.0 0.0mk42 0.0 0.0 57.2 0.0 20.8 22.0mk43 0.0 0.0 57.7 0.0 15.9 7.3mk45 0.0 0.0 45.8 0.0 6.8 0.0mk46 0.0 0.0 19.9 0.0 0.0 0.0mk47 0.0 0.0 21.3 0.0 0.0 0.0mk48 0.0 2.7 12.0 0.0 0.0 0.0mk49 0.0 0.0 44.1 0.0 0.0 0.0mkSO 0.0 0.0 42.8 0.0 0.0 0.0mkSI 0.0 0.0 13.0 0.0 0.0 0.0mk52 0.0 3.4 3.1 2.4 2.7 0.0mk53 0.0 0.0 0.0 3.7 8.1 0.0mk54 0.0 5.4 33.6 0.0 0.0 0.0mk55 0.0 0.0 43.2 0.0 0.0 0.0mk56 0.0 1.1 4.8 0.0 71.9 0.0mk57 0.0 0.0 4.1 * 0.0 67.4 0.0mk58 0.0 0.0 0.0 0.0 0.9 0.0mk59 0.0 0.0 0.0 0.0 9.1 0.0mk60 0.0 0.0 0.0 0.0 0.0 0.0mk61 0.0 0.0 39.7 0.0 0.0 0.0mk62 0.0 0.4 24.2 0.0 28.3 0.0mk63 0.0 0.9 46.1 0.0 0.1 0.0mk64 0.0 0.0 47.0 0.0 0.0 0.0mk65 0.0 0.5 60.3 0.0 0.3 0.0mk76 0.0 0.0 0.0 0.0 17.4 0.0mk77 72.4 0.0 19.9 0.0 0.5 0.0mk78 58.2 0.0 0.2 0.5 22.2 0.0mk79 54.7 0.0 0.0 1.7 21.8 0.0mk80 48.4 0.0 0.0 5.9 0.0 0.0
D1: Boulder clay/Till D2: Alluvium D3; Moraines
D4: Raised Marine Deposits D5: Peat D6 : Sand and gravel
365
üCOoœcvjcocoocopococj îcqi^oqocooscDLncJCJïT-incqpT-^T-icqpr^’ T-icqcDcDiiScôcoaiodoôcbœœcdcocDcocDT-^côi^iocôcocôcvicvir^cocôoïr^cô' it '^coô
p N p p p i - ; p p p p ^ M . p T i p p p T f o q p p p ' < ) ; T - % p p M p p M p p p c \ jcDc\ia>o6 i --^cor^cdooc>c\jcoocdcb'^a)iri t^i--co'^Lrir^cDc:jr^cdoa>r^c\iLr>Lrio r CDCOCOCOCOCOCDlO'iï-inCJCMCVJCOCO Tf c\j CVJ T- (N
cro+X
p p p p r ^ p p p i ^ p ' ^ p ' ^ p p i ^ i v . p o î o p p p p p h v . p p p o q p ' ^ r ^ pcO'^'^lf)COO)I^COinCDCOCOhN!c\iC3îO>LOLOCvjcD't--LO' '^'^'^OCjcOCOC)OrJ-Lr>T-'
1— 1— 1— 1— C\J T- C\J C\J CM T— C\l CVj T— T— T- T— CM T—
Î3üco
o o o o o o o o o o o od d c i i c D C D d d d d d o ô
O O O O O O O O C D O C O C M C M O O O O O O O O Od d d d c M d c M - ^ o o M o ô d d d d d d d d d d
T— t— co
O
ÇO03>OCO
S Ic3CD
ÜCO
Î3üco
dco
p p p p p p p p p p p p ^ t p l ^ p p ^ t p p p p p p p p p p p ' ^ p h ' . r f f ^d i r jT^dddddddcMcôt^cMLr i i r i dr^cdc j îoc)CDLOT-^côddddLnodr^Ti *d T- cMoor^r^'^cMinLococomcoN-c3>co cMcoh-r^cor-
«^ p p p p p p p p p p p ^ p p M p p M p p p p p p N p T f T f C M p p p C M CMdodiocddcoddcdoôcoi^r^dcMCMCM'^LricDioi—d d d o î c v i d ' ^ d c M L o d LO' ^co r^oooo î Ooor ^T- t- ■M’ uo cm co cm oiOJor^-r-cMCMt-co
C O O O î O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O Od i r i r ^ d d d d d d d d d d d d d d d d d d d d d d d d d d d d d d dCM
coCD_3CO >
•4— >
CCD£
IüC\JCDCOCO
S ICL
CD
XycCDQ.CL<
üCO
2ceco
co
co
Xco
OCN’ C35C0O'^OOd CM _CM CO T- CM
- 0 0 0 3 C 3 > C O L n C M C M O O O C O O O O C O T l - C O O O C M O O C M nr r->inr>rr> — nn— rpir^(—>0 0 0 0 0 0 0 r^ 0 0 T-0 0 '^
P p p T - p p p p p p p p p p p p p T - p p p p
T^cd'<e-CMC7>ddddddd'«e-d'M-coiricc5cj5CMddTf CMC M O O O O O O ' ^ O O C O Oc d d d d d d d d d d d d
c M p p p p ' ^ t p p p i - ^ p p p r ^ p h ^ p p p p p p p p poddoôdcdcMMlr^dcdcdcdcdidi—i - i - ^ ^ d d d d d d d T - ^ d d o O f - c o o ^ l o r ^ o o o î L O T — T- CM 1 1— 1— CM "M" co T-
■^ CO( 3>OOCMCMOCM d id d d
"M" CM
h» ■4— lO d cd cd
1-
- p T ^ r s . T f p p p p p p p ' ^ p p p p ^ p p p p i - T - p T f c M O ) i - % p N -'M'cdcMCMcdcdcMCMcdT^dcdcM'^cddcdddddcdcMcdcd'^dcd'sj- '^
COU5h~C3îr^C3>CM f-CO t-CM lOLOCMh-T-CM CM
0)
p ' « d ; p p p p ' ^ p œ p p t ^ _ p p p p p p p T ^ p t ^ p p p p p p p ^ p r « - C M C Md i d d d d d d d c v i d d ’i—cdcM'^ indcdcdi^cdcdcdddi r i r ^ ' ^d idcdh^Lnt - CM1— œr^r^'^cMmcoco<3îc:îC35C3)00 cMincMcot^t^cejco
?-CMCOTfincONOOC:)OT-CMCO^mCOM.OOCRC3T-CMCO^lOCONOOC3)O^CMCO'(f
E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E E
366
oCO
T - ; T - i r > i n p c D p p i n p c \ j c \ J c q c : ) p p p p p o > p c : ) p p o r ^ p p c v i o > p p p L n r ' - p po c o ' ^ r ^ c o c o c o c o c o c o i r l c o c D c o c o c o c o c o c o c ô o ô c ô o d o ô c ô c o c ô c o c ô c N i c o c ô c ô c J L O o d L O
co ^ f C V J p p p r ^ p C V l N p p r t C M O p - ' - p p p p p p p ' ^CD Cvi'^C0h^COI^CT>t^cÔ'^cd'^Cûr^*<1’ LO'^^J-C0O5LOi—;nO co t— T— T— 1— T— O J t— C \ | l — C \ I C \ j T - ? — 1— r ? — C \jT— T - T — T—
p N p p p p T f T j | p ( S . p p [ S .
l O T t c Ô ' ^ ^ ' < t C D ' ^ T ^ C \ i c J > l f ) C OCVJ T - ^ T - CVJ co co
CT03+X
P p p t ^ P P P p P p T - ^ p t ^ P P P P p P P P P P P p P p p p P p p p - ^ t - l - ^ T - ;LOCT)cvic>oc)cy>'^cdcocj)r'Icot^t^o>o>cT>a>c33cdcviinT^T^ir>i^cdcor^o>cbc3>T-^cocvjf^co
if]üco
O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O t- O O O ...............................CDOC>C3C><3joC3C3C3C>C3CD0C5O0(I>OC::>C3OC:3Cr>C0C3O(DCDC:3C:3C>o o o o o
oco
O h » C V j C 0 O O C 0 O 0 3 C M C 0 O O O O O C J 3 O O O O O O I ^ l 0 O C \ J O O O O ( 3 3 C M O C 3 3cjcôajr^cDcicdcJt^ojr^cDcjcjocJoôciJci icâcDcjcDr^cocDci ir^cDÔcDcococDciir ^ i — CJ3 0 > O O C 3> O L 0 0 3 C O O O O O O C 3 ) 0 0 3 0 3 0 0 3 0 3 0 0 0 0 0 ' ^ CO
f3üco
c \ j L n o i r 3 C D O O o o o o o o o o o o o o c v j o o 3 0 0 o o o o o o o o o o r ^ O T - i-^T-^oincDcocôcocoocococoocoocôcococdcdczncococoocôcococococoocO'r-^ocn 03 co 00 03 o (33 o o L O O C O
füüco
O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O OC O c d Ô C D C O C D C O C Ô C O c d c O C O C D C D C D C Ô c J c O C O C D Ô C D d C D C Ô C D C O C Ô C D C O C O C O C D O C D C D C O
oco
O O O O O O O O O t - O O O C O O O C D O O O O O O O O GOCDCOCOCOCÔCOCi'r- COCvicOCvicDCDCDCDCDCDCOCÔCOCOCOCDCvicodcicÔCO CD d cS d CD c5G O O O O O O O O O > ' O t - 0 0 ' ^
2coco
o o o o CM CVJ d d d d d CM
o o o o o o o o o o o o d d d d d d d d d d d d
p p p p p p pd d d d d d d
LOO ' T - O O O O O O O Od d d d d d d t o d d
coO O O O O O O O O N O - r - C O T j - O O O O O O C O O i - O O O C O O O C O O O O O O O O O
d i - ^ dc v i dc v i ddd dddc oddd ddc v i dddd dddddd d"M" T-03 o o o o o o
CO r-( 3 3 L O C M i n ( O O O O O O O O O N O O O O O C 0 0 3 0 C O O O O O O O O O O C O O O T j ' ( 0
" d d d d d d d d d d d i o d ' M - d d d L o i o d c d d d c p d d d d d d d d d d ^N 00 c o CM T - l ô
Xco
p p p J ^ p p p p p p p C V j p p p p p p p p T - i p M p p M p p p p p p C M O O C O T f03oôo3côo3 j ^ddoôd i '~Zo) |s l Tfd (dddo6d 'M'03côddT-^c6do3dddo3ddo3 i f jT - C 3 3 C M T - 0 3 0 3 0 0 0 3 0 L f ) 0 3 0 0 0 0 0 0 3 0 0 0 3 C 0 0 3 C 3 3 0 3 0 0 N ( 3 3 0 0 3 0 0 0 0 3 0 0 3 0 0 ^
T** T— 1— T— T— T— T— T— T— T— <«—
0)lf)(0M 00 03OCMC0L0(0N 00 03OT-CMC0^lf3C0M 00 03OT-CMC0Tj-if)C0N 00 03O'T-CMC0 COCOCOCOCO'M-TfTf^'ifTj-'^^lOlOlOLOlOlOlOlOlOlOCOCOCOCOCOCOCOCOCOCOh.h.h.ME E E E E E E E E E E E E E E E E E E E E E E E E S E E E E E E E E E E E
367
wo \00
site SH SM SL Bare SCL1 SCL2 SCL3 SCL4 SCL5 H+eq %BS SCL
mk74 26.2 73.8 0 . 0 0 . 0 0 . 0 0 . 0 73.9 26.1 0 . 0 9.3 45.0 6.7mk75 1 0 . 1 89.9 0 . 0 0 . 0 0 . 0 0 . 0 86.3 13.7 0 . 0 7.2 43.0 7.3mk76 73,9 26.1 0 . 0 0 . 0 0 . 0 0 . 0 16.8 74.3 9.0 15.0 23.8 3.6mk77 90.9 9.1 0 . 0 0 . 0 0 . 0 0 . 0 0 . 0 0 . 8 99.2 15.4 10.9 0 . 0
mk78 92.1 7.9 0 . 0 0 . 0 0 . 0 0 . 0 0 . 0 0 . 0 1 0 0 . 0 1 2 . 0 13.3 0 . 0
mk79 1 0 0 . 0 0 . 0 0 . 0 0 . 0 0 , 0 0 . 0 0 . 0 28.0 72.0 9.6 17.2 0 . 8
mk80 95.9 4.1 0 . 0 0 . 0 0 . 0 0 . 0 4.1 46.0 49.9 1 0 . 6 19.1 1.7
See Section 4.6.6 for details relating to the derivation of these variables See Table 6.3 for units of measurement"NB; Only variables relating to data at soil map unit scale shown here (see Appendix 6.3b)
Appendix 4.11: Miscellaneous data for Phase 2 catchments
site Area SJen Dist Aits Aitm Sdep NDep Rain
mkOI 1.55 2.35 3.21 7 30 0.613 0.697 983mk0 2 5.66 7.71 5.15 1 0 57 0.613 0.697 983mk03 8 . 8 8 30.61 5.84 1 0 105 0.613 0.697 983mk04 3.91 9.21 7.17 1 0 117 0.613 0.697 983mk05 1.70 3.18 10.50 25 81 0.613 0.697 983mk06 3.28 3.81 19.28 140 350 0.647 0.77 1032mk07 1 0 . 2 1 19.31 24.32 60 445 0.647 0.77 1032mk08 2 . 6 8 5.82 22.14 55 275 0.647 0.77 1032mk09 1.99 4.62 2 1 . 2 2 50 295 0.647 0.77 1032mklO 1.45 1.27 18.93 1 0 0 313 0.647 0.77 1032m kll 0.47 1.54 20.93 40 90 0.468 0.593 782mk1 2 3.83 6.50 24.66 70 290 0.468 0.593 782mk13 6.75 13.47 25.92 90 403 0.468 0.593 782mk14 4.71 7.67 25.94 75 403 0.468 0.593 782mk15 5.38 9.55 23.95 60 141 0.468 0.593 782mk16 3.44 1 1 . 1 2 38.95 170 669 0.912 1.408 1854mk17 3.07 8.77 40.60 80 640 0.912 1.408 18ë4mk18 5.76 13.61 43.34 190 825 0.912 1.408 1854mk19 1.16 3.72 46.05 2 2 0 625 0.912 1.408 1854mk2 0 1.85 4.54 46.87 450 870 0.912 1.408 1854mk2 1 11.14 17.36 45.90 410 1068 0.715 1.113 1168mk2 2 5.66 16.00 48.86 470 1068 0.715 1.113 1168mk23 5.27 6.60 45.61 150 864 0.715 1.113 1168mk24 5.60 7.55 49.34 400 975 0.715 1.113 1168mk25 2 . 8 6 3.76 51.06 450 933 0.715 1.113 1168mk26 5.71 6.82 37.41 190 359 0.473 0.786 709mk27 3.22 3.11 41.31 2 1 0 448 0.473 0.786 709mk28 6.91 7.31 40.26 170 490 0.473 0.786: 709mk29 1 . 1 0 1.30 39.20 156 318 0.473 0.786 709mk30 1.63 2 . 1 1 42.87 170 ' 470 0.473 0.786 709mk31 4.19 5.15 46.88 2 1 0 570 0.496 0.819 782mk32 6.17 9.33 44.62 2 2 0 670 0.496 0.819 782mk33 3.27 3.09 39.76 370 721 0.496 0.819 782mk34 1 . 8 8 3.08 37.47 345 521 0.496 0.819 782mk35 4.14 8.18 40.47 2 2 0 439 0.496 0.819 782mk36 3.54 5.27 18.12 175 356 0.383 0.532 764mk37 1.95 2.90 15.71 60 255 0.383 0.532 764mk38 5.72 11.53 9.43 95 339 0.383 0.532 764mk39 5.81 9.95 15.30 155 356 0.383 0.532 764mk40 3.09 5.50 14.69 145 326 0.383 0.532 764mk42 3.00 3.97 27.46 251 838 0.377 0.649 1425mk43 10.52 10.51 30.43 213 693 0.377 0.649 1425mk45 4.97 7.22 33.65 189 617 0.377 0.649 1425mk46 2.23 5.14 3.57 145 896 0.78 0.966 2929mk47 1.67 3.70 4.14 46 597 0.78 0.966 2929mk48 5.13 19.71 3.01 15 510 0.78 0.966 2929mk49 4.64 14.22 0.81 30 549 0.78 0.966 2929mkSO 3.72 11.40 3.13 61 488 0.78 0.966 2929
369
site Area SJen Dist Alts Altm Sdep NDep Rain
mk51 1.13 3.21 0.05 1 0 230 0.644 0.785 2419mk52 5.17 14.75 0.50 2 0 590 0.644 0.785 2419mk53 2.62 5.89 0 . 2 0 2 0 2 0 0 0.644 0.785 2419mk54 4.34 10.53 4.20 180 680 0.644 0.785 2419mk55 1.23 3.19 0 . 1 0 2 0 730 0.644 0.785 2419mk56 6.34 13.70 3.98 74 551 0.896 1.131 3039mk57 4.11 8.32 4.21 84 613 0.896 1.131 3039mk58 1.59 3.00 5.11 63 466 0.896 1.131 3039mk59 3.50 8.33 1 . 2 2 95 546 0.896 1.131 3039mkSO 1.76 5.54 0.63 126 557 0.896 1.131 3039mk61 0.87 4.02 24.02 1 0 381 1.105 1.416 3319mk62 2.73 8.74 19.32 1 0 635 1.105 1.416 3319mk63 2.36 6.74 2 0 . 0 0 30 6 6 8 1.105 1.416 3319mk64 2.48 10.44 26.91 1 1 0 620 1.105 1.416 3319mk65 19.67 84.80 24.67 1 0 635 1.105 1.416 3319mkSB 9.37 21.13 76.05 251 1070 0.635 0.909 1948mk67 5.75 17.16 75.86 251 1070 0.635 0.909 1948mk6 8 3.03 4.89 71.22 274 1070 0.635 0.909 1948mkS9 9.17 17.88 69.19 251 1130 0.635 0.909 1948 fmk70 0.62 1.41 68.77 274 488 0.635 0.909 1948mk71 1.59 7.82 23.97 50 897 1.242 1.579 3015mk72 1.97 5.52 16.60 60 250 1.242 1.579 3015mk73 3.53 9.67 23.39 70 250 1.242 1.579 3015mk74 2 . 1 1 4.89 22.07 70 258 1.242 1.579 3015mk75 1.40 4.23 20.38 50 172 1.242 1.579 3015mk76 1.41 3.15 3.83 1 0 0 267 0.783 1.154 1703mk77 5.65 13.00 0 . 0 2 30 536 0.783 1.154 1703mk78 5.85 14.07 0 . 0 2 2 0 439 0.783 1.154 1703mk79 2.73 6.53 0 . 1 0 1 0 187 0.783 1.154 1703mk80 1 . 2 2 3.74 0 . 1 0 5 2 0 0 0.783 1.154 :1703
See Table 6.3 for full description and units of measurement for these variables
370
Appendix 5.1 : Summary statistics for transformed chemistry/response variables* (c= constant) /7=954
Mean SD Min. Max. c
pH 6.53 1.10 3.82 9.21Aik 2.66 0.41 0.00 3.81 211Cond 1.99 0.37 1.04 3.08 *Na* 2.52 0.32 1.63 3.67 *K* 1.30 0.45 0.00 2.70 1Mg'* 2.13 0.45 0.95 3.61 *
Ca'* 2.39 0.64 1.08 3.91 *
Cl 2.60 0.35 1.58 3.75 *
NO,- 0.76 0.82 0.00 3.18 1SO,'- 2.17 0.46 1.15 3.78 *TOC 0.54 0.38 -1.00 1.59 *A LNL 0.95 0.53 0.00 2.35 1A LL 0.72 0.63 0.00 2.77 1HCL 0.54 0.36 0.00 1.54 1DCL 0.60 0.49 0.00 1.94 1
* Determinands which were transformed without a constant.
371
Appendix 5.2 : Summary statistics for transformed catchment/explanatory variables*(c = constant) n=954
Mean S.D Min. Max c
Sdep 0.91 0.22 0.49 1.62 1Ndep 1.05 0.20 0.58 1.66 1Alt 15.00 5.95 0.00 35.35 naRain 3.10 0.20 2.67 3.57 *
Dist 4.10 2.02 0.39 8.83 1
* No constant required
372
Appendix 5.3: Exploratory data analyses of the sensitive sub-set (Ca <200p.eq I )
A5.3.1 Response data
A5.3.1.1. Summary statistics
Summary statistics for the water chemistry data are presented in Table 5.3.1. It is clear that,
whereas with the 954 dataset the calcium gradient is the longest, here it has a range of only
188peq 1' while the DCL range is reduced from 85.29 to 2.43keq H"" ha* y r \ Conversely
Na+ (1484peq l''\ Mg (521 peq M), Cl' (1760peq I''), SO g. (766peq I ') and AI-NL (594peq
r ) all vary along substantially wider gradients. The mean and standard deviations are
reduced for all determinands except the aluminium species which are negatively correlated
with Ca^ in the 954 dataset and TOC. This will inevitably affect the relationships both within
the chemistry dataset and between the chemistry and the catchment attributes. The main
axes of variation are unlikely to be correlated to any great degree with Ca or DCL which is
likely to have significant repercussions in terms of the explanatory power of the RDA model.
The 954 dataset contains most sites in LVb classes 2, 5 and 6 (ca. 32, 19 and 37%
respectively) whereas the 469 dataset is dominated by class 6 , the upland moor and
grassland (ca. 60%). With regard to SCL, the full dataset is characterised by a fairly even
spread across the sensitivity classes (with the exception of SCL4 at 49.1%). However,
85.2% of sensitive sites are in either SCL4 or SCL5 reflecting the correspondence between
SCL class and freshwater sensitivity.
373
Table A5.3.1 : Summary statistics for untransformed chemistry - sensitive sites
Mean SD Min. Max.
pH 5.76 0.87 3.82 7.30Aik 28.45 73.02 -210.00 556.00Cond 66.82 40.90 11.00 247.00Na+ 328.54 261.80 43.00 1527.00K+ 12.95 11.69 1.00 167.00
82.36 66.23 9.00 520.00Ca"* 85.21 47.41 12.00 200.00CI 385.42 294.67 38.00 1798.00N Q / 5.11 12.28 B.D 134.00SO / 92.48 67.09 14.00 780.00TOC 3.90 3.32 0.10 22.40AI-NL 21.80 25.43 B.D 201.00Al-L 27.85 73.43 B.D 594.00HCL 1.01 0.98 0.00 6.75DCL 0.69 0.49 0.00 2.43
BD = Below detection limit, SD = Standard deviation. Min. = Minimum, Max. = Maximum
A5.3.1.2 Correlation structure
Correlation coefficients were calculated for the transformed chemistry determinands and
these are shown in Table A5.3.2 Those coefficients significant at the 1% level are
highlighted. There are significant differences between this matrix and the one produced for
the larger dataset. A substantial increase in the number of coefficients that are not significant
following the removal of sites towards the positive extreme of the ionic gradient is evident.
Correlations between Ca "" and the other determinands are much reduced as a result of a
more severe gradient contraction in the former instance.
A5.3.1.3 Principal components analysis on more sensitive sites
Principal components analysis (RCA) was undertaken on the sensitive (469 site) dataset.
The results are summarised in Table A5.3.3. As with the previous analysis the two derived
374
a
Cond -0.190
Aik 0.738 -0.144
Na" -0.066 0.915 -0.003
K" -0.109 0.708 -0.030 0.647
Mg:" 0.019 0.846 0.067 0.825 0.633
Ca:" 0.555 0.331 0.415 0.265 0.249 0.409
Cl -0.141 0.915 -0.040 0.971 0.643 0.805 0.247
NO; -0.075 0.066 -0.099 -0.061 0.049 0.191 0.081 -0.039
S O / -0.272 0.706 -0.252 0.510 0.527 0.649 0.385 0.509 0.355
TOC -0.316 0.286 -0.197 0.177 0.248 0.141 0.165 0.168 0.280 0.166
AI-NL -0.585 0.140 -0.412 0.071 0.147 0.015 -0.172 0.091 -0.059 0.174 0.629
Al-L -0.684 0.133 -0.507 -0.029 0.108 0.022 -0.261 -0.014 0.198 0.392 0.135 -0.385
HCL 0.632 -0.100 0.520 -0.058 -0.067 0.119 0.578 -0.141 0.013 -0.125 -0.038 -0.323 -0.385
DCL 0.721 0.030 0.579 0.033 0.004 0.141 0.860 0.002 -0.039 -0.020 0.111 -0.288 -0.438 0.742
pH Cond Aik Na" K" Mg:" Ca:" Cl NO; SO/ TOC AI-NL Al-L HCL
Table A5.3.4: Matrix of Pearson product-moment correlations matrix for 14 transformed water chemistry from 469 sensitive sites (Ca<200iieq/1*)
critical load variables were made passive. The differences between this and the PCA carried
out using 954 sites are substantial. The first two eigenvalues now account for only 42% of
the variation in water chemistry compared with 70.6% for the full dataset suggesting that
other directions of variation are relatively more important. The first PCA axis is most
associated with Na"", Cl', IVIg ", and conductivity, indicating a sea-salt gradient. The second
PCA axis appears to represent an acidity gradient with alkalinity and the aluminium species
exhibiting the highest scores. PCA Axis 3, accounting for 6.5% of the variation, has high
negative scores for pH and alkalinity with high positive scores for TOC and non-labile
aluminium. This also suggests an acidity gradient but one which is driven by organic acids.
NOg' and TOC have high positive and negative scores, respectively along Axis 4 which
explains 7% of the variation in the data. Table A5.3.3 shows that the eigenvalues for Axis
4 is greater than that for Axis 3. Usually eigenvalues are in order of decreasing value. This
is a result of a failure of the iterative ordination algorith within CANOCO to extract the
ordination axes. However, when eigenvalues are close the results of the PCA can still be
trusted (ter Braak, 1990).
The results of the PCA are presented graphically in Figure A5.3.1, a PCA correlation biplot
of the water chemistry determinands plotted against the first two PCA axes. This illustrates
the orthogonal sea-salt and sensitivity gradients The much reduced variance exhibited by
Ca^ , DCL and HCL is shown by their short vectors. The relationship between DCL and Ca ""
is weaker compared with the analysis on the full dataset. This may be an artefact of the DCL
calculation method (see chapter 3). DCL is calculated by dividing calcium zero (Ca^) by 94,
the critical acidification ratio. Ca^ results from subtracting the change in S O /' multiplied by
Henriksen’s F factor for Ca . The F factor is not applied at sites where Ca^ is greater than
400peq 1' (C.Curtis, pers comm). At these sites Ca "" is equal to Ca^ and therefore perfectly
correlated with DCL. For the sensitive sites the relationship between Ca "" and DCL is
dependent on SO/'.
376
Table A5.3.3 : Results of PCA on transformed water chemistry determinands 469 sensitive sites
r C A Axes
P2ij;envalue .2539 1683 .0650 .0739C iiiu % variance\'a r ia b lc loadings (correlations)
25.4 42.2 48.7 56.1
pH .3092 -.3205 - 5733 -.2515Aik .0749 -.5064 -.5340 -.12.53Cond .5535 ..3456 .2101 .0572Na* .7036 .3347 1760 .0387K* 4792 .3249 .1980 .0669
.6068 ..3917 .0493 1653Ca-* .4811 -.1580 -.2972 -.2548Cl .6646 ..3075 1759 .0465N O , -.0816 -.0463 -.4487 .6993S O / .4024 .2798 .0162 .3753TO C ,1321 .4243 .7472 -.5973A i-N L -.1938 .4912 .7799 -.2851A i-L -.3286 .4691 .3960 .4014
H C L (passive) .1809 -.1891 -J 3 6 8 -.24.3t)D C L (passive) 3134 -.2708 -3 4 0 0 -.4039
Figure A.5.3.1 PCA bi-plot of 15 water chemistry determinands from 469 sensitive sites
0.60
A I-N LA l-L
0 45 — TOCMg
Cond NoS 0 4
0.15 —
0.00 —
N 03
•0.15 — Co
HCL
•0.30 — OCLpH
•0.45 —
Aik
•060
■045 •0.30 •0.15 000 0 15 0.60 0.75
P C A A xis 1 = 0 .2 5 4
377
A5.3.2 Exploratory data analysis - explanatory variables
A5.5.2.1 Summary statistics
Catchment attribute summary statistics are presented for the subset of 469 sites (Tabie
A5.3.4). Prior to analysis these underwent the same transformations as the 954 dataset. The
number of sites in each of the classification datasets is presented.
Table A5.3.4: Summary statistics for untransformed catchment attributes - sensitive sites
Mean S.D Min. Max.
Sdep 0.85 0.44 0.24 2.47Ndep 1.17 0.48 0.34 2.75Alts 342.56 203.04 10.00 1000.00Rain 1796.22 688.95 598.00 3749.00Dist 17.57 16.19 0.15 66.05
Sdep and Ndep mean values (0.85 and 1.17keq H'" ha' yr respectively) are much the
same as the full dataset. This is not surprising as sensitivity is independent of deposition and
the removal of non sensitive sites should have little effect on the deposition gradient. The
increase in the means for rainfall (1796 from 1414mm yr^) and altitude (343 from 259m) is
intuitive following the negative correlations between these two variables and the ionic
concentrations. Upland sites will tend to be characterised by base poor surface waters and
higher rainfall regimes.
A comparison between the number of sites in each of the LVb classes for the full (954) and
sensitive (469) datasets is provided by Table 5.3.5a. Table 5.3.5b shows how the two
datasets differ with regard to SOL classes.
378
Table A5.3.5: Number of sites in each nominal class for the 954 and 469 datasets a) LVb classification
954 dataset 469 datasetClass No. of sites % No. of sites %
1 21 2.2 12 2.72 306 32.1 38 8.13 40 4.2 11 2.44 47 4.9 19 4.15 185 19.4 111 23.66 355 37.2 277 59.1
b) SCL classification
954 dataset 469 datasetClass No. of sites % No. of sites %
1 173 18.1 9 1.92 109 11.4 30 6.33 119 12.5 30 6.34 462 49.1 326 69.55 91 9.6 74 15.7
A5.3.2.2 Correlation
Table A5.3.6 is a correlation matrix showing the relationship between the transformed
continuous independent variables. These are largely similar to those from the full dataset.
However, altitude and rainfall exhibit a much poorer correlation. The altitude range has
remained the same but the rainfall has shifted towards the wetter end of the gradient.
379
Table A5.3.6: Matrix of Pearson product-moment correlations matrix for transformed catchment attributes for 469 sensitive CLAG sites
Sdep 0.289
Ndep 0.315 0.943
Rain 0.119 0.272 0.231
Dist 0.573 0.318 0.374 -0.346
Alts Sdep Ndep Rain
All correlations have P values <= 0.01
380
Appendix 6.1: Pearson product-moment correlations for 16 transformed water chemistry determinands (full dataset, n = 78).
A correlation matrix (Table A6.1) quantifies the individual relationships between the
transformed water chemistry determinands together with the derived critical loads values.
The significance level of each correlation co-efficient was established by determining the p-
value. A p-value greater than 0.05 indicates that a correlation is not significant at the 5%
level. Correlation coefficients which are statistically significant at this level are shaded.
Although a 1 % significance level was selected for the Phase 1 dataset the 5% level selected
here reflects the much reduced number of observations. There are high positive correlations
between the base cations and Cl'. pH is highly correlated with alkalinity, Ca "" and the critical
load measures. These in turn exhibit strong relationships with conductivity, Mg "" and Ca "".
The aluminium species are negatively correlated with all other determinands, with particularly
high negative coefficients between non-labile aluminium and pH, alkalinity and diatom critical
load. Nitrate shows higher correlations with a number of determinands in comparison with
the Phase 1 analysis, particularly with K , IVIg , Ca '' S O /, conductivity and diatom critical
load. This may be a consequence of the higher proportion of less sensitive sites in the
calibration dataset. The CLAG programme targeted the 'most sensitive site’ for sampling
(Kreiser et a i, 1993). Absorbance at 250nm which can be considered as a surrogate for
organic acidity is highly correlated with both total monomeric and non-labile aluminium.
381
Aik 0.935
Cond 0.588 0,755
Na* V 0.250 0:441 , 0.830
K* 0.280 0.480 , 0.800; 0,709
Mg^ 0.606 .. .0.75S ; 0.939 .. 0,765 . V b .7 2 7
Ca^ CVO.773 0:893 ^ 0.^27 .0.611' ' 0.714 X 1 0.850.
Cl 0.M0 0.970 {: :'Ô.7& . 0.736;, ' 0.600.
NO,- : 0.492 0.633 " 0.786: 4 = m i ! < 0.807 . 0.544'
SO4*- ; 0.599 X': 0.894 \ 0:709 0.785 . . : 0.619 0.837;;
AI-TM
AI-NL
Al-L - a ™
-0.108
-0.147
0.064
-0.286'
- 4.344
-0.032
-0.448
-0.079
' .-0.575
.< 4 .6 7 3
-0.198
-0.098
-0.139
0.066
.4.627
4.675
> 4.314
,4 .3 5 2 :
% o . ; i6 i
-0.031 < {0 .752 . "^'0^59?
ABS-250
HCL
0.251
0.559
-0.065
X- 0:422,
-0.136
' 0.788
: 4.324 '
0,841
,0.262
‘ , 0.497
4.356
0.539
-0.100
0.439
0.782
-0.518
0.782:
4 # r
-^:0.566:
-0.185
DCL ' 0 ,3 « .X :0 .m ';:o ,5^o ' '0.785 $ 0 . 7 & ' ;..:^4.703^ -0.240 :0.89Qi
pH Aik Cond Na" K" Mg'* Ca'* Cl NO,- s o / AI-TM AI-NL Al-L ABS-250 HCL
Table A6.1: Matrix of Pearson product-moment correlations for 16 transformed water chemistry determinands (full dataset, n = 78).
Appendix 6.2: Comparative analysis of alternative sub-sets of the catchment data
The aim of this section is to select sub-sets of the catchment data for inclusion in the Phase
2 analysis. An attempt is made to quantify the relationships between these datasets and the
catchment data in order to evaluate which should be used within the predictive model
exercise.
During the early stages of the Phase 1 analysis a preliminary investigation was undertaken
to examine the relative importance of the classification system imposed on a type of
variable. For example, the land cover data is available at 1km and at 25nf resolution. In
addition to the land cover data, a land classification dataset could also be used. Both these
datasets can be aggregated at a number of hierarchical levels. A number of RDA’s were
produced looking at datasets separately in order to facilitate the decision as to which
classification system (for those based on similar attributes) and at what aggregation, should
be input into the multivariate analysis. As with the Phase 1 analysis the objective here is to
reduce the number of explanatory variables to limit spurious explanation.
The classifications and aggregations relating to soil and land use data are discussed in
Chapter 4. The analyses presented here relates specifically to land cover resolution and
aggregation, land classification aggregation, the scale at which soil is mapped and
superficial drift deposits. The following analyses are undertaken.
1. An assessment of the land use variables and selection of the most suitable for
inclusion in subsequent analysis.
2. A comparison of the explanatory powers of the catchment attributes where the soil
data is mapped at association scale and at soil map unit scale.
383
3. An examination as to whether the inclusion of drift data has any bearing on this
predictive power.
A6.2a Redundancy Analyses (PDA) on land cover and land classification variables
A number of separate PDA analyses are undertaken using the water chemistry
determinands as the response variables and the various land use classifications described
in Chapter 4 as the explanatory data. Each variable is expressed as the percentage of the
land area of the catchment covered by that variable. The results of these analyses are
displayed in Table A6.2.1.
The sum of canonical eigenvalues (and thus the level of explanation offered by the
catchment variables) for these analyses ranges from 0.302 for the land classification
aggregated to four classes, to 0.672 for the 16 class (25m^ resolution) land cover data.
Excluding these, the amount of chemistry variation accounted for is fairly similar for each
analysis. Given this similarity, land classification datasets were subsequently excluded from
further analysis because these include geology, soil, and morphological components. The
land cover data relates specifically to vegetation and as such is more useful in a multivariate
analysis. Ultimately, the classification with the coarsest aggregation ( 6 classes) was selected
for inclusion in the Phase 2 analysis despite explaining 17% less of the variation in water
chemistry because this enables a reduction in the number of variables. A forward selection
(in tandem with Monte Carlo Significance Tests) on the 16 class dataset produced a sum
of canonical eigenvalues of 0.471 with only three of the classes significant. A similar
exercise on the 6 class dataset produced a value of 0.439 reducing difference in the level
of explanation to only 3% thus reinforcing the choice of this classification for inclusion in
subsequent PDA models.
384
Table A6.2.1: Results of RDA on selected land use classifications
E ig e n v a lu e s
Classifications Axis 1 Axis 2 Axis 3 Axis 4 ^eigenvalues
Land classification at 1km resolution (15 classes,Lcl 18-32)' .377 .069 .024 .012 .501Land classification at 1km resolution (4 classes)^ .275 .020 .005 .003 .302Land cover at Ikm^ resolution (16 classes (LVaO-25)^ .448 .038 .028 .010 .542Land cover at Ikm^ resolution (9 classes (LVb 1-9)^ .430 .026 .017 .007 .487Land cover at 25m resolution (16 classes (LVcO-25)'’ .475 .071 .053 .036 .672Land cover at 25m resolution (9 classes (LVdl-9)^ .432 .040 .023 .020 .521Land cover at 25m resolution (6 classes (LVel-6)’ .438 .039 .013 .011 .505
’ The dominant classes along axis 1 are Lcl25 (intensively farmed lowlands) and Lcl26 (mainly good grassland) while axis 2 is dominated by Lcl23 (high altitude with limited open range grazing) and Lcl30 (open range grazing on peatlands). This classification explains 50% of the water chemistry variation.
Of the four classes in this classification the percentages of upland and the percentage of arable land in the catchment have the highest loadings (-0.8498 and 0.9(X)6) respectively.
The highest variable loadings for axis 1 are LValO (open shrub moor, -0.7986) and LVal8 (Tilled land, -.0.7536). Approximately 54% of the variation in water chemistry is explained.
“ Axis 1 is dominated by LVb3 (arable, 0.7717) and LVb7 (lowland semi-natural grass/moor;shrub, -0.8118). On the second axisthe highest variable loading is exhibited by LVb8 (upland semi-natural grass/moor;grass, -0.7694).
LVclO and LVcl8 also have the highest variable loadings at this resolution. LVc6 (mown/grazed turf is also highly correlated (0.7689). The variance explained by catchment attributes is 67%, 48% by axis 1
*' LVd7 (as above, -0.7916) and LVd3 (as “ above, 0.8617) together with LVd2 (agricultural grass, 0.8815) have the highest axis1 loadings)
’ Axis 1 is polarised in terms of the dominant classes with LVe2 (agricultural grass and arable) and LVe6 (upland semi-natural grass/moor) exhibiting the highest variable loadings (0.9389 and -0.7258 respectively. On axis 2 LVe4 (coniferous woodland) has the highest loading (0.7528)
A6.2b Redundancy analysis using chemistry and two soil classifications
In Chapter 4 the derivation of the various soil variables is discussed in detail. Soil maps are
utilised both at soil association scale and soil map unit scale. Each soil association
comprises a number of soil map units. To produce attributes at the association scale, soil
map unit areas in each catchment were aggregated into the appropriate soil associations.
The sensitivity and soil critical load classes for each association are based on the classes
most common among the soil map units comprising the association.
385
In this section two RDA’s were run, one using the association scale soil data and one using
the soil map unit data. All other catchment attributes are excluded. Table A6.2.2 summarises
the RDA results on the association scale and the soil map unit scale data.
The sum of all canonical variables (X^) for the association scale dataset is 0.449 (/a, the
explanatory soil variables explain 44.9% of the variation in the water chemistry). However,
the soil map unit dataset explains 58% (X^O.513) of water chemistry variation. The first
canonical axis, in both instances, dominates the structure of the RDA. This is not surprising
given the unidirectional variation of the soils data (as witnessed by the PCA on the
catchment attributes in Chapter 6 ) along a gradient of sensitivity.
Table A6.2.2: Results of RDA’s on soil association and soil map unit data
Canonical Eigenvalues
Scale Axis 1 Axis 2 Axis 3 Axis 4 XT
Soil association data .355 .053 .019 .010 .449Soil map unit data .457 .072 .023 .013 .580
The disparity between the levels of explanation offered by the two analyses may be the
result of differences in mapping scale. Although each soil association overlies a distinctive
parent material, the soil map units within that association may be characterised by different
component soils, landforms and vegetation. The Lochinver association, for example, is
formed on drifts derived from Lewisian gneisses. Within the association, map unit 386
(component soils; brown forest soils, humus-iron podzols, some non-calcareous gleys, peaty
gleys and rankers) is typically found on valley sides and undulating lowlands characterised
by arable and permanent pastures. By way of contrast, map unit 396 (subalpine podzols and
386
peat) covers moderately rocky mountain slopes. Typical vegetation includes mountain heath
communities and mountain blanket bog. As a consequence any fine level variation
parameter variation which might occur as a result of these map unit differences will not
emerge at the association scale. Additionally, the map unit data includes catchment values
for soil H"' concentration and percentage base saturation. These are available for most soil
map units and may account for a substantial proportion of water chemistry variance. A
catchment weathering rate value has been calculated and is included among the association
scale data. However, this is only available for 18 major soil associations (Table 4.10).
Given the higher level of explanation obtained by using the soil map unit data, soil variables
at this scale are utilised throughout the direct gradient analyses presented in Chapter 6 .
A6.2C Redundancy analysis incorporating data relating to superficial deposits
The primary aim of this section is to assess the effects of incorporating data relating to
superficial drift deposits in a redundancy analysis. The data were derived in the same way
as that for the soil and solid geology variables. Drift maps were digitised and catchments
characterised according to the percentage of different drift types in each. Six classes of
superficial drift were identified from the map legends (see Chapter 4). However, maps of
superficial drift data in Scotland do not provide as comprehensive a coverage as solid
geology and soil maps. As a consequence, only 48 of the 78 calibration sites had drift data
available. If the analyses were confined to these sites the soil, geology and land cover
representativeness intended during the research design would not be achieved. As such the
analyses in this section seek to investigate whether the omission of a superficial drift
component in the model development substantially reduces the level of explanation.
387
The results of RDA undertaken, using the soil map unit data alone, and with the inclusion
of the superficial drift classification (see Chapter 4) are shown in Tabie A6.2.3. These
include analyses with and without fonA/ard selection. Monte Carlo Permutation Significance
tests are employed to identify soil and drift data that are significant (at the 95% level) in
terms of explaining the water chemistry variation.
Table A6.2.3: Results of RDA’s on soil association and soil map unit data
C a n o n ic a l E ig e n v a lu e s
Axis 1 Axis 2 Axis 3 Axis 4 r
Soil data only .505 .141 .045 .018 .728Soil and superficial drift data .521 .189 .052 .025 .819Soil data with forward selection .448 .086 .025 n/a* .560Soil and drift data with forward selection .493 .157 .034 .007 .699
* Only three variables were found to be significant using soil data only and consequently only three constrained axes were derived. Axis 4 is unconstrained
The RDA run using drift data increases 'ZX from 0.728 to 0.819 an increase in explanation
of approximately 9%. The drift classes closely associated with high positive Axis 1 scores
are D1 (boulder clay/till) and D4 (raised beach and marine deposits) although the latter
occurs in very few sites. D3 (moraines) is associated with high negative axis 1 sores. Thus
the chemistry determinands which vary positively along Axis 1 (including conductivity, Mg^*,
Ca "" and DCL) appear to occur in higher concentrations in those catchments rich in boulder
clay deposits. For the RDA’s using forward selection ZÀ is reduced to 0.560 and 0.699 for
the soil only and soil/drift datasets respectively. The soil/drift data thus explains
approximately 14% more of water chemistry variation than the soil data. Figure A6.3.1 is a
correlation biplot which illustrates the relationships between the soil/drift variables and the
water chemistry determinands within the context of the first two canonical axes. The
significant soil variables (from both analyses) are, area covered by low sensitivity soils (SL),
the derived soil critical load variable (SCL) and soil water concentration. These dominate
388
Axis 1 with biplot scores of 0.9036, 0.7902 and -.8300 respectively and account, to a large
extent, for the sensitivity gradient identified in Chapter 6 . The drift variables D5 (peat) and
D4 have the highest scores on Axis 2. D5 is associated with sites with high Abs-250 linking
peat coverage with waters with high organic content. D4 occurs more in sites characterised
by high Na'" and Cl' which might be expected in coastal areas. RDA with forward selection
using DCL as a sole response variable (not reported) identified SL and SCL3 as significantly
accounting for critical load variation. None of the drift classes were found to be significant.
The question as to whether to use drift data hinges on the importance attached to the
increased explanation offered by the RDA model which included the drift classification.
Additionally consideration must be given to the implications of the reduction in the number
of calibration sites that would be necessary to incorporate drift into the predictive modelling
exercise. The possibility that the potential influence that superficial drift has on surface water
chemistry is replicated by other variables in the dataset also requires attention.
The addition of 6 explanatory variables to a RDA is always likely to increase explanation and
it is necessary to be aware that this may simply be a result of the adding more variables and
not of a meaningful increase in explanation. Additional explanatory variables will not result
in lower eigenvalues. The use of forward selection (see Chapter 4) during RDA rejects those
explanatory variables whose contribution to the overall variation in the response data is not
statistically significant. This showed that soil variables were most associated with Axis 1
while drift variables were most associated with subsequent axes.
By reducing the calibration dataset from 78 to 48 sites (and adding to the number of
catchment attributes) the number of explanatory variables will become very large relative
to the number of sites. This may cause instability when running the analyses and lead to
spurious results.
389
The role of drift deposits in determining surface water chemistry is based, to a large extent,
on the provision of base cations through the processes of weathering. Where a base rich
drift deposit overlies an area of poorly buffered solid geology the acid sensitivity
Figure A6.2.1: RDA correlation biplot showing soil and drift variables (dashed vectors) and water chemistry determinands (solid vectors) n=48
1.00
A b s -2 5 0
0 .75 —
No
Al —tmA l—nl
A l - I0 .5 0 — ^ D 4
D5OII CondS 0 4(N—
<<Û
0 .2 5 —
Mg
N 0 30.00 —
D3HCL
SCLD2
-0.25 — Aik
H+
-0.50
- 1.00 -0.75 -0.50 -0.25 0.00 0.25 0.50 0.75 1.00
RDA Axis 1 = 0.493
of surface waters in the catchment is reduced. (Shilts, 1981; Norton, 1990). Clearly the
nature of superficial deposits are important in this context. However soils in Scotland are
classified at the association scale according to underlying parent material. In most instances
this will be drift material of some sort. An examination of soil and drift data for sites included
in the analysis here shows that while the latter provide information relating to surface
coverage of, for example, boulder clay or moraine, soil maps based on soil associations
390
provides information relating to the derivation of the drift material. According to the relevant
drift geology map Site mk12, for example, has approximately 80% boulder clay coverage.
The soil map for the same area shows that the dominant soil associations within the
catchment are Strichen (parent materials predominantly drifts derived from arenaceous
schists and strongly metamorphosed argillaceous schists) Forfar (Water sorted drifts derived
from Lower Old Red Sandstone) and Gourdie/Callander/Strathfinella (drifts derived from acid
metamorphic rocks and Lower Old Sandstone sediments with igneous rocks). This illustrates
the more detailed information provided by the soil maps.
It is apparent that the influence of drift deposits can be incorporated into the analyses by the
inclusion of the overlying soil. It is also possible that statistical problems will encountered
with low explanatory:response variable ratios. Moreover, the drift variables plotted in Figure
A6.2.1 indicate that the drift variables are associated less with the primary sensitivity
gradient than with sea-salts and organic content. Given that the primary objective of this
research is to calibrate a model which will effectively predict sensitivity to acidification the
drift variables discussed will not be used In subsequent analyses
391
Appendix 6.3 : Summary statistics for untransformed water chemistry variables(sensitive subset)
Mean S.D Min. Max.
pH 6.69 0.74 4.87 7.81Alkalinity 179.52 178.59 -12.00 689.00Conductivity 56.35 24.56 20.00 105.00Na* 272.91 129.86 98.00 643.00K+ 12.25 7.19 6.00 45.00
138.43 103.64 39.00 485.00Ca"+ 211.35 108.87 60.00 394.00CI 247.15 125.45 73.00 623.00NQ,- 5.20 14.27 B.D 95.00S O / 77.04 62.05 28.00 419.00AL-TM 46.83 58.99 B.D 350.00AL-NL 36.20 45.15 B.D 258.00AL-L 10.63 15.63 B.D 92.00ABS-250 0.39 0.34 0.04 1.48HCL 6.43 5.44 1.79 24.37DCL 2.33 1.46 0.58 5.67
B.D = Below detection limit
392
Appendix 6.4: Matrix of Pearson product-moment correlations for 16 transformed water chemistry determinands (sensitive subset, n = 46).
wCDGO
Aik 0.931
Cond 0.249 0.399
Na* -0.048 0.097 0.852
K* -0.317 -0.164 0.452 0.418
Mg^ 0.339 0.477 0.909 0.752 0.383
Ca^ ■= 0.734 .0.821 0.711 > 032 3 0.152 0.648
Cl -0.116 0.022 0.824 0.980 0.475 0.702 0.273
n o ; 0.184 0.242 0.221 -0.014 0.314 0.092 0.422 -0.051
S O / -0.278 -0.111 0.346 0.135 0.707 0.338 0.283 0.208 0.480
Al-TM -0.664 -0.561 0.060 0.389 0.358 -0.025 -0.404 0.440 -0.391 0.169
AI-NL . -0.692 -0.597 -0.004 0.360 0.339 -0.088 -0.459 ■ 0.408 -0.397 0.142 0.977
Al-L : -0.479 ‘ -0.365 0.252 0.369 0.338 0.161 -0.137 0.407 -0.250 ; 0Î307 . 0.768 0.642
ABS-250 ■ -0.593 -0.410 0.422 0.648 0.469 0.332 -0.136 0.679 -0.195 0.317 0.803 0.806 0.598
HCL 0.745 * : 0.758:: ^iÿp':558:' : ' 0.359 -0.213 0.592 0.695 0.270 0.037 -0.360' -0.431 ' -0.492 -0.192 -0.224
DCL ■ 0.834 0.863: : 0.619 0.280 -0.093 0.606 0.905 0.207 0.286 -0.076 -0.532 -0.586 -0.284 -0.244 0.855
pH Aik Cond Na* K* Ca^ Cl n o ; S O / Al-TM AI-NL Al-L ABS-250 HCL
C o rrela lion coeffic ients w ith /j-va lues ^ 0 .0 5 are shaded. These are significant at the 9 5 % level
Appendix 6.5: Summary statistics for untransformed catchment attributes - siteswhere ca " <=400^eq
Mean S.D Min Max
Catchment area (km^) (Area) 4.24 3.41 0.62 19.67Stream length (km) (SJength) 10.34 12.29 1.41 84.80Distance from sea (km) (dist) 25.43 23.23 0.02 76.05Site altitude (m) (Alts) 140.57 123.77 5.00 470.00Maximum altitude (m) (Altm) 616.80 267.93 172.00 1130.00Total S dep (89-92) (keq H*ha‘‘ y f ‘) (Sdep)
0.78 0.25 0.38 1.24
Total N dep (89-92) (keq H*ha' y f') (Ndep)
1.08 0.31 0.53 1.58
Rainfall (1989-92) mm y f' (Rain) 2157.83 826.87 764.00 3319.00Geology Class 1 (%) (01) 70.89 35.27 0.00 100.00Geology Class 2 (%) (G2) 17.02 25.47 0.00 87.61Geology Class 3 (%) (G3) 11.79 31.29 0.00 100.00Geology Class 4 (%) (G4)
Soil variables
0.25 1.24 0.00 7.53
High Soil Sensitivity (%) (SH) 88.22 21.91 10.09 100.00Medium Soil Sensitivity (%) (SM) 8.01 17.23 0.00 89.91Low Soil Sensitivity (%) (SL) 1.44 5.18 0.00 31.73Bare ground (%) (BARE) 2.32 7.99 0.00 45.87Soil Critical Load Class 1 (%) (SCLl) 0.97 3.05 0.00 18.51Soil Critical Load Class 2 (% ) (SCL2)^ 0.00 0.00 0.00 0.00Soil Critical Load Class 3 (%) (SCL3) 21.03 35.44 0.00 100.00Soil Critical Load Class 4 (%) (SCL4) 67.41 39.15 0.00 100.00Soil Critical Load Class 5 (%) (SCX5) 8.37 23.64 0.00 100.00WA* H* concentration (H*peq 1') (H+) 17.01 5.59 7.10 29.30WA* Base saturation (%) (%BS) WA* Soil Critical Load
16.57 7.69 4.76 42.98
(keq H+ ha' y f') SCL 4.09 2.39 0.00 11.32
Land Cover Class 1 (%) LCl 5.88 9.29 0.00 38.00Land Cover Class 2 (% ) LC2 1.98 4.54 0.00 27.69Land Cover Class 3 (%) LC3 2.09 4.52 0.00 25.39Land Cover Class 4 (%) LC4 4.11 8.50 0.00 39.97Land Cover Class 5 (% ) LC5 50.23 17.34 0.00 100.00Land Cover Class 6 (%) LC6
* Calculated using weighted averages SCL2 does not feature in the sensitive sub-set
35.70 18.39 0.00 100.00
394
Appendix 6.6: Matrix of Pearson product-moment correlations for 28 transformed catchment attributes (sensitive subset, n = 46) SCL2 not present in any catchment).
%ÜI
S > n
Dht 021# 0.102
AKm r î ; oj#i
S 4 tp •0260 0.127 O M 9 0 1 )9
NArp -0200 013) 0040 0 0 )2 0.952
Rtin -0250 0114 t 0202 0091 ■ 0.76) f 0M 2
Gl 0.079 0.116 0246 0.064 0242 OI#1 0.265
C2 -0 001 -0.042 fp0A20^ OOO) 02)# r 0271 0.249 0049
G3 -COM < o A i) : 0.126 003# 0075 0.224 •0860 0.344
G4 0.069 0.1)0 0.117 0044 O.IIS 0198 0008 00)3 0253
0.170 0465 O 142 ,V 01)96 / 0 4 5 ) •0.136 0.456
SM -0004 00)2 0 0 2 ) 0260 i?,OJ75? r 04)2 0.064 0.343 0.741
SL 047) 0.157 02 IS 0119 0.155 0 277 0001 0 0 )0 ; 0213 0058 ^^ojo9: >;0J16f
Bart -0.161 -0092 0290 0.190 0.113 0.324 0.115 0051 ^ r o ji) 0.101 t> 4 3 4 3 i 0.147 ' O A K
SCLl 0.1#) 0250 016) 0.116 0127 0222 0017 0065 0256 004# <f440S‘ J ? :s 9 ^ r : 0 900 * 0201
SC U -4164 -0.1)7 0006 0.104 ^ 0 4 6 9 0.159 0726 z 023* K'O-n) 0.16) 0JO2 Î 0.18) 0241
5CL4 0467 0423 0242 ^ 7 ; 0764 0.152 > '4 6 )4 0086 0.124 0.021 0.122 0.027 i^ - a 6 U f
S C U 0491 0449 0157 OOS) 0.021 0115 0244 0201 0087 0249 0038 0.149 0011 0132 0.062 ■an# 0.184
H * 0279 0.165 0I#6 #0L462r #&4iO; 0252 O.066d •0.15) 0056 ^0.4061 0.100 0.0)9 -0061 # 4 4 0 ) ^ ff&497 0.148
%BS •0210 -015# 0.0)6 m m t 0.210 H i .o M : 0.119 0.109 0024 0004 0124 028) -0183 01575
SCL ■0.146 •0.094 0200 0456 0.260 0.2#0 % ;0299/ \^ 0 M 7 0.047 %'OL40#t 0.196 # 'Ü ) 4 4 | ■a%20 > 0 3 ) 7
LC l 044) 0.147 m o-*9u 0454 0005 0252 f ;0 J 2 4 t 0 2 4 ) 0 262 0.1)2 0252 0.284 0.244 ^ ^4 3 4 7 . 0204 #0 J 0 8 * 0025 0009 0006 0184
t o -045# -0210 0.127 0.229 0216 0.22# 0276 0.109 0089 0.19) 0008 0.001 01)3 0 0 )3 0.026 0.029 0.14) -aoo) 0089 0098 0191 0122 0160
t o -0.074 ■0170 0.069 0211 0.2## ir-0223. ^XUO# 0.167 0.099 0.069 0.104 0.164 0.102 0117 00)9 0066 0.248 -0217 0375 ' -0127 0084 0 1 8 )
LC4 4.1)9 -002# 0457 0015 0090 0.068 0.237 0.1)7 0.139 0.118 0.111 0081 0058 0044 0.016 0 248 Y 0)20 0.212 0158 0.10) 0 0 4 ) 0025 0007 ^<=0362
t o 0.174 0244 0.10# 0262 0.006 0.072 0047 ^<^0463 ' 0076 '^ '4 5 0 6 0.074 0.020 0022 0029 0.060 0015 0282 0246 0.020 0018 t ; -A )0 4 i 002) 0003 01 50 0225
tC 4 ■0.116 -0.099 043# 0409 0.070 T .0J73: 0 0 4 ) £V 0J4I 0087 0.189 0086 0.105 0.12) 0 097 0269 # '■ 0 35 0 ; 0.269 0 1 )7 0216 0218 V o j9 j : 0190 0009 0169
1 Otal Akin Sdtp Ndtp Gl G2 GJ G4 SH SM SL B tr t SCLl s e t) SC U sets H* %BS set t c i LC2 t o 1X4 u . 1
Correlation cocflicients with p - values <0.05 arc shaded. These are .significant at the 95% level.
Additional references
Bache, B.W. (1983) The implications of rock weathering for acid neutralization. In Ecological Effects of Acid Deposition : National Swedish Environment Protection Board - Report PM1636
Berden, M., Nilsson, S.I., Rosen, K. and Tyler, G. (1987) Soil acidification, extent, causes and consequences. Report No. 3292. National Swedish Environmental Protection Board.
BIbby, U.S., Douglas, H.A., Tomasson, A.J. and Robertson, J.S., (1982) Land Capability Ciassification for Agriculture Soil Survey of Scotland Monograph, Macauley Institute for Soil Research, Aberdeen.
Bricker, O.P., Godfrey, A.E. and Cleaves, E.T. (1967) Mineral water interaction during the chemical weathering of silicates. In Trace Inorganics in Water, A.C.S. Advanced Chemistry Series No. 73.
Chrlstopherson, N. and Wright, R.F. (1981) Sulfate budget and a model for sulfate concentrations in streamwater at Birkenes, a small catchment in southernmost Norway.Water Resources Research 17, 377-389.
Cresser, M.S. and Edwards, A C. (1987) Acidification of freshwaters Cambridge University Press.
Cresser, M.S., Edwards, A.C., Ingram, S., Sklba, U., and Plerson-Smlth, T., (1986) Soil- acid deposition interactions and their possible effects on geochemical weathering rates in British uplands. Journal of the Royal Geological Society, London 143, 649-658
Cronan, C.S., Reiners, W.A., Reynolds, R.C. and Lang, G.E. (1978) Forest floor leaching: Cintributions from mineral, organic and carbionic acids in New Hampshire subalpine forests. Science 200, 309-311.
Davis, R.B., Anderson, D.S., and Berge, F. (1985) Palaeolimnological evidence that lake acidification is accompanied by loss of organic matter. Nature 316, 436-438.
Gherlnin, S.A., Mok, L., Hudson, R.J.M., Davis, G.F., Chen, C.W. and Goldstein, R.A.(1985) The ILWAS Model: Formulation and application. Water, Air and Soil Pollution 2S, 425- 459
Galloway, J.N., Norton, S.A. and Church, M.R. (1983) Freshwater Acidification from Atmospheric Deposition of Sulfuric Acid: A Conceptual Model. Environmental Science and Technology 17, 541A-545A
Goldlch, S.S. (1938) A study in rock weathering. Journal of Geology 46,17-58.
Hall., J., Hornung, M., Freer-Smlth, P., Loveland, P., Bradley, I., Langan, S., Dyke, H., Gascoigne, J. and Bull, K. (1997) Current status of UK critical loads data - December 1996 Report to the Department of Environment prepared under contract PECD7/10/90.
Harrlman, R. (1988) Patterns of surface water acidification in Scotland, in Acidification in Scotland Proceedings of a Symposium organised by the Scottish Development Department, Edinburgh, 1988 pp72-79
Hemond, H.F. (1990) Wetlands as the source of dissolved organic carbon to surface waters. In Perdue, E.M. and Gjessing, E.T. Organic Acids in Aquatic Ecosystems. Dahlem Workshop report LS 48. Wiley, Chichester.
396
Hemond, H.F. (1994) The Role of Organic Acids in Acidification of Freshwaters. In Steinberg, C.E.W. and Wright, R.F. (Eds.) Acidification of Freshwater Ecosystems \N\\ey and Sons, UK.
Jackson, M.L., Tyler, S.A., Willis, A.L., Bourbeau, G.A. and Pennington, P.P. (1948) Weathering sequence of clay size minerals in soils and sediments. J. Phys. ColL Chem., 52, 1237-1260.
Jenkins, A., Renshaw, M., Helliwell, P., Sefton, C., Perrier, P. and Swingewood, P.(1987) Modelling surface water acidification in the UK; Application of the MAGIC model to the Acid Water Monitoring Network. Institute of Hydrology, Report No. 131.
Johnson, D.W. and Cole, D.W. (1977) Sulfate mobility in an outwash soil in western Washington. Water, Air and Soil Pollution 7,489-495.
Kâmâri, J., Forsuis, M., Kortelainen, P., Mannio, J. and Verta, M. (1991) Finnish lake survey: Present status of acidification. Ambio 20, 23-27.
Kinninburgh, D.G. (1986) Towards more detailed methods for quantifying the acid susceptibility of rocks and soils. Journai of the Geological Society, London. 10, 679-690.
Krug, E.G. and Frink, O.P. (1983) Acid rain on acid soil: a new perspective. Science 221, 520-525
Kullberg, A., Bishop, K., Hargeby, A., Jannson, M. and Petersen, Jr. P.O. (1993) The ecological significance of dissolved organic carbon in acidified waters. Ambio, 22, 331-337
Likens, G.E., Gormann, F.H., Pierce, P.S., Eaton, J.S. and Johnson, N.M. (1977) Biogeochemistry of a forested ecosystem Springer, New York.
Mayer, P., and Ulrich, B. (1977) Acidity of precipitation as influenced by the filtering of atmospheric S and N compounds - its role in the element balance and effect on soil. Water, Air and Soil Pollution 7, 409-416
Miller, H.G. (1981) Forest fertilisation, some guiding concepts. Forestry 54,157-167
Peid, J.M., MacLeod, D.A. and Cresser, M.S. (1981) The assessment of chemical weathering rates within an upland catchment in north-east Scotland. Earth Surface Processes and Landforms 6, 447-457.
Peuss, J.O., Chrlstopherson, N. and Seip, H.H. (1986) A critique of models for freshwater and soil acidification. Water, Air and Soil Pollution 30, 909-930
Pobson,, J.D., and Thomasson, A.J. (1977) Soil water regimes Soil Survey Technical Monograph No. 11, Rothampstead Experimental Station, Harpenden.
Seip, H.M. (1980) Acidification of freshwaters: Sources and mechanisms. In Drablos D. and Tolland, A. (Eds.) Ecological Impact of Acid Precipitation SNSf Project, Norwegian Instistue for Water Research, Oslo, Norway. pp358-365.
Skeffington, P.A. (1983) Soil properties under three species of tree in southern England in relation to acid deposition of throughfall. In Effects of Accumulation of Air Pollutants on Forest Ecosystems Re\6e\, Boston. pp219-231.
Theis, H. (1990) Acidification studies at northern Black Forest cirque lakes. In Hydrology in Mountainous Regions. I Hydrological measurements; The Water Cycle.\AHS Publication No. 193.
397
Ulrich, B. (1983) Soil Acidity and Its Relation to Acid Deposition. In Effects of Accumulation of Air Pollutants on Forest Ecosystems Reidel, Boston. pp127-146.
Van Breeman, N., Driscoll, C.T., and Mulder, J. (1984) Acid deposition and internal proton sources in acidification of soils and waters. Nature, 307, 599-604.
Whitehead, P.O., Bird, S., Hornung, M., Cosby, B.J., Neal, 0. and Parcols, P. (1988) Modelling long-term stream acidifcation trends in upland Wales at Plynlimon. Hydrological Processes 2, 357-368.
Wilson, M.J., Lilly, A. and Nolan, A.J. (1988) Vulnerable soils and their distribution. In Acidification in Scotiand Proceedings of a Symposium organised by the Scottish Development Department, Edinburgh, 1988 pp60-70
398