Modelling the Hydrological Sensitivity to Land Use Change in a Tropical Mountainous Environment by Mauricio Edilberto Rincón Romero April 2001 A thesis submitted to the University of London for the degree of Doctor of Philosophy Department of Geography King’s College London
358
Embed
Modelling the Hydrological Sensitivity to Land Use … the Hydrological Sensitivity to Land Use Change in a Tropical Mountainous Environment by ... a global issue 1
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Modelling the Hydrological Sensitivity to Land UseChange in a Tropical Mountainous Environment
by
Mauricio Edilberto Rincón Romero
April 2001
A thesis submitted to the University of London for the degreeof Doctor of Philosophy
Department of GeographyKing’s College London
To my lovely wife
And sons
Manuel Felipe and
Miguel Angel
Tambito River (Rincón-Romero, 1997)
i
Abstract
The main subject of this thesis is the production of a sensitivity
analysis to land use and land cover change (LUCC) for a tropical
montane cloud forest (TMCF) environment on the basis of flux
responses in a hydrological model. Human pressure is one of the
main causes of LUCC in the TMCF which often results in important
consequences on natural resources like reduction of water quality,
loss of biodiversity, micro-climatic change or ecosystem
degradation (Koning et al., 1998). Deforestation of the tropical
cloud forest is an activity of recent decades that is modifying the
landscape significantly. The impact of this deforestation, rather
than the deforestation itself, is studied here by comparing variation
in fluxes of erosion and overland flow derived from different land
uses within a mountainous tropical forest catchment. A physically
based hydrological model of the Tambito watershed, Cauca-
Colombia and 5 LUCC pattern scenarios are implemented for the
study. A 2.5D dynamic surface hydrological model integrated with
a Geographic Information System (GIS) working on an hourly time
step is designed for the catchment, to assess flux variability in time
and space. The hydrological model includes the following sub-
modules: solar radiation and energy balance, evaporation,
interception and effective precipitation, infiltration, soil
hydrological balance, overland flow, recharge and erosion. Three
hydro-meteorological stations installed on experimental plots
collect basic model information for parameterisation and
validation. Experimental description, methodology, field data,
model implementation and analysed results are presented. Each
LUCC scenario uses 15 to 22 consecutive GIS iterations, which
transform forest to pasture within the catchment. Summaries of
annual average hydrological flux variations are used in the
sensitivity analysis. Multiple linear correlation was carried out for
ii
flux variations and hydrological sensitivity with landscape physical
properties of the deforested area by iteration for each scenario, in
order to determine the correlation between landscape catchment
physical properties and hydrological flux sensitivities. This process
also facilitated the identification of the topographic characteristics
of the most sensitive areas within the catchment to LUCC. The
model and statistical analysis provides a means of assessing the
contribution of different landscape units to hydrological change in
the face of LUCC. The impact of LUCC is assessed in terms of
catchment hydrological changes and the areas within the
catchment with more hydrological sensitivity to LUCC are
identified.
iii
ACKNOWLEDGEMENT
This thesis was funded by the Higher Education Programme of the
Office of Presidency of Republic of Colombia, through
COLCIENCIAS, and the “Instituto de Investigaciones Biológicas
Alexander Von Humboldt” for the development and application of
GIS-modelling technologies in Colombia. Additionally some field-
work expedition were possible due to the help of the University of
London Central Research Fund.
I want to express my special thanks to my supervisor Dr. Mark
Mulligan for his unconditional support in all fields including
academic, logistic, personal and moral. Without his direct
assistance this thesis would not have been possible. Also, I would
like to express my gratitude to the late Alvaro Jose Negret, of
Fundacion Proselva and the University of Cauca, Popayan, for his
enthusiastic support and permission to use the Tambito field site
and its facilities. I am also grateful to other Colombian
organisations that provided logistic support, in particular the
International Centre of Tropical Agriculture (CIAT), the Regional
Corporation for Cauca (CRC) and the ‘Instituto de Hidrología
Meteorología y estudios Ambientales’ (IDEAM), this last one who
brought meteorological information of the region.
I would like to make a special mention to the other KCL students
who came to the Tambito Reserve, to provide their assistance in the
field, and additionally, who gave a pleasant touch to the difficult
iv
experience, making the situation bearable and enjoyable. They
were Koulla Pallaris, Andrew Jarvis, Jorge Rubiano, Robert Stein
Rostaing, Matthew Letts, Juliana Gonzalez, Sim Reaney and Lydia
Bruce-Burgess. Also there were some special people who gave local
support in the campaign activities particularly Quintin and Olga.
This work would not have been possible without constant and
valuable support from my family, particularly my lovely wife who
day by day was behind my shoulders encouraging me and feeding
my hopes to get successful results. Also, to my beautiful and
innocent son Manuel Felipe for his constant stimulation and for
showing me the sense of our life. Also, my father and my brothers
who have been constantly interested in my progress / development
with this thesis. Thanks to all of them.
It is impossible to pass without mentioning the ‘DUNGEON’
friends, Andy, Sotto, Benny, Elias, Matt, Jim, and Christos,
because through the circumstances, we became a family, giving
our support to each other in both personal and academic aspects.
They were the ones who made London a pleasant place to live, and
the dungeon an agreeable palace in the middle of an eighteenth
century building. I would also like to thank the Pallaris family, who
welcomed me in their home during the last stages of the writing up
this thesis.
v
Table of contents
Chapter I Introduction
1.1 Land use and cover change (LUCC): a global issue 11.2 LUCC: global impacts 31.3 A review of models for LUCC 5
1.3.1 Methods for identifying the impact of LUCC 71.3.2 Strategies for evaluating hydrological fluxes in the
assessment of LUCC impact 81.4 LUCC: issues and impacts in tropical montane
environments outside Colombia 91.5 LUCC in Colombia: History and impacts in hillside
areas 111.5.1 Historical review of LUCC in Colombia 111.5.2 The hydrological impacts of LUCC in Colombia 16
1.6 Structure of the thesis 20
Chapter II Literature review of hydrological models applied to LUCC impacts research
2.1 Structure of this chapter 232.2 General concepts of hydrological models 232.3 A general classification of hydrological models 242.4 Handling spatial variability in hydrological models 262.5 A review of hydrological models related to LUCC impact 272.6 Hydrological modelling in tropical montane
environments 372.6.1 A review of modelling studies of the hydrologicalimpact of LUCC in Colombia 38
2.7 Research approaches to LUCC impacts 392.8 Main objective 402.9 Specific aims 402.10 Rationale 41
vi
Chapter III Methodology
3.1 Structure of this chapter 423.2 Description of the study area 433.3 Experimental strategy 493.4 Land use change scenario generation for this thesis 50
3.4.1 Estimating initial vegetation cover for LUCCScenarios 51
3.4.2 Scenario descriptions 543.5 Field methodology 64
3.5.1 Plot scale 643.5.1.1 The hydrological weather stations 673.5.1.2 Data collected from the weather stations 70
3.5.2 Catchment scale 713.5.2.1 Soil data 713.5.2.2 vegetation data 74
3.5.2.2.1 Leaf area index 753.5.2.2.2 Vegetation cover 763.5.2.2.3 Canopy water storage capacity 76
3.5.3 Other spatial data 783.6 Hydrological Modelling methodology 83
3.6.1 Introduction 833.6.2 Strategy 833.6.3 Consideration for modelling process 87Climate3.6.4 Solar Radiation sub-model 90
3.6.4.1 Hourly extraterrestrial solar radiationModel 91
3.6.4.2 Hourly cloud-cover attenuation model 923.6.4.3 Net solar radiation function 98
Hydrology3.6.5 Evaporation sub-model 1003.6.6 Canopy storage, interception and throughfall 1083.6.6.1 The Rutter model 1113.6.7 Sub-surface water sub-model 115
3.6.7.1 Modelling flow of water in porous media 1163.6.7.2 Soil water retention and matric potential 1173.6.7.3 Pedotransfer functions 119
3.6.9.1 Sub-model description 1313.6.9.2 Surface component of overland flow at the
catchment scale 1343.6.10 Erosion sub-model 134
vii
3.7 Integrating the sub-models in the 1D and 2,5D model 1393.7.1 Module sequence 1393.7.2 Data used in the model 1403.7.3 Parameters used in the model 145
Chapter IV Model results, sensitivity analysis andvalidation
4.1 Structure of this chapter 1474.2 Model results 148
4.2.1 Model results at the plot scale 1484.2.2 Model results at the catchment scale 156
4.3 Sensitivity analysis of the hydrological model at theplot scale (1D model) 1604.3.1 Sensitivity to parameter A of net radiation 1624.3.2 Sensitivity to parameter B of net radiation
equation 1654.3.3 Sensitivity to parameter light extinction K 1664.3.4 Sensitivity to parameter leaf area index (LAI) 1684.3.5 Sensitivity to parameter maximum canopy water
storage capacity 1704.3.6 Sensitivity to parameter vegetation cover 1724.3.7 Percent of variation due to soil texture 1764.3.8 Sensitivity to parameter soil porosity 1784.3.9 Sensitivity to parameter soil depth 1804.3.10 Sensitivity to parameter erodability factor, K1 1834.3.11 Sensitivity to parameter m factor of erosion
equation 1844.3.12 Sensitivity to parameter n factor of erosion
equation 1854.4 Summary of 1D sensitivity analysis 1864.5 2.5D model sensitivity analysis 187
4.5.1 Definition of topographic characteristics 1884.5.2 Sensitivity analysis at the catchment scale 189
4.5.2.1 Sensitivity analysis of overland flow to LUCC at the catchment scale 192
4.5.2.2 Sensitivity analysis of erosion toLUCC at the catchment scale 207
4.6 Summary of 2.5D sensitivity analysis 2204.7 Model validation 222
4.7.1 Organisation of this section 2224.7.2 Field data set for validation 2234.7.3 Parameters used in validation 2234.7.4 Validation of net solar radiation 2254.7.5 Validation of soil moisture 228
viii
Chapter V Summary, conclusions and further work
5.1 Summary of key finding in this thesis 2325.2 Conclusions and their implications 2325.3 Further research and model development 245
Bibliography 248
Appendix I LUCC scenarios 286
Appendix II Collected data from the pasture plot 294
Appendix III Summary of soil analysis samples 298
Appendix IV Summary of vegetation samples for:canopy water storage capacity, vegetationcover, and LAI for grassland 300
Appendix V Tambito daily rainfall data 304
Appendix VI Example of input data file for the model 310
Appendix VII Extraterrestrial solar radiation model 315
Appendix VIII Mean value of cloud cover 326
Appendix IX Hydrological of PCRaster program Code 328
Appendix X Summary of physical variables and modelvariables response for all scenario 336
ix
List of tables
Table 1.1 Land use census data comparison for Colombia
between 1960 and 1995. 15
Table 3.1 Average NDVI values for classification of land
use classes 54
Table 3.2 Rates of deforestation per iteration of the
different scenarios (values in ha.) 62
Table 3.3 Periods during which data were collected 71
Table 3.4 Classes of slope and land use 73
Table 3.5 Leaf area index samples for grassland 76
Table 3.6 Vegetation parameters 77
Table 3.7 Soil erodability factor (taken from Morgan and
Kirkby, 1980) 137
Table 3.8 Soil parameters used in the physically-based
hydrological model 146
Table 4.1 Parameters used in the physical hydrological
Model 149
Table 4.2 Parameters used in the physical hydrological
model 160
x
Table 4.3 Hourly average values of model variables for a
year simulation in 1 m2 161
Table 4.4 Colour code of the degree of sensitivity 162
Table 4.5 Sensitivity to parameter A in the net radiation
equation 163
Table 4.6 Sensitivity to parameter B in the net radiation
equation 165
Table 4.7 Sensitivity to light extinction 166
Table 4.8 Sensitivity to LAI 168
Table 4.9 Sensitivity to maximum canopy storage capacity 170
Table 4.14 Sensitivity to erodability factor k1 183
Table 4.15 Sensitivity to m factor of erosion equation 184
Table 4.16 Sensitivity to n factor of erosion equation 185
Table 4.17 Summary of 1D sensitivity analysis by classes
with the colour code 186
xi
Table 4.18 Summary of data used in OF sensitivity
Analysis 205
Table 4.19 Multiple regression analysis of overland flow
for all scenarios. Significant relationships
are highlighted 206
Table 4.20 Summary of data used in Erosion sensitivity
analysis 217
Table 4.21 Multiple regression analysis of erosion for all
scenarios 219
Table 4.22 Parameters used in model validation 224
Table 5.1 Overland flow and erosion model results for the
original vegetation (from Landsat TM, 1989)
comparison with other research 236
xii
List of figures
Figure 3.1 Location of the Tambito watershed and prevailing land uses 45
Figure 3.2 Monthly average rainfall of the nearest weatherstation to Tambito (3km distance) 47
Figure 3.3 NDVI radiance from Landsat TM for Tambito Catchment 52
Figure 3.4 Transition of LUC by scenarios(a) Scenario 1. Forest conversion to pastures from cellular automata.(b) Scenario 2 Forest conversion with a fixed distance from river channels.(c) Scenario 3. Forest conversion with a fixed distance
toward river channels.(d) Scenario 4. Forest conversion with a fixed altitudinal distance from lower point in up hill
direction.(e) Scenario 5. Forest conversion with a fixed
altitudinal distance from higher point in downhill direction. 55
Figure 3.5 An example of an iteration for SC1 56
Figure 3.6 An example of an iteration for SC2 57
Figure 3.7 An example of an iteration for SC3 59
Figure 3.8 An example of an iteration for SC4 60
Figure 3.9 An example of an iteration for SC5 63
Figure 3.10 Distribution of plots and weather stations 66
Figure 3.11 Location of gutters in plots 66
Figure 3.12 Throughfall collector 68
Figure 3.13 Weather station in deforested areas. 68
Figure 3.14 Classification map for collecting soil samples 72
Figure 3.15 Basic cartography of the area (source from IGAC, 1985) 80
xiii
Figure 3.16 Digital elevation model for the study area derived from digitised contours using Arc/Info 7.3 80
Figure 3.17 Slope map derived the digital elevation model 81
Figure 3.18 Aspect map derived from the digital elevation model 81
Figure 3.19 LUCC map for Tambito watershed from Fundación Proselva (Museo de História Natural 1996) 82
Figure 3.20 Landsat image TM for the study area, false colour (5,4,3) 82
Figure 3.21 Schematic diagram of the hydrological model 86
Figure 3.22 Hourly cloud cover 96
Figure 3.23 Range of modelled cloud cover 97
Figure 3.24 Linear relation between measured and modelled cloud cover 97
Figure 3.25 Regression for computing net radiation in the model. 98
Figure 3.26 Diagram of net solar radiation model 99
Figure 3.27 Evapotranspiration variation with respect to Net Solar Radiation using the Penman-Monteith equation 106
Figure 3.28 Evaporation variation with respect to atmospheric and stomatal conductance 106
Figure 3.29 Flow diagram for potential evaporation 109
Figure 3.30 Diagram of the Rutter model (Jetten, 1994) 112
Figure 3.33 Diagram of soil hydrologic characteristics 123
Figure 3.34 Diagram of infiltration sub-model 130
xiv
Figure 3.35 Diagram of runoff sub-model. 133
Figure 3.36 Diagram of erosion sub- model 138
Figure 3.37 A map of simulated rainfall distribution for Tambito watershed 142
Figure 3.38 One year of hourly rainfall from Tambito weather station (1995) 143
Figure 3.39 Histogram distribution for Tambito rainfall using simulated data of 1995 144
Figure 4.1 Modelled evaporation with 1D model for forestand grassland LUCC compared with the rainfallevents 151
Figure 4.2 Modelled canopy interception with 1D model forforest and grassland LUCC, compared withrainfall events 151
Figure 4.3 Modelled matric potential with 1D model forforest and grassland LUCC, compared withrainfall events 152
Figure 4.4 Modelled hydraulic conductivity with 1D modelfor forest and grassland LUCC, compared withrainfall events 152
Figure 4.5 Modelled infiltration with 1D model for forestand grassland LUCC 153
Figure 4.6 Modelled soil moisture with 1D model for forestand grassland LUCC compared with rainfallevents 153
Figure 4.7 Modelled overland flow with 1D model for forestand grassland compared with rainfall events 154
Figure 4.8 Difference between modelled overland flow forboth forest and grassland LUCC, compared withthe rainfall events 154
xv
Figure 4.9 Modelled erosion with 1D model for forest andgrassland, compared with rainfall events 155
Figure 4.10 Difference between modelled erosion for forestand grassland, compared with rainfall events 155
Figure 4.11 Changes in overland flow due to LUCC (units inmm) for a modelled year. 157
Figure 4.12 Changes in erosion due to LUCC (units inmm m-2) in a modelled year 159
Figure 4.13 Sensitivity to parameter A in the net radiationequation 164
Figure 4.14 Sensitivity to parameter B of the net radiationequation 165
Figure 4.15 Sensitivity to light extinction 167
Figure 4.16 Sensitivity to LAI 169
Figure 4.17 Sensitivity to maximum canopy storagecapacity 171
Figure 4.18 Sensitivity to vegetation cover 174
Figure 4.19 Sensitivity to soil textures 177
Figure 4.20 Sensitivity to soil porosity 179
Figure 4.21 Sensitivity to soil depth 182
Figure 4.22 Sensitivity to erodability factor k1 183
Figure 4.23 Sensitivity to m factor of erosion equation 184
Figure 4.24 Sensitivity to n factor of erosion equation 185
Figure 4.25 Modelled soil moisture with different initialconditions for the same rainfall pattern 191
Figure 4.26 Overland flow sensitivity in scenario 1(deforested pattern with cellular automata) 194
Figure 4.27 Mean topographic variables for deforested areasin SC1 194
xvi
Figure 4.28 Overland flow sensitivity in scenario 2 (forestconversion with a fixed horizontal distance fromriver channel in uphill direction) 195
Figure 4.29 Mean topographic variables of deforested areasin SC2 195
Figure 4.30 Overland flow sensitivity in scenario 3 (forestconversion with a fixed horizontal distancetowards channel rivers in downhill direction) 197
Figure 4.31 Mean topographic variables of deforested areasin SC3 197
Figure 4.32 Overland flow sensitivity in scenario 4 (forestconversion with fixed distance of altitude, inuphill direction from the lowest to the highestpoint) 199
Figure 4.33 Mean topographic variables of deforested areasin SC4 199
Figure 4.34 Overland flow sensitivity Scenario 5 (forestconversion with fixed distance of altitude, indownhill direction from the highest to thelower point) 201
Figure 4.35 Mean topographic variables of deforested areasin SC5 201
Figure 4.36 Erosion sensitivity in scenario 1 (deforestedpattern with cellular automata) 209
Figure 4.37 Mean topographic variables for deforested areasin SC1 209
Figure 4.38 Erosion sensitivity in scenario 2 (forestconversion with horizontal a fixed distancefrom river channel uphill direction) 210
Figure 4.39 Mean topographic variables of deforested areasin SC2 210
Figure 4.40 Erosion sensitivity in scenario 3 (forestconversion with horizontal a fixed distancetowards channel rivers downhill direction) 212
Figure 4.41 Mean topographic variables of deforested areasin SC3 212
xvii
Figure 4.42 Erosion sensitivity in scenario 4 (forestconversion with a fixed distance of altitude,in uphill direction from the lower to the highestpoint) 214
Figure 4.43 Mean topographic variables of deforested areasin SC4 214
Figure 4.44 Erosion sensitivity in scenario 5 (forestconversion with a fixed distance of altitude, indownhill direction from the highest to the lowerpoint) 215
Figure 4.45 Mean topographic variables of deforested areasin SC5 215
Figure 4.46 Modelled and measured solar net radiation forvalidation 226
Figure 4.47 Linear regression between modelled andmeasured net radiation in validation 226
Figure 4.48 Hourly average solar net radiation during theday for validation 227
Figure 4.49 Linear regression between hourly average ofsolar net radiation modelled and measured forvalidation 227
Figure 4.50 Modelled and measured soil moisture forvalidation 230
Figure 4.51 Linear regression of modelled and measured soilmoisture for validation 230
Figure 6.52 Daily soil moisture comparison betweenmodelled and measured, for validation, inJuly of 1999 231
Figure 4.53 Linear regression between measured andmodelled daily soil moisture, for validation 231
1
Chapter I Introduction
Land use and cover change (LUCC) has been recognised as a
modifying agent of landscapes. Some of the effects of LUCC on the
ecosystem are studied in this thesis, particularly those related to
the hydrological cycle.
In this chapter, first LUCC is investigated as one of many globally
important environmental changes. Later, in Colombia is analysed,
as a historical process (including socio-economic and political
factors) concentrating on its effects the hillside areas.
Subsequently, the thesis structure is discussed, indicating briefly
the content of the thesis chapters.
1.1 Land use and cover change (LUCC): a global issue
Of the world’s 12000 million ha. of tropical forest in 1988, 3600
million ha were tropical rain forests, 40% being located in Latin
America (Koning et al., 1998).
The tropical rain forest (TRF) is one of the world’s richest
ecosystems in plant and animal diversity (Jetten, 1994) but is also
one that is threatened by human pressure (Park, 1992; Dale, 1997)
where LUCC is mainly driven by population increase (Sinha, 1997).
Land use and land cover (LUCC) change plays an important role in
this ecosystem when compared with natural events, and can
impact upon water quality, biodiversity, regional climate, and
ecosystem degradation (Koning et al., 1998).
The conversion of TRF to pasture and the subsequent succession of
pasture to secondary forest has a significant effect on canopy
2
cover, canopy height, species composition, and biodiversity
(Reiners et al., 1994). Increasing food demand and changes in land
management and land tenure have pushed forward the agricultural
frontier in many tropical countries. Subsistence farmers are
continuously being displaced and forced to clear new areas for
cultivation on steeper slopes.
Throughout history, the land surface has undergone changes in
use. However, over the last decades, these changes have not only
been rapid but also drastic. The forces behind land cover changes
include population growth, which leads to an increased demand for
food and, as result, agricultural expansion, but economic and
technological development are also important (Dale, 1997; FAO,
1997; Sinha, 1997). Most of these processes start at the micro-
level, but because of indiscriminate replication over large areas,
they soon become a global problem (Lambin, 1997). One of the
major expressions of LUCC is deforestation for agriculture and
grazing and to provide wood for housing and fuel (Sinha, 1997). In
most cases, deforestation is the result of complex chains of
causality, originating outside the forestry sector (Lambin, 1997).
Park (1992) reported that 23 million km2 of the earth’s surface was
covered by tropical forest and woodland. In the mid-80s, Latin
America accounted for about 11 million km2 of the world total area.
A historical review of deforestation conducted by Houghton (1994)
revealed that approximately 28% of the forests in Latin America
vanished between 1859 and 1985. During the same period,
croplands and pastures had increased from 3.5 million to 9.2
million km2. FAO reported that 0.15 million km2 of forest are lost
each year (FAO, 1997). Therefore, the area dedicated to agriculture
today is twice that 90 years ago, half of which is accounted for in
the tropics in the last 50 years (Houghton, 1994). Tropical
deforestation can be viewed as a growth process whereby the forest
conversion rate is regulated by the density of deforested areas; the
3
larger the deforested area, the more likely that deforestation will
continue to expand and spread outwards (Lambin, 1997).
This problem has been addressed by FAO’s Global Terrestrial
Observing System (GTOS). The following limitations to the
accurate prediction of LUCC have been found: the lack of data on
terrestrial ecosystems and on the changes occurring within them,
and the lack of technical capacity to identify operative solutions
(FAO, 1997). The knowledge and understanding of the processes
involved in LUCC are fragmented and, in many cases, restricted to
a given area (Watson, 1997). All approaches to the analysis of
LUCC yield only information on specific aspects of the process.
According to Lambin (1997), several essential questions must be
addressed when studying LUCC such as: Why does LUCC occur?
What variables contribute to these changes? Where does LUCC
occur? (In other words, which locations are affected by LUCC?)
When does LUCC occur? , and At what rate does LUCC take place?
1.2 LUCC: global impacts
Land conversion and intensification through human intervention
brings about changes in the ecosystem’s balance, generating a
response in the system (Dale, 1997). System alterations include
increased air temperatures, increased atmospheric CO2, release of
nitrogen to the atmosphere, soil salinity, soil compaction,
pronounced changes in erosion rates, and even soil degradation
and water contamination. In some cases recovery can take from
100 to 500 years, or these effects on the ecosystem may be
irreversible (Dale, 1997).
Variations in vegetation cover in hillside areas generate changes in
hydrological cycles, soil properties and atmospheric fluxes, and
4
meso-climatic conditions, as well as the loss of biodiversity.
Researches have shown that tropical forests play a major role in
regulating the earth’s climate, whereby the elimination of forests
can have enormous implications on local, regional and global
climate (O’Brien, 1996).
Several factors determine the impact of deforestation on the
climate. O’Brien (1996) argues that different controls affect the
climatic system in different ways, and that it is not easy to predict,
analytically, just how deforestation will change the climate.
Furthermore, the impact appears to vary depending on local
conditions, such as topography and proximity to oceans. As a
result, neither the magnitude nor the direction of climatic change
associated with deforestation can be considered definite. One
important factor that affects the climate is the change in
concentrations of atmospheric gases. Deforestation increases
atmospheric CO2 because of reduced sequestration of CO2 through
photosynthesis and emissions of CO2 through burning and
decomposition (Melillo et al., 1996; Tinker et al., 1996; Sinha,
1997). Methane (CH4) production, including the variation of
nitrous oxide (N2O), is another significant atmospheric flux that
occurs when forest or grass covers are converted to croplands
(Mosier et al., 1997; Sinha, 1997).
Pitman et al. (1993) evaluated different ways of assessing climate
response to deforestation in South America and Asia. By
comparing the output of six general circulation models (GCM), each
based on different scenarios, whereby forest areas were replaced by
different land uses. The short-term global effect, the global area
affected, and the climate response were all very significant.
Changes in climate variables involved increased air temperature
and reduced annual precipitation and annual evaporation.
5
In contrast to the effects occurring in lowland forests, deforestation
in hillside areas has drastic effects on soil stability and
hydrological cycles, in particular the increasing water runoff and
erosion as well as impacting nutrient cycles (Dale, 1997).
Dale (1997) argues that LUCC has a greater effect on ecological
variables compared with climatic change and LUCC has little to do
with climatic change or even with climate. Man will change the
land use, and especially land management practices, to adjust to
climatic change. The ecological impact of these adaptations is
therefore more significant. However, it can be argued that climatic
and hydrological factors affect, to a certain degree, ecological
factors. Therefore, even small changes in these factors will have a
significant impact on ecosystem ecology.
Both considerations are applicable to long-term LUCC change.
Changes will occur and one way to understand these changes is
through the application of simulation models.
1.3 A review of models for LUCC
Modelling activities on LUCC impacts have taken on more and
more importance in recent decades. Modelling has become an
important tool for understanding physical and hydrological
processes and impacts (Bronstert, 1999). The most common
reasons for applying simulation models are:
1. To monitor and assess potential impact. Impact of LUCC is
assessed by comparing model responses to the incorporation of
different scenarios of land cover (Mosier et al., 1997).
6
2. To conduct sensitivity analysis. To understand which
processes or landscape properties are the key determinants of
hydrological response to land use change (Johnes and Burt,
1990). Sensitivity analysis also illustrates the effects of LUCC
on single model variables or on groups of these variables, giving
the degree of change in the modelled response (LeBlanc et al.,
1997).
3. To predict and forecast. The modelling of hydrological fluxes
and the variation in ecosystem response to LUCC can be used
as a forecasting tool in the short term (Kirkby, 1990; Crohn,
1995; LeBlanc et al., 1997). Forecasting tools are used in the
assessment of water resources for flood risk and hazard.
4. Better understanding of the system. The need for a much
better understanding of the underlying driving forces behind
LUCC and its impact (Turner et al., 1994). Modelling is a mean
of rapid and inexpensive experimentation with model systems to
understand the relationships between variables, especially over
spatially heterogeneous landscapes.
5. Integrate processes. The interaction between climatological
and hydrological mechanisms in hillslope physically-based
models produces an integration of a number of diverse
processes in the disciplines of forest and land management
(Bonell, 1993). Models are a tool for the formal integration of
research applications across disciplines.
7
1.3.1 Methods for identifying the impact of LUCC
A benchmark land cover (LC) must be considered before LUCC can
be modelled. The LC is obtained by direct field observation or
remote sensing, which is clearly defined as a reference point. A
baseline inventory then identifies LC distributions in the recent
past. This initial scenario can be based either on physical or on
biological conditions. The initial hydro-climatological or biological
conditions produced by the reference scenario of LUCC are the
state budget or flux initial conditions, which will be used in the
comparison process.
LUCC can be studied at different scales; global, regional,
watershed, and plot scale (Kirkby, 1990; Dunn and Mackay, 1995;
Johnes, 1996; Leemans et al., 1996). Each scale has related
constraints, such as data availability, and the most appropriate
model type (Kirkby, 1990). Large-scale models and data are
usually integrated with small-scale models for parameterisation
and validation, for example, plot to watershed scale, watershed to
regional scale, regional to global scale (Johnes and Heathwaite,
1997; Bronstert, 1999).
The method most used to determine hydrological LUCC impact
involves the comparison of sequential land-cover maps, which
allows subtle changes to be detected.
Results of spatial statistical models of tropical deforestation show
that single-variable models based on landscape data from a
previous time period provide forecasts information of spatial
deforestation patterns and trends. Predicting the spatial pattern of
deforestation is therefore a much easier task than predicting future
rates of forest clearance (Lambin, 1997). Spatial statistical models
primarily identify location predictors of areas with the greatest
8
propensity for LUCC change. They do not predict when the change
will occur, they only identify proximate causes of LUCC change.
1.3.2 Strategies for evaluating hydrological fluxes in theassessment of LUCC impact
Numerous hydrological models have been adopted to estimate the
hydrological impact of LUCC. Studies on how LUCC affects the
hydrological environment must involve the response of these fluxes
to LUCC. Different approaches can be used to assess the
variations in hydrological fluxes, with respect to initial flux
conditions. Some approaches are summarised below:
1. Real-time LUCC scenarios. The model is used to integrate
scenarios with different land uses; the scenarios can be
generated from measured or remotely-sensed data or from land
use change models. Parameters for each land use must be
previously identified, to be included in the model. The model
includes land cover conversion, with real-time variations in the
scenario during the simulation period (Gustard and Wesselink,
1993).
2. Off-line LUCC scenarios. Remotely-sensed data or models are
used to generate a series of LUCC conditions which are applied
to the hydrological model off-line as alterations of the
equilibrium rather than transient change experiments (Frohn et
al., 1996).
9
1.4 LUCC: Issues and impacts in tropical forestenvironments outside Colombia
About half of all the world’s forests are in the Tropics. Though
there are many different types of tropical forest (moist forest, dry
forest lowland forest, upland forest) some of the characteristics of
this environment are precipitation greater than 1500 mm a year,
dense vegetation, an abundance of epiphytes, and dense under
stories of smaller trees, and shrubs, with harbouring high
biodiversity (Whitmore, 1998). This type of forest can be found in
America, Asia, Africa and Australia.
The main areas of remaining tropical rain forest, particularly
lowland forest are in Brazil and a number of other Latin American
countries, Congo and its neighbours, Indonesia, and Malaysia.
Tropical rain forest are known, as the world’s most productive
plant communities, with giant trees up to 60m in height supporting
thousands of other species of plants and animals.
Montane RainforestMontane rain forest differs in some characteristics growing at
higher elevations, where the climatic and topographic
characteristics are diverse and include extreme wetness
environment and steep slopes. Changes in environmental
characteristics change the forest appearance and structure.
Canopy height decreases with the elevation, reaching up to 35m in
the lower part of the montane zone, but only 9m above 3000m of
elevation (Whitmore, 1998). The structure is simpler than lowland
forest, with large buttresses, branches and epiphytes, which
become more numerous with increasing the elevation (Whitmore,
1998). Temperature can range from 10°C to 25°C according to
elevation (from 1000 to 3000 masl) and latitude. Climatic
conditions are also characterised by low ground level clouds,
10
particularly at different times during the day. The combination of
these creates a particular environment that is known as tropical
montane cloud forest (TMCF).
FAO estimated that for the period 1981-1990 the annual forest loss
in tropical highlands and mountains was 1.1%, much more than
other tropical forest, including lowland forest (Singh, 1994). The
main reason for the disappearance and degradation of this
environment was conversion to grazing land and temperate
vegetable cropping, trimber harvesting and wood production at
unsustainable rates (Bruijnzeel, 2000). Researchers have
expressed that the conversion of TMCF to other land uses could
result in significant declines in overall river flows (Brown et al.,
1996).
LUCC in tropical montane environments and its is becoming
important due to the deleterious nature of its consequences.
Tropical forests are considered a global climate regulator due to the
interaction between land surface processes and atmospheric and
climatic activities (Lambin, 1997).
Some consequences of tropical deforestation have been identified in
the literature including changes in surface and subsurface fluxes,
reduction in infiltration and water retention capacities, ecological
changes and loss of biodiversity, diminished cloud water
interception, increased runoff and thus soil erosion (Scatena and
Larsen, 1991; Brown et al., 1996; Scatena, 1998; Pounds et al.,
1999; Bruijnzeel, 2000; Sperling, 2000),
There are very few well monitored TMCFs in the world. Some of
these important studies areas are: Monte Verde Cloud Forest
reserve in Costa Rica (Pounds et al., 1999) where the forest
conversion to pasture and its effects have been studied by Pounds
11
et al. (1999), and Cusuco National Park in Honduras (Brown et al.
1996). Other notable examples are Sierra de Minas in Guatemala
(Brown et al., 1996) and Mt Kinabalu Sabah in Malaysia for forest
clearance for vegetable cropping. Luquillo Mountains in Puerto
Rico (Scatena, 1998), the Blue Montains in Jamaica (Tanner,
1977), and Talamanca in Costa Rica (Calvo, 1986) among others.
1.5 LUCC in Colombia: History and impacts in hillside areas
Several human activities have affected the vegetation cover in
Colombia, particularly LUCC for subsistence of the majority of the
population. Those activities are driven by socio-political factors
which control land tenancy and land use, and as a consequence
the environment is threatened. A brief historical review of
Colombian agrarian conflicts and land tenancy is presented to
understand the evolution of land use processes in the country.
Then some of the land use activities in the hillside of Colombia are
discussed to provide the context of the national problem and the
importance of a better understanding of hydrological process
affected by LUCC.
1.5.1 Historical review of land use change in Colombia
In Colombia as in most of the Latin American countries the
agrarian problems go back by the time of the great Spanish
conquests in the New world. The conquistadores were amply
rewarded by the Spanish Crown for their efforts on its behalf
through grants of land. The land was granted to the
conquistadores through the system of capitulaciones. Which were a
type of contract in which privileges over lands were granted by the
sovereign to the discoverers (Duff, 1968). By the end of eighteenth
12
century a Spanish system of large landholdings had replaced the
former small areas of communal Indian land tenure system. From
the Independence from the Spanish until the end of the nineteenth
century the land tenure changed ownership with owners,
increasing in number (because the land passed from the Spanish
Crown to the Colombian State, and then to the bourgeoisie
landholders currently in ownership (Duff, 1968).
In the XX Century after the First World War and up to 1920’s the
Colombian bourgeoisie efforts went to build the basis of an
industrial framework for an international open market of
agricultural products. The commercial activities were based in the
latifundio (large extension of land) created through two centuries of
the New Great Colombia, and to take the advantage of increasing
international prices in agricultural products (Bejarano, 1977). The
tendency was for property concentration and land monopoly,
including land holding of good land without production. Those
activities changed the way that the land has been used. In that
time the working population is divided into the workers of the
incipient industrial process in Colombia in the cities and the
farmers that have been attached to small lands and isolated from
the market for the prevailing commercial conditions imposed by the
monopoly (Buitrago, 1977). As consequence, on the one hand the
increasing populations in the cities where the factories improve the
production, and for the other hand, the remaining part of the
landholdings who were expropriated from their own land, were
pushed to colonise new lands meanwhile some of them remain in
the land but becomes land workers (without land) in the big
properties of terratenientes. In this time the government gave the
chance to landowners to make official the land tenancy by owning
property titles through new laws. Using the guise of ‘giving the
owning title to the small farmer’, but in reality they wanted to make
official their big properties which they had appropriated before (Ley
13
de tierras 200 of 1936), where the new structure of land property
were the couple ‘big property – small property’ (latifundio and
minifundio) (Cartier, 1990).
After the Second World War several social and political problems
kept the peoples attention, hiding the agricultural problem. In that
time, the Currie Mission, was assigned to build a program for
Colombia’s development. They highlighted the problem that large
flat extensions allocated on fertile valleys were used for cattle while
people in the hillsides fight for a piece of land to crop their own
food, which meant that the best land was used in the wrong way
(cattle instead of intensive agriculture). In the fifties and sixties the
fast rhythm of agricultural industrialisation and mechanisation,
displaced the field workers from the big farms to the hills, adding
to these hill areas more necessities, been the low yield production a
characteristic these land. This changes the land tenancy stage
from small production to commercial agriculture (as an industry)
and traditional agriculture (survive crops) (Banco de la República,
1951).
As a consequence of those processes the workers displaced from
farming became unemployed in the cities and hillside areas
increasing the social and agricultural problems. The bourgeoisie
hiding behind the government realised the problem and in the
sixties and seventies a project of law (The Agrarian Reform, Law
135 of 1961) was proposed. This law with the face of ‘justness for
the people’ allocated the land redistribution with equal
opportunities, offering the unoccupied lands and the worse types of
land to the population that did not have land (because the better
lands were already occupied by the terratenientes). That law of
land had the purpose to, stop the people that started to invade the
bourgeoisie productive lands. In this way, the agricultural labour
force that were not absorbed by the small industries (because of
14
the mechanisation of agriculture) were neutralised temporally
(Bejarano, 1977). The land problem remained in the country,
displacing the population to colonising new lands, logging the
forest, or on occasion practising slash and burn to increase the soil
fertility for a few years, and then moving to a new forest area and
repeating the procedure.
In the last decades, people that occupied unproductive land were
forced to move to the forest and agricultural frontiers, colonising
and deforesting new land, to supply the necessary food to survive.
However between 1978 and 1992, the proportion of the rural
population in extreme poverty (the countryside farmers) declined
fairly slowly (from 38% to 31%) (The World Bank, 1996), due to the
socio-political conflicts between gerrilla and paramilitares, which
both razing several towns in the countryside and killing people with
the excuse that they were collaborators with the enemy. As
consequence those farmers escape from the countryside to main
cities.
In addition, the most recent problem (illicit cropping of coca leaf
and amapola (opium poppy), as a fast solution to economic
problems), in combination with narco-economy, paramilitaries and
increasing violence, become others factors adding to the problems
of land use change. Also the growth of populations and their
migration to marginal areas, increased the pressure on the forest,
changing the forest to land with low agricultural potential,
producing environmental impacts such as degrading the soil,
natural resources and, vegetation followed by abandonment.
(Fajardo, 1996).
Nowadays these problems are still hitting most of the poor
population and the effects of bad agrarian practices are appearing
markedly on the hillsides areas, characterised by the high density
15
of minifundios (small parcel less than 3 ha), increasing land
degradation and ecosystem instability.
No much information in the country has been published about
deforestation rates. The first agricultural census carried out by the
“Departamento Administrativo Nacional de Estadística” DANE,
which is the national institution with the responsibility of produce
this type of information, was in 1960, covering small parts of the
country, only for the productive area, without including large areas
such as savannas, forest and deserts. The most recent agricultural
census was in 1995, covering less of a half of the country. A
Comparison from those sources is included in table 1.1.
Year 1960 % of area 1995 % of area
Census area 27’337,827 100 51’865,996 100
Agriculture 5’047,088 18.4 4’430,018 8.5
Pasture 14’605,954 53.4 35’527,873 68.5
Forest 6’387,024 23.4 10’088,071 19.4
Other uses 1’297,751 4.7 1’820,034 3.5
Table 1.1 Land use census data comparison for Colombia, between 1960 and
1995. (area values in ha., area total of the country 114’174,887 ha).
Data from table 1.1 show that despite the census area in the 1995
census being twice as much as the 1960 area, the area used for
agricultural exploitation decreases over time. The area used in
pasture increase more than twice, meanwhile the forest area
increases in almost 4 million ha (these reults largely a function of
the different census areas). It is clear that in 35 years 24.5 million
ha. were incorporated to the productive system, of which 85% was
for pasture (20.92 million ha), and just 3.7 million ha were
identified as forest. As the new area came from wild and natural
forest as well the native savannas, the deforestation activities were
significant. The deforested area used in illicit crops is not counted
16
in these assessments, but the rate of deforestation for this activity
is estimated reach up to 60,000 ha per year. Winograd (1995)
reports that the deforestation rate between 1980 to 1990 reached
up to 60% more than previous decade, meanwhile the agriculture
area decrease in a rate of –0.5 % a year and the areas used in
pastures increase in +3.4 % a year.
1.5.2 The hydrological impacts of LUCC in Colombia
Colombia is one of the richest countries in hydrological resources
with abundant rivers and natural resources, which are well
distributed geographically. Colombia occupies the fourth place
after Soviet Union, Canada and Brazil in hydrological richness,
with more than 88% of the total area (1’141,748 km2) with
precipitation over 2000 mm a year, and an average of 3000 mm a
year. The mean evaporation in Colombia is 1150 mm a year, and
the total runoff could average 2,112 km3, which is 67 m3 s-1
approximately (annual values for the whole country area) (Marin-
Ramirez, 1992).
During the last decades water resources have become a problem,
with watershed management in the Andean hillside areas,
producing ecological, social and economical damages due mainly to
population growth, changes in vegetation cover, industrial
development and land use change (Marin-Ramirez, 1992). The
obvious consequences that can be mentioned are, among others:
loss of biodiversity in relation to the rapid loss of natural forest
cover; loss of wild relatives of useful crop species; soil instability
and landslides. Soil erosion, principally loss of topsoil due to water
erosion. Nutrient loss through leaching, with monocultuves and
badly-managed sown pastures. Water quality issues, associated
17
with high sediment load in head waters are also a growing problem
(CIAT-Hillsides Program, 1994).
The World Bank estimated that 45% of the rural Colombian
population were predominantly in hillside areas in the beginning of
the 1990’s, with 23% being the indigenous population. Rural
impoverishment has increased for those areas relative to the
country as a whole (Cepal, 1990).
Poor agricultural practices on the hillsides are used extensively
such as fallow rotation systems in which forest or bush are cleared
for cropping, and then are returned to pasture or bush fallow once
yields decline to a level that is not economically useful.
Deforestation, overgrazing and agricultural activities are also
causes of degradation in the hillside agro-ecosystem.
Environmental degradation in the hillsides has serious implications
not only for the viability of agricultural production in the ecosystem
itself, but for “downstream” lowland agriculture and coastal
ecosystems affected by soil erosion and agrochemical pollution in
the uplands. Soil erosion, sedimentation and major land
degradation caused by deforestation and cropping without use of
soil conservation practices affects watercourses originating in the
hillsides. The most irreversible and potentially damaging with
major social cost caused by hillside environmental degradation, is
the loss of biodiversity due to the disappearance of montane forest
which amounts to 32% of the forest area in the Colombian Andean
Region. The rate of deforestation in hillsides is higher than in the
lowlands. Causing a loss of 90% of the original montane forest
cover by 1990 (CIAT-Hillsides Program, 1994). Montane forest has
very high biodiversity, which is considered important to conserving
wild crop genetic resources in-situ. In ecosystems where the land
use is intensive the most important environmental degradation is
18
the excessive use of agrochemicals which is a characteristic of
agricultural intensification, causing soil and groundwater pollution
(CIAT-Hillsides Program, 1994).
A CIAT study carried out on the hillsides in the Andean Region in
Colombia, was centred in the Rio Ovejas watershed in the Cauca
Department. This watershed covers 100,000 ha. and encompasses
a diverse range of Andean hillside systems ranging from indigenous
slash and burn cultivation to peri-urban, high-input horticulture,
and includes CIAT commodities (Knapp and Buitrago, 1994).
Consequently, the assessment of the location and extent of the
erosion problem in the hillsides was an additional activity
undertaken by CIAT in the study area, as well as the ex-ante
impact assessment of land use change and development of a
diagnostic simulation model of alternative technological
interventions. The model considered impact on soil erosion,
nutrient loss, crop productivity and water quality (Knapp and
Buitrago, 1994).
The relationship between soil erosion and productivity remains
poorly researched and little understood in tropical soils. It is
identified as a need for research focused on improving
methodologies for characterising the extent and cost of soil
degradation. In addition systematising the available data requires
regional collaboration, due to the diversity of the hillside land use
classes found in the country.
To improve crop productivity and forage availability, to enhance
erosion control and soil physical rooting conditions, and to
increase water infiltration, water-holding capacity, and nutrient
retention of the soil, the incorporation into hillside production
systems of practices for soil conservation and regeneration are
being energetically promoted (Knapp and Buitrago, 1994).
19
Knapp and Buitrago (1994) also points out that while farmers
consider the monetary benefits of erosion control, such as yield
increases, they are unlikely to consider non-monetary benefits
such as soil resilience, or downstream benefits which accrue to
others.
Hillside agro-ecosystems are a mosaic of diverse micro-edapho-
climatic regimes, user circumstances and cultures. In any one-
area the results of technological innovation will be location-specific.
An essential task is to develop a replicable approach to innovation,
based on strategic understanding of how to intervene in the hillside
agro-ecosystem and how to make transitions to ecologically-sound
and economically-viable alternatives, acceptable to users.
Determining why some technological options are more acceptable
to farmers than others, and the trade-off between production and
conservation objectives this involves, requires technology testing
which is embedded in a community based participatory framework
(Knapp and Beltran, 1994).
The hillside approach is focused on the effects of soil degradation
that involve diagnostic research to better identify problems and set
priorities amongst them with respect to biophysical and economic
aspects of soil degradation due to agricultural practices and
catchment management. In addition, the design of decision-
support systems incorporating different types of models, including
knowledge-based models drawing on indigenous technical
knowledge and research results that can be introduced into models
to facilitate the understand of LUCC effects in the watershed (CIAT-
Hillsides Program, 1994).
20
1.6 Structure of the thesis
Physical hydrological fluxes are dynamically modelled from the
atmospheric interface to the soil bedrock interface. A 1D dynamic
hydrological model was initially developed at the plot scale for each
type of land cover. The 1D model is parameterised and validated
on the basis of data from hydrological stations in pasture, primary
and secondary forest. Lessons learned from the production and
sensitivity analysis of this model were applied in the development
of a 2.5D distributed hydrological model, integrated within a
Geographic Information Systems (GIS). This was then applied to
understanding the impact of LUCC at the catchment scale. A
sensitivity analysis of the 2.5D model was performed to identify
hydrological flux variation with land use change to determine key
variables of the ecosystem that are affected by different spatial
patterns of LUCC. Five different scenarios of LUCC were used
within the analysis, to assess the hydrological flux sensitivity and
to determine the most sensitive areas in the studied catchment.
Chapter 1 introduces the topic of LUCC in this thesis. First a
discussion about the impact of LUCC in general terms and
additional information about LUCC modelling is provided,
including methods and tools. Then the LUCC impact on tropical
montane forest is discussed in a global context. Subsequently, the
development of LUCC modelling in Colombia are also presented,
and provides brief background of the LUCC in Colombia,
historically and the actual situation of the hillsides research, and
finally the thesis structure is presented. Chapter 2 presents the
literature review of hydrological models applied to LUCC. The
strategy for estimating LUCC is discussed. Then the literature
review of the hydrological models is discussed: characteristics,
classification, types and results, and also a brief review of some of
the best known contemporary hydrological models with their main
21
features. Also hydrological models in tropical montane
environments are reviewed and finally, understanding the problem
and the research approach are presented in the thesis and the
thesis objectives, main goals, and the obtained achievements
discussed.
Chapter 3 describes the methodology used in this thesis. This
chapter has two marked sections: the first is related to the
collection of the information for modelling, the second is related
with the construction of the hydrologically-based model. Initially
the structure of the chapter and the study area are presented. The
research and experimental strategies are provided and detailed
description of the scenarios of LUCC used in combination with the
hydrologically-based model are given. Then the fieldwork
methodology is discussed for plot and catchment scale studies; the
installation of hydrological stations, and the field methods used for
the collection of data are illustrated. The data collected in the field
are presented and additional data used for model parameterisation,
experimentation and for model verification and also for validation
are discussed. Secondly the modelling aspects are discussed. This
section describes the development of the 1D and 2.5D models,
together with a description of the following sub-model components:
solar radiation, energy balance, evaporation, canopy storage and
interception, infiltration, soil water hydrology, overland flow, and
erosion. Each component is explained in detail and source
equations, flow diagrams, and data requirements are indicated.
The inter-relationship between components and information flow is
also indicated. After describing the sub-model, model performance
and initial conditions are explained. Then model integration with
Geographic Information Systems (GIS) is also described.
In Chapter 4 the model results are presented. 1D and 2.5D model
results are shown to discuss the model characteristics and some
22
implication of landscape properties on the hydrological response.
Then 1D model parameterisation and sensitivity analysis is
discussed, and subsequently 2.5D model sensitivity analysis for
overland flow and erosion is shown; the relationship between those
variables and the topographic variables is evaluated. A summary
of TMCF sensitivity to LUCC is presented in terms of overland flow
and erosion sensitivity. Finally, validation of some output variables
is carried out to evaluate the model goodness of fit.
Chapter 5 gives the summary and the conclusions, the objectives
evaluation and the achievements, including the recommendations
for estimating hydrologically sensitive areas to LUCC for the TMCF
environments, and then the conclusions are drawn with further
model applications and future research possibilities elaborated.
23
Chapter II Literature review of hydrological models appliedto LUCC impacts research
2.1 Structure of this chapter
This chapter presents the literature review of hydrological models,
which begins with the general concepts used in the modelling
activities, particularly with the issues related to hydrologically-
based simulations. Then a classification of these models is
presented, including the importance of spatial variability as a
characteristic of modelling the surface water fluxes. A complete
review of the existing commonly used hydrological models related
to LUCC impact is presented, and finally the main objective and
the specific aims of this thesis are numerated.
2.2 General concepts of hydrological models
Hydrological models aim for simplicity by selecting a system’s
fundamental aspects at the expense of incidental detail (Anderson
and Burt, 1985). A number of alternative techniques and
modelling approaches have been developed.
The first integrated hydrological model, called the Stanford
Watershed Model (Singh, 1995), was reported in the literature in
1966 by Crawford and Linsley. During the following decades,
hydrological modelling improved significantly because of advances
in technology and computer hardware.
Better hydrological models are becoming available with these
technological advances and the continuous improvement in
24
modelling strategies, such as inclusion of GIS, remote sensing or
cellular automata (MacMillan et al., 1993; Beven and Moore, 1994;
Robin et al., 1995). Many of these methods are used in
contemporary watershed models, such as TOPMODEL (Beven et
al., 1995); KINEROS, a kinematic runoff and erosion model
developed by Rovey et al. (1977) and described by Smith et al.
(1995), and TOPOG_IRM (CSIRO, 1993).
Many of the latest generation hydrological models use GIS, but, in
many cases, GIS and environmental models are not well integrated,
just used together. GISs are frequently used as post-processors to
display and further analyse model results. In turn, modelling
approaches directly built into a GIS appear rather simple and
restrictive (Fedra, 1993). Dangermond (1993) indicates that the
tendency for integration is to use specialised software systems.
“Such powerful tools without well distributed data are, at best
expensive interpolation tools and, at worst subject to GIGO
(garbage in-garbage out)” (Fedra, 1993). One of the main
restrictions on good spatial (GIS) modelling is a lack of good,
spatially detailed hydrological parameters for model
parameterisation and validation.
2.3 A general classification of hydrological models
Models can be characterised by the type of relations used within
the routines. The relationship between real and model processes
can be represented either empirically or physically.
1. Empirical models. Model relationships are based on empirical
data, not necessarily on physical processes. These models tend
to have a high predictive ability but their physical explanatory
power is often low. They are sometimes called “black box” or
25
“input/output” models. These terms are usually applied to
those models whose internal operation does not aim to directly
represent “real” operative processes, even at an abstract
mathematical level (Kirkby et al., 1993). Successful
applications of this strategy include the unit hydrograph,
extreme frequency analysis, regression analysis, and real time
forecasting models (Anderson and Burt, 1985). Statistical
analysis faces several methodological and interpretative
difficulties, such as measuring complex dependent variables,
and spatial aggregation of data in large units. The existence of a
statistically significant association does not establish a causal
relationship. Moreover, a regression model that fits well in the
region for which it was designed might not function well in other
regions, because it should not be transferred beyond the
physical limits for which it was developed, parameterised and
calibrated.
2. Physically-based models. These models, based on physical
processes, are modelled on the understanding of physical
mechanisms and often make large demands in terms of
computational time and data requirements. Nevertheless, such
models offer increased explanatory and experimental power.
However, because of the higher number of assumptions that are
necessary, their predictive capacity is often equal or worse than
that of empirical models. Beven (1989) argued that highly
complex, physically-based models are possible at smaller scales.
However, larger-scale models must be simple to allow
parameterisation. Woolhiser (1996) pointed out that simpler
models are often more accurate than physically-complex
models, but are difficult to scale up to larger watersheds.
Parameter generalisation within the watershed involves simple
representations of main model elements. Several variables such
as soil characteristics which are important at reduced scales for
26
detailed studies are also important at the watershed level,
increasing model complexity while not necessarily adding
precision to the results.
2.4 Handing spatial variability in hydrological models
Several approaches to represent spatial variability within a
watershed exist. These approaches can be classified as:
1. Lumped modelling, expressed by ordinary differential
equations that describe simple hydraulic laws. These models
do not take into account the spatial variability of processes,
inputs, boundary conditions, or the system’s geometric
characteristics. Instead, a single value for properties and
parameters is applied to the entire watershed. Some examples
are HEC-1 (Hydrologic Engineering Center, 1981) described by
Feldman (1995), RORB (Laurenson and Mein, 1995), and
SSARR (USA Army Engineer, 1972) described by Speers (1995).
2. Distributed modelling, which explicitly accounts for the spatial
variability of processes, inputs, boundary conditions and system
characteristics. The spatial distribution of features and their
spatial inter-relationships are especially important to explaining
physical processes within the watershed. Examples are SHE
(Abbott et al., 1989) described by Bathurst et al. (1995), SWMM
(Metcalf et al., 1971) as described by Huber (1995).
Models can also be classified according to the type of equation used
and the resulting output. Model results can be a singular, or a
population of answers. Processes can be described either by
deterministic or stochastic equations. Deterministic models
have just one possible outcome, whilst stochastic models have a
27
population of answers. In most cases both types of equations
occur within the same model. However, in the cases when the
relevant information for parameterisation is not available, some
processes are better modelled by stochastic equations that could
give an approximation for modelling purposes.
2.5 A Review of hydrological models related to LUCC impact
There are several hydrological models that have been created for
particular purposes or environments. The models have different
abilities, characteristics and type of results, including resolutions
in time and space. Some of the most widely used models are
discussed here, identifying some of their important features related
to the subject of this thesis, and the reasons that the models are
not used in this thesis.
SHE/SHESED
The SHE/SHESED combination is a physically based, spatially
distributed modelling system for water flow and sediment transport
to be applied at a catchment scale. The SHESED model was
developed in the University of Newcastle upon Tyne, UK, and is
based on the SHE (Systeme Hydrologique Europeen) model which
was developed by international collaboration between groups in the
UK, Denmark, and France. SHESED is used to investigate land
management especially the prediction of LUCC and climate change
impacts. SHE was designed as a flexible modelling system,
encompassing several levels of complexity, consisting of sub-
components for evapo-transpiration and interception, overland and
channel flow, unsaturated zone flow, saturated zone flow,
snowmelt and channel/surface aquifer exchange. The SHE model
28
is driven by meteorological inputs and provides inputs to the
sediment transport component (Bathurst et al., 1995).
The interception sub-component is an adaptation of the Rutter
model (1971), and the evapo-transpiration is based on the Penman-
Monteith equation. Some sub-components require more
parameters and input information than are not available for
Tambito such as the atmospheric component, sediment yield and
transport of material within the channels component, as well as
soil matric suction, raindrop impact amongst others (see section
3.6). Also the model has a number of simulation routines that are
not useful for this study. Nevertheless, this model is one of the
investigated models that could be appropriate for this thesis, but
its complexity is too high for the application intended here.
In addition to the lack of information available for parameterisation
and validation of the SHE model, most of the literature consulted
reported the use of the model for short time simulations (days)
providing good simulation results (Wicks and Bathutst, 1996), or
for bigger spatial resolution up to 4000 m of pixel size (Refsgaard,
1997). However, Wicks and Bathust, (1996) used the model for two
small agricultural catchments (5.1 and 6.4 ha) in Iowa, with good
reproduction of the observed temporal variations in sediment yield.
In contrast Refsgaard (1997) applied this model to a catchment of
440 km2 in Denmark, for which calibration and validation
processes were carried out splitting the catchment in seven
sections, and to producing better results at a pixel size resolution
of 500m.
29
SHETRAN
The SHETRAN system was developed by the Water Resource
Systems Research Laboratory (UK), based also on the SHE
(Systeme Hydrologique Europeen). SHETRAN is a 3D, coupled
of 0.6X0.6m pixel size). The vegetation texture patterns appears on
the panchromatic photographs as several granules or flat areas,
which produce different textures due to the size of the shadows
and forest trees, showing where the vegetation is forest or
grassland. According to these patterns on the panchromatic
54
photographs, they were compared with the NDVI image to select
the signature areas of the LU used in the classification.
Four reflectance classes of vegetation were distinguished from the
satellite image: two types of forest, grassland and cloud (table 3.1).
LUC class Clouds Primaryforest
Secondaryforest
Grassland
NDVI 0.33 0.59 0.8 0.72
Table 3.1 Average NDVI values for classification of land use classes
3.4.2 Scenario descriptions
The LUCC scenarios are described as follows:
1. Scenario 1 (SC1): The LUCC pattern derived from a cellular
automata, as designed and implemented by Mulligan et al.
(2000). This scenario simulates the conversion of forest to
pasture as spreading from roads and agricultural frontiers, in
an epidemiological fashion or a propagation wave through 22
iterations in this catchment. Figure 3.4 shows the trend and
the area by iteration of land conversion for twenty-two
iterations, and an example of an iteration of this scenario is in
Figure 3.5. Figure A1.1 (Appendix 1) shows the spreading
pattern over the watershed.
2. Scenario 2 (SC2): The conversion pattern is carried out by
applying a fixed horizontal distance from the river channels
taken from ‘Instituto Geográfico Agustín Codazzi’ (IGAC)
cartography to convert forest areas to grassland in an uphill
direction. This pattern was created to understand the effect of
forest buffers and hydrological connectivity on secondary flow
55
path to the major rivers. Deforestation is produced on both
sides of all rivers in 50-m. horizontal distance increments by
iteration. Each 50m is recognised as a single class for a total of
18 classes, and incrementing the same distance for each
iteration until the whole watershed is deforested; this is reached
in iteration 18. Figure 3.4 shows the LUCC pattern in this
scenario and deforested area by iteration, and an example of an
iteration of this scenario is in Figure 3.6. Figure A1.2 (Appendix
1) shows the spreading pattern over the watershed.
Scenario 1 (SC1)
0
500
10001 4 7 10 13 16 19 22Iteration
Are
a (H
ecta
res)
Grassland
Forest
Scenario 2 (SC2)
0
500
1000
1 3 5 7 9 11 13 15 17Iteration
Are
a (H
ecta
res)
Grassland
Forest
Scenario 3 (SC3)
0
500
1000
1 3 5 7 9 11
13
15
17
Iteration
Are
a (
Hec
tare
s)
Grassland
Forest
Scenatio 4 (SC4)
0
500
1000
1 3 5 7 9 11 13 15
Iteration
Are
a (H
ecta
res)
Grassland
Forest
Scenario 5 (SC5)
0
500
1000
1 3 5 7 9 11 13 15Iteration
Are
a (H
ecta
res)
Grassland
Forest
Figure 3.4 Transition of LUC by scenarios(a) Scenario 1. Forest conversion to pastures fromcellular automata. (b) Scenario 2 Forestconversion with a fixed distance from riverchannels (c) Scenario 3. Forest conversion with afixed distance toward river channels. (d) Scenario4. Forest conversion with a fixed altitudinaldistance from lower point in up hill direction. (e)Scenario 5. Forest conversion with a fixedaltitudinal distance from higher point in down hilldirection.
(a)
(b) (c)
(d) (e)
56
Figure 3.5 An example of an iteration for SC1
57
Figure 3.6 An example of an iteration for SC2
58
3. Scenario 3 (SC3): The conversion pattern from forest to pasture
is carried out by applying a 50-m fixed horizontal distance from
the watershed boundary toward river channels in a downhill
direction. Deforestation advances in each iteration by the same
distance (50m.), producing complete deforestation by the 18th
iteration. This scenario combined with SC2 would help to
identify whether or not the direction of deforestation relative to
the rivers could affect the catchment hydrological response.
Figure 3.4 shows the LUCC pattern for this scenario, and an
example of an iteration of this scenario is in Figure 3.7. Figure
A1.3 (Appendix 1) shows the spreading pattern over the
watershed.
SC2 and SC3 are related to deforestation from and toward river
channels because one of the purposes of this experiment is identify
the importance of forested areas close to river channels.
4. Scenario 4 (SC4): Conversion pattern from forest to pasture is
carried out by applying deforestation in 100-m fixed increments
of altitude, in an uphill direction, from the lower to the higher
points of the catchment. Within 15 iterations the catchment is
completely deforested. Deforestation advances with 100-m of
altitude until it reaches the highest points of the watershed.
This scenario was built to identify the elevation effects on the
catchment hydrological response to LUCC conversion. These
may be important since most of the climatic variables change
with elevation in this catchment. Figure 3.4 shows the LUCC
pattern for this scenario, and an example of an iteration of this
scenario is in Figure 3.8. Figure A1.4 (Appendix 1) shows the
spreading pattern over the watershed surface.
59
Figure 3.7 An example of an iteration for SC3
60
Figure 3.8 An example of an iteration for SC4
61
5. Scenario 5 (SC5): Conversion pattern from forest to pasture is
carried out by applying deforestation in 100-m. fixed elevation
distance, from higher to lower points of altitude, in a downhill
direction, through 15 iterations. Higher points in the watershed
are deforested in the first iterations. Figure 3.4 shows the
LUCC pattern for this scenario, and an example of an iteration
of this scenario is in Figure 3.9. Figure A1.5 (Appendix 1)
shows the spreading pattern on the watershed surface.
SC4 and SC5 show the pattern of deforestation related to elevation
change. Deforestation from and towards the highest points are
included in the analysis with the purpose of identifying altitudinal
whether the order of deforestation (top to bottom or bottom to top)
is significant.
Iterations within scenarios are not related to any concept of time
taken for LUCC. Iterations are steps in a pattern of change in
which forest is converted to pasture within the terms that the
scenarios define.
Table 3.2 summarises the extent of deforested areas by scenario
and by iteration. Values in table 3.2 were extracted using GIS
utilities developed for the study area, which is widely explained in
following chapter.
62
Table 3.2 Rates of deforestation per iteration of the different scenarios (values in ha.)
among others), shows the high variability in both stomatal
response (variation between 0.02 to 0.001 m s-1) and biological
diversity. On the assumption that the bulk physiological
conductance is equal to the conductance of all stomata acting in
parallel (big leaf), the estimation of gs as a product of ‘scaling up’
process, is uncertain due to the landscape biodiversity. The same
assumption was tested earlier by Shuttleworth, (1978) finding a
fairly close agreement. Those difficulties were also identified by
Dolman et al. (1990) in their observations. In addition, Veen and
Dolman (1989) highlight the spatial and temporal variability of gs in
the tropical forest canopy, who suggest the use of stratified
sampling procedures to create sub-layers for modelling gs.
Roberts et al. (1990) using a stratified sampling procedure in the
Amazon tropical rain forest, demonstrated that there is a strong
relationship between gs and the solar radiation, where the
emergent trees had the highest gs (which declined rapidly during
the afternoon), whilst vegetation close to the ground had lower gs,
(with little variation during the day). Dolman et al. (1990)
emphasised that the derived variation in transpiration could be
attributed up to 80% of the gs variation, been accounted for the
variability in solar radiation, which on time varies spatially and
temporally within an area, as it occurs in Tambito area, as has
105
been seen in the previous section. Under such conditions, Dolman
et al. (1990) highlighted that only solar radiation shows the main
variation that, in turn, drives the diurnal variation in temperature
and humidity deficits.
The atmospheric conductance is a function of the wind speed, the
aerodynamic roughness of the vegetation and the stability of the
atmosphere (Bonell and Balek, 1993). Within this context, Wilson
(1989) also argued that using the Penman-Monteith equation for
modelling forest transpiration Et, the ga has only limited sensitivity
to the formulae precision and is data limited.
Daily evapotranspiration response for Tambito was evaluated using
the Penman-Monteith equation (eq. 3.12), varying net solar
radiation, atmospheric and stomatal conductance, in order to
identify the equation’s sensitivity to the variation to these
parameters. Net solar radiation was varied between 0 to 400 W
m-2; atmospheric and stomatal conductance were varied from
0.001 to 0.9 m s-1, covering the range conditions which might be
present in the Tambito area (Letts, 2000).
Figures 3.27 shows the variation of evapotranspiration to changes
in net solar radiation. It is clear from Figure 3.27 that
evapotranspiration is highly sensitive to solar radiation with evapo-
transpiration nearly doubling for a doubling of net solar radiation.
106
Figure 3.28 shows the evapotranspiration change with atmospheric
and stomatal conductance and indicates that except at very low
stomatal and aerodynamic conductances (high resistances) the
impact of these variables on evapo-transpiration is low. This would
indicate that whilst the evaporation module must include spatially
variable solar radiation fluxes, these is little need to include
aerodynamic and stomatal parameters particularly since these is
no way to provide spatially varying measurements for these.
Figure 3.27 Evapotranspiration variation with respect to Net Solar Radiation using Penman-Monteith equation
Evapotranspiration variation with net solar radiation using Penman-Monteith equation
0
1
2
3
4
5
0 100 200 300 400 500
Incoming net solar radiation (W/m2)
Evap
otra
nspi
ratio
n (m
m/d
ay)
Figure 3.28 Evaporation variation with respect to atmospheric and stomatal conductance
Evapotranspiration variation w ith stomata and atmospheric conductance
0
1
2
3
4
5
6
0 0.2 0.4 0.6 0.8 1
Stomata and atmospheric conductance (m s-1)
Evap
orat
ion
(mm
/day
)
Atmospheric
Stomata
107
Thus, there is a need to implement a simplified evaporation model
based on the most sensitive variables. On de basis of the previous
discussion, the evaporation model used in this thesis is a simple
evaporation model based on net solar radiation, which was used by
Mulligan (1996) with good results.
Average annual air temperature in the study area ranges from 14°C
to 25°C. Within this range of temperature, latent heat ranges from
2.467 to 2.443 MJ kg-1; in extreme cases, the difference in
evaporation between these latent heat values is no more than 0.2
mm per day (computed with extreme temperatures recorded with
the weather station in the last 5 months in 1997 at the grass plot,
n=1656). For reasons of simplification, the latent heat parameter
(λ) was assigned an average value 2.445 MJ kg-1 in both 1D and
2.5D models. Latent heat variation with the elevation through the
catchment varies less than 2% across the catchment (temperatures
at the top of the catchment), which would produce an evaporation
variation less than 1%.
As was mentioned in the model presentation at the introduction of
this chapter, normal weather conditions in TMCF environments are
mostly near saturation (92% humidity). This means that
vegetation in the Tambito watershed is unlikely to suffer from
water stress, so net radiation rather than water availability is the
main control on evapotranspiration. The evaporation module can
therefore be simplified to the evaporation over free water as being
the only function of energy available for evaporation from the
energy balance, as follows:
mm h-1 Eq 3.13
=
λρn
p
SE
108
where Ep is the potential evaporation; Sn, net solar
radiation [MJ hour-1] is used because it is the majority of the
energy available and net radiation is not measured; λ, latent heat
as previously calculated; and ρ, water density (in this case is
assumed as 1.0 g cm-3) (Oke, 1987). The model is constrained by
the surface area of water available for evaporation in the canopy.
Net radiation, Sn, is the most important variable for evaporative
energy. In hydrology studies, all the energy available for
evaporation is assumed to be accessible by the plant canopy, and
water vapour first diffuses from the leaves, against surface (or
stomatal) resistance, rs (Maidment, 1993), and then out into the
atmosphere, against an aerodynamic resistance. Meanwhile,
sensible heat, which originates outside rather than inside the
leaves, only has to diffuse upward against aerodynamic resistance
ra.
To include the transpiration process in the model, in the cases
when the canopy is dry, the actual evaporation is calculated from
the potential evaporation applied to the soil moisture availability,
without taking into account any plant physiological activity in the
process. The design of the potential evapotranspiration sub-model
is illustrated in Figure 3.29.
3.6.6 Canopy storage, interception and throughfall
By it’s nature, vegetation has a large influence on hydrological
processes in a TMCF ecosystem (Jetten, 1994; Hafkenscheid,
2000). Trees intercept most of the rainfall, part of which
evaporates (Rutter et al., 1971; Rutter, 1975; Jetten, 1994), so that
the water flux that reaches the soil surface is determined by both
rainfall intensity and the drainage of the canopy (Jetten, 1994).
109
Therefore, when describing the water balance of a forested
watershed, interception cannot be treated as a fraction that is
simply subtracted from the rainfall, because the vegetation is
complex.
Additionally, the microclimate within the canopy produces a
particular set of conditions affecting the evaporation from the
intercepted water. Consequently, water fluxes associated with
wetting and drying processes must be quantified. There are
empirical approaches, which use regression coefficients to estimate
the percent of water loss in the interception process. There are
also physically-based models, which can be adapted for particular
forest conditions (Rutter et al., 1971, 1975). A stochastic
alternative was proposed by Calder (1986) to model the rainfall
interception, which relates the mean number of raindrops retained
on elemental surface areas to the mean number of raindrop strikes
per element, using the Poisson probabilistic distribution. A
simplification of the Rutter interception model was developed by
Gash (1979), focusing on rainfall occurring in a series of discrete
Input
Net radiation at ground level
CalculatedAvailable energy, Potential evaporation
for free water
Parameter
Latent heat
OutputPotential evapotranspiration
Figure 3.29 Flow diagram for potential evaporation.
110
storms, each of which comprises a period of wetting up, a period of
saturation an a period of drying out to empty the canopy storage.
The models of rainfall interception discussed have been applied on
tropical forest environments in combination with the Penman-
Monteith equation to estimate the rainfall interception loss, with
acceptable results. For example, Bruijnzeel and Wiersum (1987)
used the Gash rainfall interception model in West Java, Indonesia.
In contrast, Calder et al. (1986) used the Rutter interception model
in West Java without good results; the model predicted only 50% of
measured interception. Lloyd et al. (1988) compared the Rutter
and Gash models in the rain forest of Amazonas, obtaining similar
and reasonable results.
In relation with tropical montane rain forest, Herwitz (1987) worked
in north-east Queensland, and identified that the assumption of
constant canopy/trunk storage capacities was an
oversimplification, and highlighted the need to take into account
dynamic changes in storage capacities; also it was pointed out that
there is a need for additional studies concerning the effects of
forest structure parameters, interception measurement and
modelling.
In this thesis the Rutter rainfall interception model is adapted to
assess the rainfall intercepted, which combined with the
evaporation module is used to estimate the water lost by
evaporation of intercepted rainfall.
111
3.6.6.1 The Rutter model
The Rutter model (Rutter et al., 1971; 1975; 1977), was designed
as an interception model for a Corsican pine stand in Great Britain
but has been applied successfully to tropical rain forests (Calder et
al., 1986; Lloyd et al., 1988; Veen and Dolman, 1989; Jetten,
1996). Canopy water balance is calculated using empirical forest
stand parameters and potential evaporation.
The Rutter model offers several advantages: the input parameters
are relatively easy to obtain from throughfall measurements and
basic meteorological data. It uses stand characteristics rather
than properties of individual plants and calculates water fluxes on
a small time step basis, making it easy to link to a vertical water
balance model. The Rutter model was expanded by Jetten (1994)
to include canopy structure. In this extended model, called
CASCADE, the canopy is layered but uses virtually the same
parameters as the original model. The “cascade” concept is not
applied in this thesis because parameterisation of the model for
TMCF would be very difficult due to the heterogeneous vegetation
types present in TMCF (including epiphytes, for which hydrological
properties are poorly known). Therefore, vegetation has to be
assumed as a single layer for the purposes of modelling
interception. The processes involved in the Rutter model are
shown in Figure 3.30.
112
Where
P Precipitation (mm)
S Transitory storage (mm) per unit area of canopy
D Drainage (mm) per unit area of canopy
p Rainfall fraction falling directly on the ground (mm)
Th Throughfall (mm) per unit area of canopy
C Canopy storage capacity (mm) per unit area of canopy
The original Rutter model presentation is included here. The water
balance, i.e. the change in transitory storage (S) per unit area of
canopy, is calculated as the sum of the proportion of rainfall (P)
that falls on the canopy minus the drainage (D) and evaporation of
intercepted water (Ei) from the canopy:
dC/dt = (1-p-pt)P – D –Ei
p1-p
Th
D
pP(1-p)P
E
SC Canopy
Figure 3.30 Diagram of the Rutter model (Jetten, 1994)
Eq. 3.14
113
where C is in mm and the other variables in mm hour-1. The
fraction of rainfall intercepted by the canopy is calculated as the
difference between rainfall (P), the fraction of rainfall falling directly
on the ground (p), and the fraction of the rainfall diverted to stem
flow (pt) (Jetten, 1994). The canopy drainage is given by
D = 0 where C < S
D = Do eb(C-S) where C >= S
where S is the storage capacity (in mm), the amount of water
retained by the canopy when rainfall and throughfall have ceased
and the canopy is saturated. The minimum drainage rate, Do, is
the drainage rate when C is equal to S (in mm) and b is a
dimensionless parameter. Because the Rutter model was designed
for small time steps (min), the b parameter could not be adjusted
with an exponential function for this thesis in an hourly time step
because it produces model instability. Increasing the dripping
value produces more net rainfall than the real values; additionally
there are no field data to parameterise this parameter in either an
hourly or minute time step. As Rutter et al. (1971, 1975)
recognised, the canopy storage changes significantly during a 5-
min period (the time step used by Rutter et al. 1971) and this is a
further reason why the b parameter can not be adopted from their
work.
Drainage from the canopy is computed as a water balance with
canopy parameters. Then the drainage function is modified to:
D = 0 where C < S
D = C-S where C >= S
The evaporation from a wet canopy surface is considered equal to
the evaporation from an open water body. The potential
Eq. 3.15
Eq. 3.16
114
evaporation (PE), calculated for the atmospheric conditions
prevailing at the top of the canopy, can therefore be used.
Furthermore, the evaporation of intercepted water (Ei) is
proportional to the area of the wetted surface (Rutter et al., 1971):
Ei = PE * C/S where C < S
Ei = PE where C >= S
Total throughfall (Th) is the sum of direct throughfall and canopy
drainage, and is expressed as:
Th = D + (o –pt) * P
The Rutter model includes routines to compute stemflow, but this
part of the model is not included in this thesis due to the lack of
data for parameterisation. The characteristics of the stemflow
model are included here just for information. Stemflow (Sf) is the
depletion of trunk storage capacity (Ct) as compared with trunk
storage capacity (St). The excess water is completely diverted to
stemflow at the end of each time step, and evaporation is measured
as 0.02*PE (Jetten, 1994). Based on Gash et al. (1978), the model
includes a numerical solution with a finite difference
approximation of the change in canopy water storage (dC/dt).
The evaporation of intercepted water from the canopy depends on
the micro-climate inside the canopy (Jetten, 1994). The energy
available is calculated with an exponential extinction parameter
describing the cumulative leaf area, which was used by Rubiano
(1998) in the Beers Law equation.
The extinction factor (k) was determined using photosynthetically
active radiation (PAR) measured at three different heights (1, 3, and
Eq. 3.17
Eq. 3.18
115
6 m) in the secondary forest plot (Rubiano, 1998). The light
extinction coefficient (k) estimated by Rubiano (1998) was 0.27,
and is used in this module to compute the energy available for
evaporation of intercepted rainfall within the forest. Canopy water
storage capacity (S) was derived from samples of forest vegetation
as is outlined in section 3.5.2.2.
The design for the interception sub-model is illustrated in Figure
3.31.
3.6.7 Sub-surface water sub-model
Soil hydraulic properties are modelled on the basis of measured
structural properties such as texture and bulk density by the pedo-
InputsRainfallPotential evapotranspirationNet radiation at ground level
ParametersLeaf area indexLeaf capacityK, evaporative energy extinction
coefficient for vegetation forestVegetation cover
CalculatedStem interception, storage, evaporation, and drainageCanopy interception, storage, evaporation, and drainageDripThroughfallWater reaching the ground
Outputs- Water reaching the ground- Water lost from canopy by evaporation
Figure 3.31 Diagram of interception sub-model.
116
transfer function of Saxton et al. (1986) that uses the Brooks and
Corey (1964) water retention function. The Saxton et al. (1986)
method is also used to determine soil hydraulic conductivity for
recharge calculation. These model sections are described next.
3.6.7.1 Modelling flow of water in porous media
The size of soil pores through which water flows and pore-size
distribution are mainly determined by grain-size distribution
(Dingman, 1994). For many purposes, particle-size and pore-size
distribution are characterised by soil texture, which is determined
by the proportion per weight of clay, silt, and sand. Figure 3.32
illustrates the scheme for defining soil textures developed by
USDA. Soil texture is determined from soil samples after particles
larger than sand (> 2 mm) have been removed.
The definitions of soil composition, soil classification, and soil
properties, as given by Kutilet and Nielsen (1994), were used in this
study. The routine aims to determine soil hydraulic properties, for
example hydraulic conductivity, matric potential, and soil
moisture. Several attempts have been made to predict moisture
release functions from soil texture data (Van Genuchten, 1980;
Arya and Paris, 1981; Grismer, 1986). Knowledge of particle-size
distribution helps determine pore-size distribution and moisture
retention characteristics. Although this approach presents several
difficulties, it is cheaper and easier than field determination
(Campbell, 1985) but this approach must be used with care
because the soils are extremely complex and variable. The
determination of pore-size, particle-size and pore-size distribution
facilitates the calculation of the space available in the soil for water
storage and movement. Arya and Paris (1981) established a non-
linear relationship between particle-size and pore-size
117
distributions, because water is held within the soil by capillary
binding of water in the pores; then the shape of water-retention
curve depends to a great extent on the pore-size distribution of the
soil (Anderson, 1990).
3.6.7.2 Soil water retention and matric potential
Matric potential is the amount of potential energy per unit of mass
or volume of water in a system, compared to that in pure free water
at a reference elevation point. Because water movement is very
slow through soil micro-pores, kinetic energy is extremely low and
may be neglected. Potential energy therefore dominates and
CalculatedSoil moisturePonding timeDistance to wetting frontTotal soil water infiltration at ponding timeInfiltration rate after ponding timeTotal soil water infiltration between ponding time and
end of time stepOverland flowDischarge from soil to subsoil water table or directly to
drainage system
Outputs
Infiltration waterOverland flow (depth)New soil moistureRechargeLoss water by evaporation
Figure 3.34 Diagram of infiltration sub-model.
131
3.6.9 Overland flow sub-model
3.6.9.1 Sub-model description
Overland flow occurs whenever the rate of water application to the
ground surface exceeds the rate of infiltration into the soil (ward
and Elliot, 1995), or on the hillsides during rainstorm events when
surface depression storage is exceeded (Kirkby et al., 1980).
Runoff may result from short, highly intense rainfall, long low-
intensity rainfall, or a combination of both (Maidment, 1993).
Several approaches are used to estimate overland flow. Black-box
models have an input-output structure rather than physically
based transfer function. A statistical correspondence needs to be
established between input and output data. The unit hydrograph,
extreme frequency analysis, and regression analyses are examples
of this type of model. Deterministic models are based on complex
physical theory. They include several flow equations, which
produce high computational cost and significant data
requirements. They improve our understanding of the hydrological
system, regardless of their predictive success which is often not as
good as simpler models. In all cases models need to be adapted to
the problem rather than vice-versa. Conceptual models are a
combination of deterministic and black box models. Such models
are formulated on the basis of a simple arrangement of a relative
small number of components with a simplified representation of
elementary system.
To understand the processes that control overland flow, the factors
involved were systematically analysed. Four different processes
were taken in account at different times: Hortonian overland flow,
subsurface flow, saturation overland flow, and ground water
movement. Hortonian overland flow, as discussed in the
132
infiltration model (see Section 3.6.8), refers to the amount of
effective rainfall that reaches the soil at rates higher than soil
infiltration capacity. Subsurface flow or throughflow refers to the
water that infiltrates into the soil and percolates rapidly, mainly
through macropores. Saturation overland flow occurs when the
water table reaches the surface (100% soil saturation) and forms
excess water, thus generating overland flow. Ground water
movement can generally be described in two ways: vertical
movement, which includes raising the ground water table or
pumping water through wells with natural hydraulic
characteristics, and lateral movement, such as throughflow.
Rainfall characteristics exert a strong influence on overland flow
events. Rainfall intensity combined with soil water saturation and
storm characteristics have important implications for flow
generation.
Temporal and spatial variations in runoff, caused by rainfall
properties, may be greatly enhanced by spatial variations in
infiltration capacity of the soil surface. Research conducted in
humid areas indicates that the frequency and magnitude of storm
channel runoff is controlled mainly by the extent and distribution
of saturated areas (Anderson and Burt, 1985). Such areas respond
quickly even to low-intensity rainstorms. Therefore, the spatial
distribution of soil moisture cannot be regarded as a major factor
in the control of storm runoff generation, and spatial non-
uniformity of runoff generation relates significantly to spatial
variations in infiltration capacities.
Overland flow frequency and magnitude therefore depends on
several factors, including geomorphological characteristics (such as
slope, slope distance, aspect, catchment area, among others),
rainfall characteristics (frequency, intensity, duration) and soil
133
properties (hydraulic conductivity, soil texture, porosity, among
others) including the ratio of rocks to soil, on the surface, that in
turn influenced the soil hydrological fluxes. High runoff can be
predicted in those cases where the rock-soil ratio is high, while low
runoff can be expected in those cases where this ratio is low.
Based on Hortonian overland flow, the runoff model indicates water
height at a given time and point, using a simple hydrological
balance in a given time step,
where D is the water depth of overland flow [mm]; P, effective
rainfall (direct rainfall plus throughfall) [mm]; Runin, runoff
contribution from slopes above the point; I, the infiltration for that
period [mm]; E, soil evaporation [mm]; and Runout, the overland
flow outflow. Most values are calculated by other sub-modules or
by the results of previous iterations. Figure 3.35 presents the
diagram of this sub-model.
D P Runin I E Runout= + − + +( ) ( )
Input- Water depth of overland flow,
net rainfall, infiltration, pot-evaporation
OutputOverland flow
CalculatedOverland flow
Figure 3.35 Diagram of runoff sub-model.
Eq. 3.40
134
3.6.9.2 Surface component of overland flow at the catchmentscale
The integration of a surface component in the model produces an
extension of the 1D model at the plot scale to a 2.5 D model at the
catchment scale.
The overland flow sub-model is the only component within the
hydrological model that is changed in this way. Water which ponds
is allowed to flow downslope according to the local drainage
direction (LDD). The overland flow sub-model computes the
amount of outflow surface water in a down slope direction for a
particular area. The inflow water volume is computed by the
UPSTREAM routine used in the GIS component (PCRaster
software), which is the sum of all overland flow values of the
upslope direction areas. This command (upstream) uses as a
parameter the local drainage direction (ldd) network, which is a
direction network connection between areas, which indicates the
flow direction; ldd is derived from the digital elevation model
(DEM). All surface water that is not infiltrated or evaporated
moves down slope direction. The model does not incorporate a
detention storage capacity. LDD is calculated using the 8 point
pour algorithm with flow directions from each cell to its steepest
downslope neighbour. The manner in which PCRaster calculates
the LDD is explained in the user manual (Utrecht University,
1996).
3.6.10 Erosion sub-model
Soil erosion is modelled to identify areas where soil detachment
and loss by natural or anthropogenic causes occurs. The
importance of assessing and quantifying soil loss in TMEs due to
135
LUCC lies in its effects on landscape transformation and
environmental consequences.
Most of the knowledge of soil erosion mechanisms is the result of
studies carried out by the US Soil Conservation Service, which has
emphasised the prediction of erosion rates. Therefore, most of
these approaches are based on empirical equations or on
generalisations for specific scenarios. Kirkby and Morgan (1980)
compiled a number of these developments in detail. The most
common reference point in this field is the Universal Soil Loss
Equation (USLE), which estimates erosion as the product of a
series of terms such as rainfall, slope gradient, slope length, soil
and cropping factors. The equation allows individual factors,
developed from extensive observation of experimental plots in the
US, to be tabulated.
Recent developments in this field are discussed by Boardman and
Favis-Mortlock (1993), who compiled the most commonly used
erosion models and described the different approaches used and
the specific characteristics of each. Although these models often
require significant data for parameterisation and input information,
they describe erosion in detail and the accuracy of the approaches.
Despite technological advances, the development of erosion models
is still largely empirical. Further, research is required because of
the extreme complexity of the physical processes involved in
erosion.
Climatic variables, such as rainfall intensity, have a major effect on
the ecosystem and also exert an effective control on the variables
that determine soil stability. Vegetation cover also has an
important effect because it provides protection against rain splash
and sediment detachment and transportation (Kirkby and Morgan,
1980; Boardman and Favis-Mortlock 1993).
136
Musgrave (1947) developed a relationship between rainfall
characteristics and the amount of soil loss using data from several
stations. He developed a relation which involves slope parameter,
surface runoff and rainfall properties (Thornes and Gilman, 1983;
Thornes, personal communication), and is expressed as follows:
where k1, m and n are parameters (discussed below), and q the
surface overland flow per unit width (mm.h-1) (Thornes, 1990) as
defined in previous section.
Musgrave’s (1947) equation was the basis for Thornes’s model
(1985) which was developed on the basis of results of small
experimental plots (20m2) in Spain and with rainfall records
shorter than an hour, which guarantee that the rainfall properties
are important.
The erosion model proposed by Thornes (1985) was used in this
study, and focuses on the competitive interaction between erosion
and vegetation cover. This routine can be incorporated into the
hydrological model because it is based on physical characteristics
of the soil profile (Thornes, 1990). The spatial variation of erosion
on hillsides can also be examined for the entire watershed. This
method was also chosen because (a) it is a physically-based
approach that uses local data; (b) the spatial resolution of the
proposed model (25m pixel side size) which is related to the
experimental plot size used by Thornes; (c) the input data needed
are available, which uses local information at a good temporal
resolution; and (d) the parameters are easy to estimate or adapt
from the literature.
E k q sm n= 1 Equ. 3.41
137
Parameter k1 is a coefficient that depends, among other things, on
the amount and intensity of rainfall, the effects of lithological
constraints on the availability of materials for erosion within
textural soil properties and organic matter, as related to the size of
material to be transported (Thornes and Gilman, 1983). k1 can be
determined through an experimental combination of rainfall
simulation and collected soil loss in the field, which was not
carried out in this project. Instead of this, Thornes and Gilman
(1983) suggest that k1 be used as a constant value of 0.02, or a
linear coefficient adjusted to empirical data, or use the erodability
factor derived from USLE tables. In the model, the USLE soil
erodability factor is used as the k1 parameter, and is determined by
the soil’s physical properties (texture and organic matter).
Therefore, k1 for sandy loam soil with an organic matter content
higher than 3.5% is 0.19, as derived from the USLE erodability
factor table 3.7 (Kirkby and Morgan, 1980).
Organic matter contentTexture class < 0.5 per cent 2 per cent 4 per centSand 0.05 0.03 0.02Find sand 0.16 0.14 0.10Very fine sand 0.42 0.36 0.28Loamy sand 0.12 0.10 0.08Loamy fine sand 0.24 0.20 0.16Loamy very finesand
Table 4.16 Sensitivity to n factor of erosion equation
Erosion sensitivity to n factor of erosion
-44
-24
-4
16
-100 -50 0 50 100
Percent variation of parameter n of erosion equation
Eros
ion
sens
itivi
ty
%∆
%∆
Figure 4.24 Sensitivity to n factor of erosion equation
186
4.4 Summary of 1D sensitivity analysis
From the parameter sensitivity analysis in the 1D model, the most
important parameters within the model are vegetation cover, soil
texture, soil porosity and soil depth. Erodability factor k1 and m
and n parameters of the erosion equation produce important
changes only in erosion. From this, it is clear that erosion is the
most sensitive variable from the model. Other parameters produce
small changes in the hydraulic variables, which are taken into
account in the analysis. Vegetation cover protects soil from direct
rainfall and according to the type of vegetation, change in OF and E
can be very significant within the watershed. Table 4.17 shows a
summary of the sensitivity of the different variables to parameters
variation.
Variables Parameters
Soil moisture
Matric Potential
Hydraulic conductivity Infiltration Evaporation
Overland flow Erosion
A of Rn eq.B of Rn eq.Light extintionLAIMx. Canopy water storage capacityVeg. CoverSoil textureSoil porositySoil depthK erodability factorm of erosion eq.n of erosion eq.
Not sensitive < 2%
Slightly sensitive 2% - 7%
Sensitive 7% - 20%
Moderately sensitive 20% - 100%
Severely sensitive 100% >
Table 4.17 Summary of 1D sensitivity analysis by classes with the colour code
187
4.5 2.5D model sensitivity analysis
Sensitivity analysis at the catchment scale is carried out to identify
the area characteristics within a given LUCC scenario, which
produce the highest impact on the model hydrological variables.
Based on the sensitivity analysis at the plot scale (1D model), the
most sensitive hydrological variables to LUCC were overland flow
and soil erosion. These two variables are then used in this part of
the analysis process as an indicator of LUCC impact. The
parameter (vegetation cover) was identified from the 1D sensitivity
analysis as one of the drivers of LUCC impact, and in essence,
changes in vegetation cover are the same as changes in LUCC. So
vegetation cover at the catchment scale is going to be included in
the model with the LUCC scenarios designed for this thesis.
Physical soil property parameters (soil texture, soil porosity and
soil depth, as well as the soil parameters in the erosion equation)
are assumed uniform across the whole catchment irrespective of
land cover since the objective of this modelling is to understand
landscape sensitivity resulting from topographical variability and
hydrological connectivity in combination with LUCC. These
topographic variables are used in the sensitivity analysis at the
catchment scale, to identify how they control catchment sensitivity
to LUCC and thus which areas within the catchment are more
sensitive to LUCC.
Further up, the topographic characteristics used, are defined and
described. In addition, overland flow and erosion sensitivity
analyses are presented.
188
4.5.1 Definition of topographic characteristics
In order to analyse the flux variation with changes in scenarios,
physical properties of deforested areas within the watershed
between iterations in the scenarios were summarised and
averaged.
Topographic variables have been used to explain and assess some
physical events that occur in the environment. Quine and Walling
(1993) used topographic variables to assess the landscape
sensitivity to erosion and deposition. Gerrard (1993) used specific
relief values like maximum slope angle, stream density, stream
frequency and stream order to assess the landscape sensitivity.
McKenzie and Ryan (1999) combined environmental variables from
the landscape to predict spatial soil properties with good results.
The variables taken into account in this part of the analysis were:
slope, aspect, topographic index, altitude and proximity of the
deforested area to river channels.
Slope: the degree of rate of change of elevation per unit of
horizontal distance. Slope can be derived from a Digital Elevation
Model1 (DEM, which is a raster2 image whereby each grid cell has
an elevation value).
Aspect is the direction of the maximum slope in a given point with
relation to a geographical north direction (given in degrees). It is
also derived from the DEM.
Altitude or elevation is the vertical distance (m) of a given point in
relation to a reference point, usually mean sea level.
Distance to rivers (m) was computed using the raster image of the
river channels, which was classified into 18 classes using a 50m
buffer of horizontal distance either side of the rivers channels.
1 Interpolated surface derived from elevation points.2 An image surface conformed by pixels or cells of uniform size
189
Topographic index was first proposed by Kirkby (1975) and then
developed as a part of a complete hydrological model by Beven and
Kirkby (1976, 1979). This index represents the propensity of any
point in the catchment to develop saturated conditions. High
values will be caused by either long slopes or upslope contour
convergence, and low slope angles. It can be used as a guide for
water and sediment movement. It has proven a useful index for
predicting soil properties within the landscape (Mckenzie et al.,
1999). It is defined as:
TopIndex = ln (Ac / tan Β)
Where Ac is the specific contributing area expressed in m2 per unit
width orthogonal to the flow direction, and B is the slope angle.
Normally both Ac and B are derived from the analysis of digital
terrain model, in which the evaluation of pixel connectivity is
produced, and integrated by the accumulative area of upslope
direction (Beven et al., 1995).
All variables were calculated with the GIS at a 25-m pixel size and
averaged for each of the deforested areas of each iteration in each
scenario. A summary of average values of topographic variables
from deforested areas, by iteration per scenario, is given in
Appendix 10.
4.5.2 Sensitivity analysis at the catchment scale
Sensitivity analysis was carried out for five LUCC scenarios. Each
iteration of each scenario was run for a year at an hourly time step,
using the 2.5D model developed in PCRaster (Utrecht University,
1996) for the whole catchment. Model initial conditions were
taken from modelled results produced at the end of a one-year pre-
Eq. 4.1
190
run. Overland flow (OF) and erosion (E) were the variables taken
into account in the analysis. The last nine months of the
simulated year were summarised using the one year average by m2
for each of the flux variables. This was done to avoid the inclusion
of data from the period when the model was adjusting to initial
conditions. Three months were shown to be enough time for model
recovery. Three different initial soil moisture conditions were used
to run the model with the same rainfall events; as is shown in
Figure 4.25 the soil moisture takes similar pattern after the first
600 hours. The model was parameterised with the parameters
outlined in the previous section 4.3 and with the initial image of
LUCC for scenarios. The simulated period includes two rainy
seasons and one dry season, and accounts for more than 6000
time steps in the model process.
Graphical analysis of each variable (OF and E) within each scenario
(SC1 to SC5) was undertaken and presented in a set of 6 graphics.
The graphics contain:
1- Pixel average for the catchment of one year total yielded by the
variables (OF in mm and E in mm).
2- Percent variation for each variable between each LUCC iteration,
given as a percentage.
3- Sensitivity of the variable to LUCC, which is the percent of
variation between two consecutive iterations divided by the change
in deforested area between the same iterations. This gives the net
response per unit of deforestation. They are shown on the same
scale for all scenarios to allow comparison of the sensitivities.
4- Total deforested area by iterations (ha.) compared with mean
altitude of deforested area by iteration (masl).
5- Mean slope and aspect of deforested area by iteration. Both are
presented in degrees.
6- Mean topographic index and mean distance to river of the
deforested area.
191
191Figure 4.25. Modelled soil moisture with different initial conditions for the same rainfall pattern
Modelled soil moisture response to different initial conditions
4.5.2.1 Sensitivity analysis of overland flow to LUCC at the catchment scale
Variation in OF between iterations was calculated and then divided
by the deforested area between iterations to get the OF sensitivity
to LUCC by scenario (table 4.18, page 205). Table 4.18
summarises the total OF by iteration for each scenario, the percent
of variation and the sensitivity to LUCC. The sensitivity is
highlighted with colours to identify the highest sensitivities.
Figures 4.26 to 4.35 show total of OF by iteration, the percent
variation of OF between iteration and the OF sensitivity for the
scenarios.
The pattern of deforestation between iterations by scenario is
different. Consequently the OF yield is also different for each
scenario throughout the iterations. Although SC1 (cellular
automata scenario, page 54) has one of the largest deforested areas
in the initial iterations (see table 4.18), it is ranked third due to the
average yield of OF. Also SC1 was the most uniform in
deforestation pattern because the percent variation was the lowest
(83%).
The scenarios with the lowest yield in OF during the simulation
period were SC3 and SC4 (averaging per iteration at 7919 and
7927 mm respectively). The percent variation of these two
scenarios were relatively uniform (132 and 135 %).
Scenarios SC2 and SC5 produce the highest average OF by
iteration (7965 and 7964 mm respectively). Scenarios SC3 and
SC4 both have the same deforestation pattern in opposing
directions, SC2 starts form the lower part of the catchment, and
193
SC5 starts from the top of the catchment. The percent of variation
were similar for both (132 and 125 %).
The LUCC pattern in SC1 is very varied (see Figures 4.26 and
4.27). Deforestation occurs at the beginning in the lowest part of
the catchment (lower mean altitude) and where the slope is small.
More than half of the area (816.4 ha) in SC1 is deforested in the
first four iterations, which produces an additional 35 mm in OF
and 10 mm in E. For this reason, the percentage of variation of
OF and E decreases rapidly. Mean altitude and mean slope of the
deforested area increases gradually through the iterations, but the
area deforested per iteration decreases. A few oscillations of slope
in the deforested area at the end of the scenario, have some
relation with the variations in OF sensitivity. The decreasing trend
of mean topographic values and mean distance to the river also
have some similarities with the OF sensitivity in the last iterations.
Despite those variations, the OF sensitivity in SC1 is very low,
without large changes (range 0.2 to 1.7).
In SC2, 1109 ha (78% of the area) were deforested in the first five
iterations producing 4 mm OF (4% of additional OF generated by
deforestation) and 18 mm in E ( 81% of additional E generated by
deforestation). In these iterations, while OF sensitivity remains
constant, E sensitivity decreases until iteration 2 and then remains
constant. This means that E is affected by other additional
variables compared to OF. The percent variation in OF is related to
the amount of OF. However, it is important to highlight that the
biggest changes in mean altitude, deforested area, and topographic
index occur in the first iterations (see Figures 4.28 and 4.29).
Between iterations 8 and 12, OF sensitivity changes significantly.
These changes are related to a decreasing mean slope in deforested
area, and increasing mean altitude, topographic index, and
distance to rivers of the deforested area.
194
Figure 4.27 Mean topographic variables for deforested areas in SC1
D eforested area and m ean altitude in S C 1
0
100
200
300
400
500
0 5 10 15 20 25Iteration
1400
1900
2400
2900
D eforested areaA ltitude
M ean slope and aspect of defo rested area in
SC 1
0
100
200
300
0 5 10 15 20 25Iteration
0
10
20
30
40
50
60
Aspect
Slope
M ean to p.index and distance to rivers in S C 1
6
7.5
9
10.5
12
0 5 10 15 20 25Iteration
0
3
6
9
12
15
18
Topographic Index
Distance to river
194
Figure 4.26 Overland flow sensitivity in scenario 1 (deforested pattern with cellular automata)
T otal o verland flow by iteratio n
7850
7880
7910
7940
7970
8000
0 5 10 15 20 25Iteration
% variation of o verland flow
betw een iterations
0
0.05
0.1
0.15
0.2
0.25
0.3
0 5 10 15 20 25Iteration
O verland flo w sensitivity
0
5
10
15
0 5 10 15 20 25Iteration
195
D efo rested area and m ean altitude in S C 2
0
100
200
300
400
0 5 10 15 20Iteration1400
1900
2400
2900
D eforested area
A ltitude
M ean slo pe and aspect o f defo rested area in
S C 2
0
100
200
300
0 5 10 15 20Iteration
0
10
20
30
40
50
60
A spect
Slope
M ean to p.index and distant to rivers in S C 2
6
7.5
9
10.5
12
0 5 10 15 20Iteration
0
3
6
9
12
15
18
Top. index
Distance to rivers
Figure 4.29 Mean topographic variables of deforested areas in SC2
195
Figure 4.28 Overland flow sensitivity in scenario 2 (forest conversion with a fixed horizontal distance from river channel in uphill direction)
T otal O verland flo w by iteratio n
7850
7880
7910
7940
7970
8000
0 5 10 15 20Iteration
% variatio n o f overland flo w
betw een iteratio ns
0
0.05
0.1
0.15
0.2
0.25
0.3
0 5 10 15 20Iteration
S ensitivity o f O verland flo w
0
5
10
15
0 5 10 15 20Iteration
196
After iteration 12, OF sensitivity has a big oscillation, decreasing
and then increasing to its maximum value, as a result of the
change in distance to rivers, mean altitude, and topographic index.
This wave on the graphic of OF sensitivity could be explained by
the combination of those factors, in particular the high values of
mean slope and mean altitude, which in the last iterations produce
the highest variation in OF sensitivity. Overall, the highest OF
sensitivity is produced in SC2, and ranged from 0 to 16 (see Figure
4.28).
In the SC3 (see Figure 4.30), the deforested area within the first 12
iterations is 148 ha (10% of the total area), which produces 30 mm
of OF (29% of additional OF generated by deforestation) and 2.8
mm of E (12% of additional E generated by deforestation). The OF
sensitivity in SC3 is a mirror view of the OF sensitivity of SC2 with
small variations. That is expected because the deforested areas are
very similar but in reverse directions. Although the areas are
similar, the OF sensitivity is larger in SC3 than SC2. This could be
due to the biggest change in mean slope and mean topographic
index occurring in the first three iterations, and the longer distance
to rivers. In SC3, the OF sensitivity in the first iterations is high
and extensive, with some variability during the last few iterations.
The big oscillation in the OF sensitivity occurs in iterations 5, 6
and 7, and could be due to changes in percent variation of OF,
because none of the other aspects have the same trends (see Figure
4.31).
197
D eforested area and m ean altitude in S C 3
0
100
200
300
400
0 5 10 15 20Iteration
1400
1900
2400
2900
D eforested area
M ean altitude
M ean slo pe and aspect o f defo rested area in
S C 3
0
100
200
300
0 5 10 15 20Iteration
0
10
20
30
40
50
60
A spect
Slope
M ean top.index and distant to rivers in S C 3
6
8
10
12
0 5 10 15 20Iteration
0
3
6
9
12
15
18
Top. Index
D istance to river
Figure 4.31 Mean topographic variables of deforested areas in SC3
197
Figure 4.30 Overland flow sensitivity in scenario 3 (forest conversion with a fixed horizontal distance towards channel rivers in downhill direction)
T o tal overland flo w by iteratio n
7850
7880
7910
7940
7970
8000
0 5 10 15 20Iteration
% variatio n o f overland flo w
betw een iteratio ns
0
0.05
0.1
0.15
0.2
0.25
0 5 10 15 20Iteration
S ensitivity o f O verland flo w
0
3
6
9
12
15
0 5 10 15 20Iteration
198
The OF sensitivity for SC4 ranges from 0.6 up to 7, which is half
compared to SC2 and SC3. This means that the areas close to
river channels are more hydrologically sensitive to LUCC than the
deforested areas created with the elevation pattern (SC4 and SC5).
The deforested area in SC4 between iterations 1 to 8 was 735 ha
(52% of the catchment), which produces 29.4 mm of OF (27% of OF
generated by deforestation) and 9.7 mm of E (42% of E generated
by deforestation) (see Figures 4.32 and 4.33). The variation of OF
sensitivity in the first eight iterations (see Figure 4.32) is relatively
constant and small. The areas deforested during these iterations
are the lowest (in terms of altitude) in the watershed; the elevation
of these areas ranges between 1400m to 2000m (see Appendix 10).
After the eighth iteration, the OF sensitivity changes in proportion
with a number of oscillations and increases until the fifteenth
iteration. Between iterations 10 and 15 the percentage variation of
OF increases and decreases, but the OF sensitivity in the same
range always increases. Those variations in the OF sensitivity
could be due to a combination of a decrease in mean slope and
mean aspect and an increase in mean distance to rivers (see Figure
4.33) of the deforested area. Also, the mean elevation, which
increases constantly through the simulation, might have some
effects on the OF sensitivity since rainfall is distributed as a
function of elevation. The large variations in OF sensitivity are in
the last 5 iterations, even though the slope values of the deforested
areas in the last iterations are small, and those areas are further
away from rivers channels, but do have high rainfall receipts.
199
D eforested area and m ean altitude in S C 4
0
100
200
300
400
0 5 10 15Iteration
1400
1900
2400
2900
D eforested area
M ean altitude
M ean slope and aspect of defo rested area in
SC 4
0
100
200
300
0 5 10 15Iteration
0
10
20
30
40
50
60
Aspect
Slope
M ean to p.index and distant to rivers in SC 4
6
8
10
12
0 5 10 15Iteration0
3
6
9
12
15
18
Top. Index
D istance to river
Figure 4.33 Mean topographic variables of deforested areas in SC4
199
Figure 4.32 Overland flow sensitivity in scenario 4 (forest conversion with fixed distance of altitude, in uphill direction from the lowest to the highest point)
T otal o verland flo w by iteratio n
7850
7880
7910
7940
7970
8000
0 5 10 15Iteration
% variation overland flo w
betw een iteratio ns
0
0.05
0.1
0.15
0.2
0.25
0.3
0 5 10 15Iteration
S ensitivity o f O verland flo w
0
5
10
15
0 5 10 15Iteration
200
As in the case for SC2 and SC3, SC5 is the mirror of SC4, but in
this case the range in OF sensitivity of SC5 is a half of SC4. The
deforested area in SC5 in the first 6 iterations was 527 ha (37% of
the catchment), which produces 63 mm of OF (64% of OF produced
by deforestation) and 10.5 mm of E (45% of E produced by
deforestation). The OF sensitivity is highest for SC5 in iteration 2
(Figure 4.34); the increase in mean slope and decrease in mean
distance to rivers combined with decrease in elevation of the
deforested area produce the variation in the OF sensitivity in the
first 5 iterations. Beyond that, values of the topographic attributes
of the deforested areas change, but the OF sensitivity remains low
(see Figure 4.35). It means deforestation in the highest areas of the
watershed at the beginning of the scenario produces more change
in overland flow by area than occurs with SC5, despite the OF yield
being lower in SC4 than in SC5.
Under SC1, areas in most of the iterations are not sensitive to
LUCC, with the exception of the last iteration, where the terrain is
the steepest and highest in elevation. In SC2, despite the
deforested areas at the beginning producing most of the excess of
OF in the catchment, these areas are not particularly sensitive to
LUCC, but the deforested areas between iteration 7 and 13 as well
as the areas in the last 3 iterations are more sensitive. The highest
values of mean topographic index combined with high mean
elevation as well as increasing distance to the rivers with
increasing mean slope values, produce high values in OF
sensitivity. Conversely, in SC3 the most sensitive areas coincide
with the areas in SC2; these areas are the highest in elevation with
the steepest slopes. The same conclusion can be made with SC4
and SC5, which show the most sensitive areas in the last iteration
in SC4, which are the deforested areas at higher elevations with
greater distance to rivers even though mean slope and mean aspect
decrease. These areas coincide with the initial areas of SC5.
201
Figure 4.35 Mean topographic variables of deforested areas in SC5
D eforested area and m ean altitude in S C 5
0
100
200
300
400
0 5 10 15Iteration
1400
1900
2400
2900
D eforested area
M ean altitude
M ean slope and aspect of defo rested area in
SC 5
0
100
200
300
0 5 10 15Iteration
0
10
20
30
40
50
60
Aspect
Slope
M ean to p.index and distant to rivers in SC 5
6
8
10
12
0 5 10 15Iteration
0
3
6
9
12
15
18
Top. Index
D istance to river
201
Figure 4.34 Overland flow sensitivity Scenario 5 (forest conversion with fixed distance of altitude, in downhill direction from the highest to the lower point)
T otal o verland flo w by iteratio n
7850
7900
7950
8000
0 5 10 15Iteration
% variation of o verland flow
betw een iteratio ns
0
0.1
0.2
0.3
0 5 10 15Iteration
S ensitivity o f O verland flo w
0
5
10
15
0 5 10 15Iteration
202
This highlights that the areas with highest OF sensitivity to LUCC
within the catchment are at elevations higher than 2200 m.
Despite the fact that high overland flow is produced in the highest
elevations, the stronger OF sensitivity is shown in SC2 and SC3,
which are related with distance to rivers. Clearly, as elevation
increases so does distance from rivers.
To identify whether or not physical variables are involved in the OF
sensitivity explanation, first of all a Kolmogorov-Smirnov test to
determent the topographic variables normal distribution which
they are, and then a multiple regression analysis was performed for
all scenarios, using the statistical package STATISTICA 6.0
(produced by Statsoft Inc., USA). Table 4.19 summarises and
compares the results of the multiple regression analysis for the
scenarios on the basis of data from Appendix 10. The dependent
variable was OF sensitivity and topographic variables were used as
independent variables. The analysis included calculation of the
explanatory coefficient of determination (R2) of the OF sensitivity, in
relation to independent variables, and the t-coefficient for
statistical significance for each variable (Rincon-Romero, 2000).
The multiple linear correlation coefficient (R) shows how dependent
variables (OF sensitivity) can be explained as a function of the
linear combination of independent variables (topographic
variables). Comparing the R values from the five scenarios (see
table 4.19), the R of SC4 (0.98) shows that 98% of the OF
sensitivity can be explained by linear combination of the
topographic variables for the deforested area, while for SC1 the R is
just 51%, only 51% of the OF sensitivity can be explained by linear
combination of topographic variables. The highest coefficients of
determination are in SC4 and SC5, which are the scenarios based
on elevation, followed by SC3 and SC2 (0.84 and 0.83, respectively)
203
and the lowest is SC1. In the same way, R2 is the proportion of the
variation in the dependent variable that can be attributed to the
variation of the combined independent variables. The maximum
value of R2 is in SC4 (R2 = 0.97) and the minimum in SC1 (R2 =
0.26).
The critical value of a statistically significant F value, with 95%
probability with (5,15 in SC1), (5,11 in SC2 and SC3), (5,8 in SC4
and SC5) degrees of freedom is 1.89. From table 4.19, it is clear
that none of the computed F values from all scenarios surpassed
the critical value, and as a result, the null hypothesis was not
rejected in any scenario. This confirms that there is no reason to
believe that the independent variables are correlated with each
other. The probability that R would have fortuitously occurred if
the null hypothesis held true was less than 0.05 in all scenarios
with the exception of SC1. The criteria to argue that each variable
helps in the explanation of dependent variable when its used in
combination with the other variables is the t-critical value.
Assuming α=0.5 the t-critical value for the explanation of OF
sensitivity by the landscape properties discussed is 2.131 for SC1
(N=15), is 2.201 for SC2 and SC3 (N=11) and is 2.306 for SC4 and
SC5 (N=8). If the computed t-value exceeds the t-critical value, it
means that the variable is a significant contributor to explanation
of the dependent variable when it is used in combination with the
other variables. The significant contributor for SC1 is slope, for
SC3 is aspect and for SC4 and SC5 is the distance to rivers. For
SC2 none of the topographic variables appear as a significant
contributor in the explanation of the OF sensitivity. This does not
mean that the other variables are not involved in the explanation of
OF sensitivity; it simply says that the associated probability that
the relationships of those variables could occur by chance, if the
null hypothesis were true, is less than 0.05.
204
From this analysis, it can be concluded that the topographic
variables, which are involved in greater proportion in the
explanation of the OF sensitivity, are slope and distance to rivers.
Aspect is not an important variable, a part of that paradoxically, in
SC3 and also SC2, where is the greatest contributor to the
explanation of the OF sensitivity. Topographic index seems not to
be an important variable in the explanation of the OF sensitivity for
all scenarios with the exception in SC1. Despite the fact that SC2
and SC3 appear as the most sensitive scenarios, the variable
distance to rivers was not the most important in the regression
analysis for those scenarios. Although the degree of correlation
between dependent and independent variables in some models are
not high, further combinations of variables could be tested in order
to produce a better correlation of topographic variables and the OF
sensitivity.
205205
Table 4.18 Summary of data used in OF sensitivity analysis
F value (5,15) 1.074 (5,11) 5.145 (5,11) 5.294 (5,8) 53.775 (5,8) 21.002F
significance0.413 0.011 0.010 6E-6 2E-4
Std. Errorof estimate
0.002N = 21
0.024N = 17
0.218N = 17
0.003N = 14
0.004N = 14
t(15) Sig. t t(11) Sig. t t(11) Sig. t t(8) Sig. t t(8) Sig. tAspect -1.18 0.25 2.14 0.06 2.29 0.04 -0.17 0.87 1.89 0.10Slope -2.17 0.04 0.47 0.65 0.34 0.73 1.40 0.20 -0.97 0.36
Table 4.19. Multiple regression analysis of overland flow for all scenarios. Significant relationships are highlighted.
206
207
4.5.2.2 Sensitivity analysis of erosion to LUCC at the catchment scale
Erosion is driven mainly by overland flow, so it can be expected to
have a similar behaviour. SC3 (deforestation pattern in a downhill
direction towards river channels) produces the lowest average
erosion per iteration (77.59 mm in a simulated year) (see table
4.20). While paradoxically the highest average values were
produced by SC2 (complementary to SC3 but in an opposing
direction) with 92.94 mm. This difference is about 1.53 m3 ha-1,
which sums to 2166 m3 for the whole catchment. This value is
similar to the resulted value of the difference between erosion
yields of the iteration at the beginning and at the end of the
simulated scenario i.e. with near-full forest cover and almost totally
deforested. The total deforestation in the catchment produces
3294 m3 of additional erosion for the simulation period throughout
all scenarios. For clarity, these values are of soil transported
within the catchment, which is not necessarily equivalent to soil
removed from the catchment (because of redeposition). Most of
this soil is, in fact, re-deposited in other localities within the
catchment. Redeposition of this removed soil is not calculated here.
In the SC1, the erosion yields with the same trend as the pattern of
LUCC. The minimum erosion is at the beginning of the scenario,
and then increases gradually following the curve of LUCC. From
this, it is clear that erosion is directly related to LUCC, as is the
case in all the scenarios.
In general, erosion variation is much larger in all scenarios than
OF variation, ranging from 2 to 10%. In SC1, percent variation of
E is very similar to the percent variation of OF in the same
208
scenario. The relation between E and OF in SC1 is high, but the
erosion sensitivity changes strongly in the final few iterations of
this scenario. The erosion sensitivity in SC1 ranges from 25 to 60
(Figure 4.36). In this scenario while OF sensitivity decreases in the
last iteration, erosion sensitivity increases markedly. The erosion
sensitivity oscillations indicate that there are some differences
between the physical properties of the deforested areas in each
iteration. The topographic variables most related (see Figures 4.36
and 4.37) with erosion sensitivity in this case (SC1) are slope and
altitude. The magnitude of variation in topographic variables is not
as strong as the apparent variation in erosion sensitivity.
For SC2, the erosion yield is very similar to the LUCC pattern,
which is not the case for OF yield. The percentage variation of E
between iterations is similar to the LUCC pattern, but with a higher
rate of decrease in the first half of the iterations, followed by a
lower rate which then reduces to zero. The erosion sensitivity in
SC2 ranges from 12 to 38; the highest values are at the beginning
of the scenario, and then decrease gradually without significant
changes until iteration 16. For the last two iterations there is an
increase in erosion sensitivity (see Figure 4.38). The differences
between OF and E sensitivities are that OF sensitivity is small at
the beginning of the scenario and increases at the end whilst the
opposite is true for E sensitivity. The sensitivities of OF and E in
SC2 show that the initial deforestation near to the river channels
produces small changes in the percentage variation of OF, though
the sensitivity does not change too much. By comparing E
sensitivity with the mean topographic variables of the deforested
area (Figures 4.38 and 4.39) it can be seen that the most similar
behaviour is produced in slope, which has the same trend
throughout the iterations. The mean aspect of deforested area
does not change very much through the simulation.
209
Figure 4.37. Mean topographic variables for deforested areas in SC1
D eforested area and m ean altitude in S C 1
0
100
200
300
400
500
0 5 10 15 20 25Iteration
1400
1900
2400
2900
D eforested areaA ltitude
M ean slope and aspect of defo rested area in
SC 1
0
100
200
300
0 5 10 15 20 25Iteration
0
10
20
30
40
50
60
Aspect
Slope
M ean to p.index and distance to rivers in S C 1
6
7.5
9
10.5
12
0 5 10 15 20 25Iteration
0
3
6
9
12
15
18
Topographic Index
Distance to river
209
Figure 4.36 Erosion sensitivity in scenario 1 (deforested pattern with cellular automata)
T otal ero sio n by iteratio n
70
80
90
100
0 5 10 15 20 25Iteration
% variation of erosion
betw een iteratio ns
0
2.5
5
7.5
10
12.5
0 5 10 15 20 25Iteration
S ensitivity o f Ero sio n
0
20
40
60
0 5 10 15 20 25Iteration
210
D eforested area and m ean altitude in S C 2
0
100
200
300
400
0 5 10 15 20Iteration1400
1900
2400
2900
D eforested area
A ltitude
M ean slope and aspect of defo rested area in
SC 2
0
100
200
300
0 5 10 15 20Iteration
0
10
20
30
40
50
60
Aspect
Slope
M ean to p.index and distant to rivers in SC 2
6
7.5
9
10.5
12
0 5 10 15 20Iteration
0
3
6
9
12
15
18
Top. index
D istance to rivers
Figure 4.39 Mean topographic variables of deforested areas in SC2
210
Figure 4.38 Erosion sensitivity in scenario 2 (forest conversion with horizontal a fixed distance from river channel uphill direction)
T o tal ero tion by iteration
70
80
90
100
0 5 10 15 20Iteration
% variatio n o f ero sio n
between iteratio ns
0
2
4
6
8
10
12
0 5 10 15 20Iteration
S ensitivity o f ero sio n
0
20
40
60
0 5 10 15 20Iteration
211
However, there is a large increment with significant changes to
percentage variation early in the LUCC iterations. This is also the
case for erosion sensitivity. The highest erosion sensitivity to
LUCC in this scenario is in deforested areas close to the river
channels (in the first iterations) and this is where most of the
erosion is produced in the catchment. This is the opposite of OF
sensitivity, which is high at the top of the catchment, that can be
due high mean slope values in these deforested areas.
In SC3 the erosion yield by iteration does not increase a lot in the
first 12 iterations (2 mm) and the E percent variation remains
equal (Figure 4.40). Through within these iterations the erosion
sensitivity varies highly (15 to 33). Then it remains low until
iteration 7, then, between iterations 7 to 11, where the E sensitivity
increases again to its highest value (33) after which it oscillates
once more, at a higher level of sensitivity. As in the SC2, the
topographic variable most related with S-E in SC3 is slope,
showing the same trend and pattern but with more exaggerated
changes. Topographic index in the first iteration shows the
opposite trend, but in the last iteration it is the same as erosion
sensitivity (see Figure 4.41). Distance to the rivers is opposite in
trend to erosion sensitivity and this suggests that, erosion
sensitivity is highest when the areas closest to the rivers are
deforested. Mean altitude of deforested area in this case does not
have any bearing on erosion sensitivity. Overall this scenario has
the lowest erosion sensitivity values. By taking into account both
SC2 and SC3 it can be concluded that the areas within 150m of
the rivers are very important for E, in magnitude, variation and
sensitivity. The areas farthest from the river channels do not
increase the E much, but they have a big influence in decreasing
the erosion sensitivity. In the middle areas the erosion sensitivity
increases once more. Slope and topographic index seem to be the
most significant control on erosion sensitivity.
212
D eforested area and m ean altitude in S C 3
0
100
200
300
400
0 5 10 15 20Iteration
1400
1900
2400
2900
D eforested area
M ean altitude
M ean slope and aspect of defo rested area in
SC 3
0
100
200
300
0 5 10 15 20Iteration
0
10
20
30
40
50
60
Aspect
Slope
M ean to p.index and distant to rivers in SC 3
6
8
10
12
0 5 10 15 20Iteration
0
3
6
9
12
15
18
Top. Index
D istance to river
Figure 4.41 Mean topographic variables of deforested areas in SC3
212
Figure 4.40 Erosion sensitivity in scenario 3 (forest conversion with horizontal a fixed distance towards channel rivers downhill direction)
T otal ero sio n by iteratio n
70
80
90
100
0 5 10 15 20Iteratio n
% variatio n o f ero sio n
betw een iterations
0
1
2
3
4
5
6
7
8
9
0 5 10 15 20Iteration
S ensitivity o f ero sio n
0
20
40
60
0 5 10 15 20Iteration
213
For SC4, erosion yield remains fairly constant through all
iterations of LUCC. The percent variation ranges between 0 and
3%. It increases constantly through the range until the maximum
value (3%) in the eighth iteration. This then decreases in similar
proportions until the end of the scenario (Figure 4.42). Erosion
sensitivity has similar behaviour to the deforested area pattern
(Figure 4.43), with increments through iterations 1 to 9, and then
remaining high until iteration 12. This then decreases until near
to the initial value within the last three iterations. Erosion
sensitivity ranges between 15 to 36. The maximum values of
erosion sensitivity are in deforested areas between 2000 to 2400 m
of elevation (see Figure 4.43). Despite the OF sensitivity increasing
in the last iteration, erosion sensitivity decreases in the same
iterations. Topographic index and distance to river show opposite
trends to those noted to erosion sensitivity. Erosion sensitivity
decreases after the deforestation occurs up to 2400 m of elevation;
in this part slope, aspect, deforested area, and topographic index
decrease too.
For SC5, erosion yield is almost constant throughout all iterations
(Figure 4.44). The percentage variation is high for the first 5
iterations, reaching the maximum value (3.2%) in iteration 3, then
decreasing consistently to zero. Erosion sensitivity also starts with
high values, increasing up to the third iteration with a maximum
value of 42, then decreasing at the constant rate. This increase is
related to the variation of mean slope and aspect of the deforested
areas (see Figure 4.45), which increase until the same iteration.
Then those topographic variables remain more or less constant.
The mean altitude of the deforested area decreases during the
simulation process, which could relate to the decreasing trend of
erosion sensitivity after the third iteration. In that period, mean
distance to rivers of the deforested area remains more or less low
and constant after it had decreased in the early iterations,
214
D eforested area and m ean altitude in S C 4
0
100
200
300
400
0 5 10 15Iteration
1400
1900
2400
2900
D eforested area
M ean altitude
M ean slope and aspect of defo rested area in
SC 4
0
100
200
300
0 5 10 15Iteration
0
10
20
30
40
50
60
Aspect
Slope
M ean to p.index and distant to rivers in SC 4
6
8
10
12
0 5 10 15Iteration0
3
6
9
12
15
18
Top. Index
D istance to river
Figure 4.43 Mean topographic variables of deforested areas in SC4
214
Figure 4.42 Erosion sensitivity in scenario 4 (forest conversion with a fixed distance of altitude, in uphill direction from the lower to the highest point)
T o tal ero sio n by iteratio n
70
80
90
100
0 5 10 15Iteration
% variatio n o f ero sio n
betw een iteratio ns
0
2
4
6
8
10
0 5 10 15Iteration
Sensitivity of erosion
0
20
40
60
0 5 10 15Iteration
215
Figure 4.45 Mean topographic variables of deforested areas in SC5
D eforested area and m ean altitude in S C 5
0
100
200
300
400
0 5 10 15Iteration
1400
1900
2400
2900
D eforested area
M ean altitude
M ean slope and aspect of defo rested area in
SC 5
0
100
200
300
0 5 10 15Iteration
0
10
20
30
40
50
60
Aspect
Slope
M ean to p.index and distant to rivers in SC 5
6
8
10
12
0 5 10 15Iteration
0
3
6
9
12
15
18
Top. Index
D istance to river
215
Figure 4.44 Erosion sensitivity in scenario 5 (forest conversion with a fixed distance of altitude, in downhill direction from the highest to the lower point)
T o tal ero sio n by iteratio n
70
80
90
100
0 5 10 15Iteration
% variatio n o f ero sio n
betw een iteratio ns
0
2.5
5
7.5
10
0 5 10 15Iteration
S ensitivity o f ero sio n
0
20
40
60
0 5 10 15Iteration
216
and mean distance to rivers of the deforested area does change in
the same trend as erosion sensitivity.
Comparing erosion sensitivity with OF sensitivity in SC5, shows
little similarity; OF sensitivity changes drastically at the beginning
of the scenario, with small changes in erosion sensitivity. After
iteration 10, erosion sensitivity shows changes where as OF
sensitivity remains constant.
As in OF sensitivity, multiple linear correlation coefficients were
computed for erosion sensitivity analysis, as the dependent
variable and the topographic variables as the independent
variables. Table 4.21 summarises this analysis. The scenario that
produces the highest coefficient of determination is SC5 (R =
0.995), and the lowest is SC1 (R = 0.88), with slope and altitude
variables as the most correlated with erosion sensitivity. The
variation of erosion sensitivity (modelled statistically with b values)
due to the linear combination of topographic variables (R2) ranged
between 0.998 (SC5) to 0.942 (SC1) (see table 4.21). The critical
values of F significance for erosion sensitivity is the same as the
computed in OF sensitivity for the scenarios (1, 89). None of the
computed F significance in any scenarios surpassed the critical
value, which means that the independent variables are not
correlated between each other. The probability that R would have
fortuitously occurred if the null hypothesis held true was less than
0.05 in all scenarios.
The t-critical value assuming α=0.5 computed for erosion
sensitivity is the same, for SC1 is 2.131with N=15, for SC2 and
SC3 is 2.201 with N=11, and for SC4 and SC5 t-critical is 2.306
with N=8. If the computed t-value exceeds the t-critical value, the
variable is a significant contributor in explaining the dependent
variable when it is used in combination with the other variables.
217
217
Table 4.20 Summary of data used in Erosion sensitivity analysis
F value (5,15) 23.422 (5,11) 98.267 (5,11) 34.47 (5,8) 96.258 (5,8) 326.52F
significance1.4E-06 9E-09 2.3E-06 6.3E-07 5.1E-09
Std. Errorof estimate
0.038N = 21
0..013N = 17
0.019N = 17
0.012N = 14
0.010N = 14
t(15) Sig. t t(11) Sig. t t(11) Sig. t t(8) Sig. t t(8) Sig. tAspect -0.38 0.71 3.87 2E-3 3.53 4E-3 -0.51 0.07 1.93 0.09Slope 3.97 1E-3 4.25 1E-3 3.68 3E-3 3.19 0.01 5.58 5E-4
Soil texture parameters and soil porosity for the pasture plot were
extracted from eight field samples of 10cm depth in this plot, until
80cm depth which was the limitation of the instrument. Initial soil
moisture was derived from a previous model run as described. The
value for soil depth was assumed 250 mm of depth because that is
the depth where the sensors that measure the soil matric potential
and soil moisture are located and in order to compare like with
like, one must ensure that the model is simulating a similar
volume to that being measured.
225
4.7.4 Validation of net solar radiation
Modelled net solar radiation shows similarities with measured net
radiation. Figure 4.46 shows both modelled and measured hourly
net radiation for the simulation period. As it was expected, there
are some differences between modelled and measured values,
which occur throughout the simulation because of the effects of
clouds, which are stochastic in the model, however both take the
same pattern. Figure 4.47 shows the agreement between modelled
and measured hourly net radiation. The coefficient of
determination (R2) is 0.71, which means that 71% of measured net
solar radiation is explained with modelled net solar radiation, with
the assumption that the variables have a normal distribution, this
is statistically significant at the 99%. The correlation coefficient is
0.84 (n=550), which shows the level of association between
measured and modelled net solar radiation. Figures 4.48 and 4.49
show the diurnal pattern of hourly average net radiation for the
pasture site (modelled and measured). The relationship between
them, which has a coefficient of determination (R2) is 0.86, which
means that 86% of measured net solar radiation is explained by
the modelled net solar radiation, with the assumption that the
variables are normally distributed, and it is statistically significant
at 99%. The coefficient of correlation is 0.92. The agreement
between measured and modelled net solar radiation is better in the
diurnal hourly average than the simple observations, because on
using the average the stochastic variation in cloud cover is no
longer important.
170
Figure 4.46 Modelled and measured solar netradiation for validation
Hourly net radiation modelled and measured
0
1
2
3
0 100 200 300 400 500
Time (hour)
Net
rad
iatio
n (M
J)
Modelled
Measured
Figure 4.47 Linear regression betweenmodelled and measured net radiation invalidation.
Relationship between modelled and measured hourly net radiation
y = 0.68 x - 0.04R2 = 0.71R=0.84
RMS=0.35n=550
0
0.5
1
1.5
2
2.5
3
3.5
0 0.5 1 1.5 2 2.5 3 3.5
Modelled hourly net radiation (MJ)
Mea
sure
d ho
urly
net
rad
iatio
n (M
J)
226
227
Figure 4.48 Hourly average solar net radiation during the day for validation
Comparison between hourly average of modelled and measured solar net radiation
0
1
2
3
0 5 10 15 20Time (hours during the day)
Sol
ar n
et r
adia
tion
(MJ)
Measured
Modelled
Figure 4.49 Linear regression between hourly average of solar net radiation modelledand measured for validation
Relationship between hourly average net radiation
y = 1.05x + 0.54
R2 = 0.86R = 0.92n = 12
RMS = 0.580
1
2
3
0 1 2
Hourly average measured net radiation (MJ)
Hou
rly a
vera
ge m
odel
led
net
radi
atio
n (M
J)
228
4.7.5 Validation of soil moisture
Soil moisture is one of the most important variables in the
hydrological model, because it is the means of hydrological control
of the atmosphere-vegetation-soil interface and the result of the
processes of the soil water balance. Validating soil moisture
indirectly allows validation of other components of the hydrological
cycle such as net rainfall, evaporation, infiltration, recharge and
overland flow, which exercise control on soil moisture; so the soil
moisture validation could be interpreted as the validation of the
outcome of all of these processes.
Soil moisture was evaluated with the same input data set as net
solar radiation. Parameters for this validation are shown in section
4.7.3. Figure 4.50 shows modelled and measured soil moisture for
the validation period. Shape and trends of both graphs are quite
similar. Rainfall is shown in the same graph and assuming linear
correlation between rainfall events and soil moisture the response
is clear.
Figure 4.51 shows the linear regression between modelled and
measured soil moisture. The coefficient of determination (R2) is
0.84, and is statistically significant at 99%, which means that 84%
of measured soil moisture can be explained by the modelled soil
moisture, assuming that the variables are normally distributed.
The correlation coefficient is 0.91 (n=550), which indicates the
association level between measured and modelled soil moisture are
correlated in 91%. Figure 4.51 shows two important things: (i)
There is an associated dependency between consecutive
observations (consecutive time steps), and (ii) a longer validation
period is required in order to produce a more robust test of model
accuracy. Statistically significant at 99% the soil moisture
229
validation gives enough confidence to conclude that the basic soil
hydrological fluxes are reasonable represented in the model
balance.
Soil moisture validation as well was carried out using daily average
for the same period. Figure 4.52 shows the agreement between
modelled and measured soil moisture in trend and shape. Figure
4.53 shows the agreement between daily average modelled and
measured soil moisture through the simulation period for
validation. The coefficient of determination (R2) is 0.84, which
indicates that the modelled daily average soil moisture can explain
84% of the measured daily average soil moisture, assuming that
the variables are normally distributed. The correlation coefficient
is 0.92 (n=21), indicating the association level between variables.
Comparing hourly and daily analysis, the daily linear regression is
not much better than the regression for hourly time step.
230
Figure 4.50 Modelled and measured soil moisture for validation
Soil moisture validation
35
36
37
38
39
40
41
42
0 100 200 300 400 500 600Time (hours)
Soi
l moi
stur
e (%
)
0
0.5
1
1.5
2
2.5
3
3.5
4
Rai
nfal
l (m
m)
rainfall
Measured
Modelled
Figure 4.51 Linear regression of modelled and measured soil moisture for validation
Linear regression of measured and modeled soil moisture
y = 1.45x - 0.17R2 = 0.84R = 0.91
RMS = 0.63n = 550
0.34
0.36
0.38
0.4
0.42
0.36 0.38 0.4 0.42
Modelled soil moisture (%)
Mea
sure
d so
il m
oist
ure
(%)
231
Figure 6.52. Daily soil moisture comparison between modelled and measured, forvalidation, in July of 1999
Comparison of daily soil moisture
35
36
37
38
39
40
41
0 5 10 15 20 25 30
Time (days in July of 1999)
Soi
l moi
stur
e (%
)
Modelled
Measured
Figure 4.53 Linear regression between measured and modelled daily soil moisture,for validation
Relationship between daily modelled and measured soil moisture
y = 1.28x - 0.11R2 = 0.84R = 0.92n = 21
RMS = 0.490.34
0.36
0.38
0.4
0.42
0.36 0.38 0.4 0.42
Modelled soil moisture (%)
Mea
sure
d so
il m
oist
ure
(%)
232
Chapter V Summary, conclusions and further work
5.1 Summary of the key findings in the thesis
A few important advances have been made throughout this thesis,
these they are:
- The application of advanced dynamic modelling techniques to
TMCF.
- The development of a new GIS-based 2.5D hydrological
distributed model for tropical montane environments (TMEs).
- The compilation of hydrological flux data from a series of
experimental plots in a TME in Colombia.
- The combination of LUCC scenarios with a distributed
hydrological model to access the impact of LUCC.
- The identification of the importance of topographic properties on
the flux variations with LUCC.
- The strong relationship discovered between the pattern of LUCC
and erosion sensitivity to it.
- The identification of the most sensitive areas to LUCC within the
catchment, and their relationship with the landscape physical
properties.
5.2 Conclusions and their implications
LUCC is recognised as being an important control on hydrological
processes. Therefore this study sets out to determine how LUCC
impacts on hydrological fluxes at the catchment scale in TMCF
environments. This assessment was reached through the
application of a 2.5D GIS-based hydrological model coupled with
233
LUCC scenarios. The overall achievement of this study, is not only
obtaining a better understanding of the spatial distribution of
hydrological sensitivity given the spatial pattern of LUCC, but also
going someway to provide an advanced and robust tool to help
decision makers to develop and protect the environment, and
produce a basis for further research on TMCF hydrology and the
impacts of LUCC.
The importance of LUCC on hydrological fluxes has been
highlighted in this thesis. The sensitivity analysis of the spatial
variability of hydrological sensitivity within the watershed has
identified the importance of the spatial distribution of landscape
physical properties with respect to where the LUCC occurs and the
differing levels of impact if the same LUCC is applied to different
parts of a catchment.
With respect to the aims proposed at the beginning of the thesis
(see chapter 2), all the objectives were realised but to varying
extents. Specifically:
1- Collecting hydrological data in TMCF at both watershed and plot
scale for forest and grassland land uses.
The climatic conditions and the permanent difficulties in collecting
field data impeded the collection from the weather stations of the
two years data that were proposed. Instrumental problems in
humid tropical forest have been addressed also by Manley and
Askew (1993) in a review of hydrological problems for research, and
also those problems have been addressed for TMCF environments
in particular in the reassessment carried out by Bruijnzeel (2000).
Nevertheless, several months of data were collected at the plot
scale with some interruptions for both types of land uses, to be
able to carry out this research and this highly detailed dataset is
234
unusual for TMCF studies (see table 3.3, page 71). The same
kinds of problems affect the weather station that collected the river
flow data for catchment scale analysis. As a consequence no data
were collected in this scale.
2- The development of a physically based hydrological model, which
includes the most important processes of the hydrological cycle at
the plot scale, and the implementation of this model at the catchment
scale for analysis of impacts of various spatial LUCC scenarios.
Chapter 3 discusses the hydro-meteorological and landscape
characteristics of the study area and gives a clear idea of the rather
hydrologically extreme nature (very steep slopes, very high rainfall)
of the environment where the research was carried out. In general
terms, the hydrology in Tambito is very dynamic. The permanently
high rainfall and atmospheric humidity create a climatic condition,
which is unique to TMCF. An almost permanently wet canopy
means that intercepted water is available for evaporation
throughout the day, and the atmosphere is charged with this
water. The catchment has frequent low-level cloud cover but cloud
interception was not studied in this thesis due to lack of data,
though should certainly be a subject for further research in the
area. In the same way, the high rainfall in the area enhances
catchment wetness. At the catchment scale rainfall was
distributed with a function based on data from field stations at
different altitudes. The extrapolation of these data to the extremes
of the catchment meant that these areas received an exaggerated
value of rainfall at more than 10,000 mm a year (see Figure 3.38.
page 144). Whilst the rainfall parameterisation requires
improvement, this is not possible without many more stations.
Overland flow produced by heavy and persistent rainfall is of high
frequency and magnitude in Tambito and this is replicated in the
235
model, as is the resulting potentially high soil loss in non-vegetated
areas on steep slopes.
The model prediction accuracy is very dependent upon the
parameters used for the land use type (see 1D sensitivity analysis,
section 4.3), and the rainfall distribution function used at the
catchment scale. The hydrological model responds to different land
use type through the parameters, which vary between the land
uses. Since soil hydrological properties for forest and pasture were
shown to be very similar, the only parameters varying with land
use were those of the vegetation, in particular vegetation cover, leaf
area index and canopy interception capacity (which may not be as
important if cloud interception were incorporated into the model).
All of these were shown to be important parameters. At the
catchment scale the relationship between the change in land use
parameters and the soil conditions (varying with geology and
geomorphology on the basis of field measurements) was also shown
to be important (see section 4.5).
The hydrological model was validated by comparing soil moisture
at the plot scale with the 1D model results. This indicates that the
model has the ability to reproduce the hydrological balance with
sufficient accuracy. This is discussed in the validation section.
Not much research has been carried out on the impact of LUCC in
TMCF that could have reported data for comparison. Bruijnzeel
(2000) reviews the existing studies in this environment. From a
catchment close to the study area, Restrepo and Kjerfve (2000)
collected water and sediment yield data in the San Juan River
catchment, approximately 200 kilometres to the north of the
Tambito area, in the same side of western cordillera in Colombia).
From this dataset, a small sub-catchment (the Tadó river) has
similar annual rainfall (7410 mm) to Tambito and was selected in
236
order to compare the overland flow and erosion model results (see
table 5.1).
Area
(ha)
Rainfall
(mm m-2 y-1)
Overland flow
(mm m-2 y-1)
Erosion
(t ha-1 y-1)
Tambito 1411 7325 3835 60.7
Tadó 160000 7410 5144 15.7
Table 5.1 Overland flow and erosion model results for the original
vegetation (from Landsat TM, 1989)comparison with other research.
Despite the large difference in catchment size the results, on a unit
area basis, are comparable. Runoff is less for Tambito catchment
that this reflects the fact that there is more forest than in the Tadó
catchment. The soil erosion is much higher in the Tambito
catchment than the Tadó catchment, which reflects the steeper
slopes, despite the fact that the vegetation cover used in Tambito
catchment was the current (from NDVI, Landsat TM, 1989) and
thus has more forest cover than the Tado. Also, and importantly
the soil erosion in the Tambito study is soil flux from cell to cell
with no function for redeposition whereas the field measurements
are for soil loss from the catchment measured as sediment yield so
the two are not directly comparable.
On average by m2, the total increment of overland flow due to
LUCC (total deforestation in the catchment), in a year-long
hydrological simulation for is 100 mm m-2 yr-1 (2% of the total), and
in erosion is 23 mm m-2 yr-1 (22% of the total). In terms of
catchment totals, these are an increments of 14,110 m3 of water
per year in overland flow and 2245 m3 of removed soil by erosion
from the pristine initial position. This indicates that LUCC can
produce very serious consequences within and outside the
catchment because based on the model results, the erosion is
237
much more sensitive to LUCC than overland flow, but that impacts
can be minimised if land use change occurs in areas with low
sensitivity to LUCC.
Although, model results must be used with care, modelling at the
catchment scale allows us to define areas where runoff and erosion
is increased significantly by LUCC. The model limitations mainly
result from the parameters used, because those parameters are
derived from the land use types present in the Tambito watershed.
The model was designed for a TMCF environment, but with several
adjustments it could be used for other landscapes and
environments, although De Roo (1993) and Lorup et al. (1998) in
discussed this idea order to prevent large errors. The 1D model
can more readily be applied to other landscapes. An evident
limitation of 2.5D model is its spatial resolution, because the cell
size of this study was fixed to 25m pixels which according to the
available data and the type of geomorphology was the most suitable
to make the model operational. To improve this spatial resolution
requires more and better data.
In order to apply the model to different areas, several
considerations must be taken into account:
- The scale of integration needs to be related to the landscape
where the model is applied. Spatial resolution has to reflect the
required level of detail in the model in order to produce
reasonable results (see sections 3.3 and 3.6). In the same way,
cell size needs to be considered and adapted to fit the
landscape, and the computationally efficient operation of the
model.
- Vegetation type and land cover need to be parameterised
carefully within the model, in order to be representative and to
reproduce reasonable results (see section 4.3).
238
- The geomorphology of the landscape has to be taken into
account, in order to define the spatial resolution, which must
represent landforms at the level at which the study is carried
out.
- For larger areas, changes in resolution (large pixels) and scale
must be considered carefully, because model adjustments are
required in order to use the model with larger cells (i.e. erosion
sub-model).
There are two important features derived from the catchment scale
model: the effects of cell connectivity and the effects of landscape
variability on the catchment scale response to spatially distributed
LUCC. Connectivity in the model produces an increment in
overland flow in downslope areas due to the accumulation of water
along flow lines. Additionally, this increase depends upon the
distribution of land cover along flow lines and the spatial variability
of landscape properties along the same lines.
3- The parameterisation of the hydrological model at the plot scale
and identification of the most important parameters influencing
hydrological sensitivity to LUCC (see chapter 4).
From modelling at the plot scale (section 3.5.1), parameters, which
change with LUCC and create the greatest sensitivity in runoff and
erosion for this particular catchment and environment were
highlighted. The most sensitive parameters are vegetationcover, soil depth, porosity and the parameters of the erosionequation. Changing land use produces significant changes in
water fluxes (see sensitivity analysis, chapter 4). Soil depth is the
main control on the amount of soil water storage, which plays an
important role in the catchment water balance.
239
Hydraulic soil properties (such as hydraulic conductivity and
matric potential) are linked directly to the most sensitive soil
parameters used in this model, that were also identified by Kirkby
(1978), as having a critical role dominating surface water
processes. Also Ternan et al. (1987) and Elsenbeer and Cassel
(1990) in Grenada and Western Amazonia respectively emphasise
that soil hydraulic properties are an integral part of hillslope
hydrology in the tropical forest. Particularly, the permeability of
the soil in combination with the topography (steep slopes) and the
soil depth are the control of hillslope responses, with a high rate of
infiltration, with the exception of during extreme rainfalls
(Bruijnzeel, 1990), where the generated saturation overland flow is
considered the principal delivery mechanism to the rivers, in
response to steep and concave slopes combined with heavy storms
(Nortcliff and Thornes, 1981), that are a characteristic of TMCF.
The top soil layer (0.2m) is where the soil hydrology controls the
infiltration and consequently the saturated overland flow as was
described in the sensitivity analysis to the parameter soil depth
(see section 4.3.9), as is highlighted by Bonell et al (1983),
Elsenbeer and Cassel (1990) and Bruijzeel (1990). In addition, this
zone is often the most porous of any in the soil profile, with high
available soil water storage capacities. The saturated infiltration
rates ranged from 5 to 12 mm h-1 for undisturbed rain forest, as
was argued by Wierda et al. (1989) in a tropical rain forest in Coté
d’Ivoire, West African, and compares with that calculated for
Tambito (6 mm h-1) with the pedo-transfer function using the soil
texture parameters.
The most sensitive parameter related to the land use type is
vegetation cover (see section 4.3.6). Vegetation cover affects
evaporation increasing it with almost linear trend as forest
vegetation cover increases, and erosion increases in an exponential
240
way as forest vegetation cover decreases (see Figure 4.18). Under
30% forest vegetation cover, erosion sensitivity does not increase at
all, because the remaining forest areas are not significant for
erosion sensitivity as much as the complementary areas of
vegetation cover (between 80% to 30%). Other parameters change
with vegetation type change in the model, but do not produce large
effects on hydrological fluxes as does vegetation cover does.
Burt et al. (1993) came to similar conclusions after analysing
several studies where LUCC can affect the variations in runoff and
erosion, despite the fact that they were carried out on different
environments to this one (North California USA and Plynlimon mid-
Wales), where the experimental catchments were forested rather
than deforested. Those experiments showed that afforestation
reduced the runoff by 25%, this loss being attributed to increasing
water loss by evaporation due to changes in vegetation properties
that are involved in the rainfall interception process. Similarly,
decreases in evapo-transpiration and thus of increasing runoff with
the reduction of forest cover were identified by Bosch and Hewlett
(1982) and Calder (1992).
The increase of erosion with changes in vegetation cover,
particularly deforestation were also identified by Bruijnzeel (1990),
Falkenmark and Chapman, (1989), who argued that forest
conservation can prevent the occurrence of landslides in similar
slopes to those in the study area.
The soil erosion model seems to be very sensitive to the vegetation
cover parameter, and also to the erosion parameters involved in the
erosion equation of the model, such as erodibility factor. This also
has been recognised by Govindaraju (1998) in his study of effective
field scale values for shear stress and soil erodibility and their
spatial variability for physically-based models.
241
Others, mainly soil parameters, were classified as sensitive
parameters, but do not change with change in land use type within
the model (such as the assumption of one type of soil for the whole
catchment, see section 3.5) based on field measurements. Soils are
much more related to the geology and geomorphology of the area.
Within the modelling process those parameters remain constant
between land cover types, so basically the effect of LUCC is applied
in the model at the catchment scale as a change of the vegetation
cover on the catchment surface.
4- To develop methods for model parameterisation at the catchment
scale
One of the difficulties in the generalisation of model parameters at
the catchment scale, is how they vary throughout the surface. In
this respect, the parameters changing on the catchment surface
are the parameters related to the land use type. Land use type in
the catchment was identified and grouped into two classes in this
study (forest and grassland), and for each land use type
parameters were determined by field sampling (see sections 3.2,
3.3 and 3.5). To assign those parameters throughout the
catchment surface the NDVI was extracted from Landsat TM and
was used to distinguish between the main land covers as the most
suitable and economic way to collect land cover information over
the catchment, and with iterations of the scenarios through the
simulation process.
Vegetation parameters such as cover, LAI and canopy storage
capacity also vary within a cover class in response to species
variability and altitudinal change but this level of complexity could
not be taken into account in this study.
242
Model validation at the plot scale
Soil moisture was used in the model validation process, because all
hydrological fluxes contribute to the soil moisture balance. The
agreement between modelled and measured was significant (r2 =
0.83, RMS = 0.63 at one hour time resolution, and r2 = 0.83, RMS
= 0.49 as a daily average), which provides a degree of confidence in
the behaviour of the hydrological model. Unfortunately validation
was carried out over a small period due to a lack of availability of
uninterrupted good quality data.
Validation at the catchment scale was not carried out because data
at this scale were not available; but the modelling suggests that
there are significant effects of LUCC at the catchment level,
necessitating the improvement of the field methodology for data
collection at this level.
5- The application of the model to identify the location of areas
within the Tambito catchment where hydrological processes are most
sensitive to LUCC in relation to physiographic properties, combined
with different scenarios of LUCC.
Burt et al. (1993) recognised the difficulties of identifying the areas
within a catchment that are sensitive to LUCC, and they suggest
the use of hydrological simulation models for this purpose. Such a
simulation model has been developed and used for the stated
purpose here.
The combined process of sensitivity analysis in the 1D and 2.5D
models, helps us to understand first of all, which parameters
within the model are important for surface hydrological fluxes and
secondly, the properties of spatial variability of the sensitive areas.
243
Using a combination of different scenarios indicates how landscape
sensitivity responds to the pattern of deforestation. From the
analysis of the impact of LUCC, the areas within the catchment
that should be protected from LUCC can be identified, because
those areas with greatest sensitivity can, if deforested, lead to more
serious environmental consequences, such as soil degradation,
erosion and sedimentation (see Figures 4.11 and 4.12). In the
same way, hydrological models offer the possibilities to evaluate the
outcome of particular LUCC strategies. The combination of LUCC
scenarios with a distributed hydrological model also was used by
Mulligan et al. (2000) finding serious effects of progressive
deforestation on runoff and erosion yields. A similar process using
the combination of hydrological models and statistical tests was
carried out by Lorup et al. (1998), who highlight the ability to
improve the analysis of the impact of LUCC on the catchment
runoff, compared to using them separately. On the other hand,
Fahey and Jackson (1997) used a comparison approach for
catchments with different vegetation (forest, grasslands, and pine
plantations) to estimate the differences in hydrological properties.
The LUCC scenarios developed in this thesis are not intended to
represent real patterns of land use change, rather they serve as a
means of testing the sensitivity to landscape physical properties
(table 4.18 to 4.21), which the LUCC patterns produce hydrological
variation in the catchment. The relationship between landscape
properties and hydrological flux variation were also studied by
Quine and Walling (1993), where the erosion rate predictions were
derived from a statistical function of a combination of topographic
attributes. Those types of landscape attributes have also been
used to predict the soil properties with good results (McKenzie and
Ryan, 1999).
244
A reduction in the catchment forest cover in the humid tropics
produces an increment in overland flow and a decrease in water
evaporation (see sections 4.3 and 4.5), that also was identified by
Bosch and Hewlett (1982), Bruijnzeel (1990) and Fahey and
Jackson (1997), among others researches. They highlight that
forest acts as a sponge, so that forest conversion increases
flooding. Unfortunately cloud interception was not taken into
account, though this is one of the processes that can affect the
hydrological balance in montane cloud forest, where may
compensate for the extra water evaporation identified in conversion
to forest (Zadroga, 1981). Also with this respect, Calder (1998)
associates the changes in evaporation rates with the variations on
large leaf surface area and the deeper root system of forest.
The pattern of variation in erosion within the modelled scenarios is
very similar to the LUCC pattern, which indicates a strong relation
between LUCC and hydrological change that is driven by vegetation
cover change.
The importance of the landscape physical properties is identified
from modelling at the catchment scale. Slope, aspect, elevation,
and distance to rivers of the deforested area, among others,
produce an important effect on the hydrological fluxes from those
same areas. Those properties, combined with the location of the
deforested areas relative to the hydrological flow-routing network
are critical to the hydrological response to land use change.
Modelling at the catchment scale combined with the sensitivity
analysis helped to identify the topographic characteristics of the
areas most sensitive to LUCC within the catchment (see section
4.5). Those characteristics are:
245
- steep slopes, within 150m distance to rivers and with highelevation (see Slope Map, Figure 3.17, page 81).
- steep slopes furthest away from river channels, which areclose to the boundary of the catchment, where rainfall ishighest (see Elevation Map, Figure 3.16, page 80).
The same land use change can have quite different effects on the
hydrological cycle if it occurs in the high parts of the catchment as
opposed to the lower parts, or at the top of a slope as opposed to its
base.
5.3 Further research
The validation process carried out in this thesis used short periods
of data. It is necessary to collect more field information and re-
validate the model, in order to check the model predictions more
fully.
The initial aim was to produce a simple framework model for this
landscape (TMCF) where the main hydrological events were
represented, and also that the model could operate with little
information. Therefore, there is now a need to incorporate other
modules or improve some of the existing one. The improvements
required include:
- Adjusting the model rainfall equation in order to produce more
realistic results through the acquisition of more spatially
detailed field rainfall data.
- The incorporation of cloud interception by trees to integrate
additional water in the hydrological cycle. Also the interception
module needs to be calculated at shorter time resolutions, in
246
order to assess with more accuracy the loss of intercepted
water.
- Modelling soil deposition and sedimentation. Erosion in the
current module is modelled only as soil detachment; this soil
will be re-deposited or incorporated to the river flux as
suspended sediment.
- Plant transpiration routine. Evapotranspiration could have a
more significant role in the hydrological cycle in TMCFs than
assumed here. As the variation in altitude within the
catchment is significant (more than 1000m) and in the same
way the environmental conditions and vegetation change.
Plants could have a complex and differing response through the
altitudinal range.
- Throughflow may also be more important at the catchment scale
for hydrology in this particular environment (TMCFs), than
assumed here by its non-inclusion.
- Changes in soil properties produced by LUCC could be usefully
included for catchments or LUCC's where LUCC impacts on soil
as well as vegetation properties.
In terms of land use change
- To incorporate dynamic scenarios where the LUCC forces to
change in soil properties (such as porosity and bulk density) as
well as plant properties.
- Allow the capacity for forest growth and regeneration,
incorporating plant physiology processes within the model.
247
In terms of modelling
- There is still a gap between modelling activities and GIS
interfaces. The spatial representations of some processes are
not simple and require better GIS tools.
- The relationships between hydrological fluxes and physical
landscape properties have been shown through this thesis. On
the basis of the analysis, a statistical model could be
developed to identify the sensitive areas (in terms of overland
flow and erosion) within catchments with similar characteristics
and land use to the ones in which the model is developed. On
the basis of catchment physical properties such as those
studied here (slope, aspect, topographic index, altitude, and
distance to rivers) and derived from DEMs should be used in the
statistical model. This statistical model could be a fast
alternative to evaluate the hydrological impact of LUCC in
catchments with similar characteristics without carrying out the
whole process of hydrological modelling, land use
parameterisation and sensitivity analysis developed here, using
only secondary information derived from DEMs. This is the
subject of further papers based on the information and
processes developed in this thesis.
It is recognised that this model is useful to support further
research on TMCF environments. Understanding landscape
sensitivity to land use change could also assist conservation
planning and rural agricultural sustainability. These types of
research can be used by the government and also in more local
areas, by the Corporations who have the responsibility to manage
natural resources within the catchments in Colombia, as a fast
alternative to assess the LUCC impact of proposed plans of
catchment management.
248
Bibliography
AARAM Andean Amazon Rivers Analysis and Monitoring project.
1998. http:boto.ocean.washington.edu/aaram/
Abbott, M. B., Bathurst, J, C., Cunge, J. A., O’Connell, P. E., and
Rasmussen, J. 1989. An Introduction to the European System –
Systeme Hydrologique Europeen, “SHE”, 2. Structure of a
physically-based, distributed modelling system. Journal of
Hydrology, 87: 61 – 77.
Ahuja, L. R., Naney, J. W., and Williams, R. D. 1984. Estimating
soil water characteristics from simpler properties or limited data.
Soil Science Society American Journal, 49: 1100 - 1105.
Ahuja, L. R., and Williams, R. D. 1991. Scaling water
characteristic and hydraulic conductivity based on Gregson-
Hector-McGowan approach. Soil Science Society American Journal,
55: 308 - 319.
Anderson, M. G., and Burt, T. P. 1985. Hydrological Forecasting.
John Wiley & Sons Ltd. Chichester, UK.
Anderson, M. G., and Burt, T. P. 1990. Process studies in hillslope
hydrology. John Wiley & Sons. Chishester, U. K.
249
Arya, L. M., and Paris, J. F. 1981. A physicoempirical model to
predict the soil moisture characteristic from particle-size
distribution and bulk density data. Soil Science Society American
Journal, 45: 1023 - 1030.
Aston A.R. 1979. Rainfall Interception by eight small trees.
Journal of Hydrology, 42: 383 - 396.
Avery B.W., and Bascomb C.L. 1974. Soil survey and laboratory
methods. Harpenden Soil Survey of Great Britain. UK.
Banco de la República, 1951. Bases de un programa de fomento
para Colombia – Informe de una Misión dirigida por Lauchlin Currie.
Banco de la República (eds). Bogotá, Colombia.
Bathurst J. C., Wicks J.M., and O’Connell P. E. 1995. The
SHE/SHESED Basin Scale Water Flow and Sediment Transport
Modelling System. In: Computer models of Watershed Hydrology.
Edited by Vijay P. Singh. Department of Civil and Environment
Engineering Louisiana State University, Baton Rouge, LA 70803-
64405, USA.
Bejarano, J. A. 1977. Contribución al debate sobre el problema
agrario. In: El agro en el desarrollo histórico colombiano. Edutorial
Punta de Lanza, Bogotá, Colombia.
Bell, M. A. and Van Keulen, H. 1995) Division S-5 -Pedology. Soil
Pedotransfer Functions for Four Mexican Soils. Soil Science
Society. American Journal, 59: 865 - 871.
Beven, K. J., and Kirkby, M. J. 1976. Towards a simple
physically-based variable contributing area of catchment
250
hydrology. Working paper 154, School of Geography, University of
Leeds.
Beven, K. J., and Kirkby, M. J. 1979. A physically based variable
contributing area model of basin hydrology. Hydrological Science
Bulletin, 24, 1: 43 – 69.
Beven, K. J., Kirkby, M. J., Schofiled, N., and Tagg, A. F. 1984.
Testing a physically-based flood forecasting model (TOPMODEL) for
three U.K. catchments. Journal of Hydrology, 69: 119 – 143.
Beven, K. J. 1989. Changing ideas in hydrology – the case of
physically-based models. Journal of Hydrology, 105: 157 – 172.
Beven K.J., and Moore I.D. 1994. Advances in Hydrological
Processes. Terrain Analysis and Distributed Modelling in Hydrology.
John Wiley & Sons. New York. USA.
Beven K., Lamb R., Quinn P., Romanowicz R., and Freer J. 1995.
TOPMODEL. In: Computer Models of Watershed Hydrology. Water
Resources Publications. Edited by Vijay P. Singh. Department of
Civil and Environment Engineering Louisiana State University,
Baton Rouge, LA 70803-64405, USA.
Binley, A., and Beven, K. 1992. Three-dimensional modelling of
Using monochromatic vertical pictures below of canopy, wereanalysed the ratio of light – dark pixels on scanned pictures. Totalnumber of pixel by picture was 826 x 550 = 454300.
# MAO PCRaster hydrological model# Release V 1.0# October 03 2000# Dynamic model for surface hydrological fluxes# Basic analysis for Geography Ph.D. degree# King’s College London, London - UK
binding
PI = 3.141592654; # Conversion value data = dry9523.txt ; # file input data AU = 1.49597890E8 ; # {DISTANCIA AL SOL EN KM} SolarKteW = 1373 ; # { W m-2} SolarkteJ = 4921 ; # { kJ m-2 h-1} GR = 0.017453292 ; # Conversion value to rad olat = 2.5 ; # Latitude of the area olong = -76.85 ; # Longitude of the area oslope = slope.map ; # Slope map name oaspect = aspect.map ; # Aspect map name cellnum = 22780; # Number of cell in the catchment
# Parameters of interception
landuse = lucriof0.012; # Land use for this run b_canopy_drip = 0.5; # Parameter empirical value
# for drip from canopy min_drainage = 0.002; # Minimum drainage from
# canopy num_layers = 5; # Layer in the canopy canopy_retention = 0.6; # Canopy retention
iobss = iobs ; # Solar radiation at the top of the atmospherereport IobssSalida = (maptotal(iobss))/cellnum; # in Kj / m2 / hourreport SIobssSalida = timeoutput(samples.map, iobss);report iobss.map = iobss.map + iobs;
w = abs(w); # Sun angle elevation in degrees att = sin(w)**0.3333 + 0.25 * sqr(cos(w)) * normal(bol.map)*uno.map; # random values are included atenuation = att * difdtm.map * 0.0004 ; # in Kj / m2
attenuation = if(atenuation > 1 then 0.98 else atenuation );# Never it takes more than 100 %
K = 0.000002778 * (exp (pv + qv * sand.map + (rv + tv * sand.map + uv * clay.map+ vv * clay2.map) * (1 / teta))) * 1000 * 3600 ; # k and ksat are in de mm per hour
# After poundig time, infil. is proportional to ksat infiltration = if(ThroW <= 0, cero.map, infil3);
# Infiltration rate of rainfall to the soil surface mm / hour infil=infiltration;report infil.map = infil.map + infil;report InfilSal = (maptotal(infiltration))/cellnum;report SInfilSal = timeoutput(samples.map,infiltration); # Evaporation from soil moisture tot_evapo_soil = teta_antes * pot_evapo * (1 - veg_cover) ; totevapo = tot_evapo_soil + evapo_canopy;report EvapoTotSal = (maptotal(totevapo))/cellnum;report SEvapoTotSal = timeoutput(samples.map,totevapo);
#_________________________________________# Discharge drenage = K * 0.625; # en m3 : k mm/h * .m2 = m3 salida del pixel # es el area de 25 X 25 = 625 m2 * 0.001 m = m3 by cell#_________________________________________#/* Overland Flow
over1 = if(ThroW - infiltration <= 0 then cero.map else ThroW - infiltration);# OVER1 IS THE HORTONIAN OVERLAND FLOW en mm
335
over2 = teta_antesmm + infiltration - K - tot_evapo_soil; over2 = if (over2 < 0 ,cero.map , over2); # Soil water balance mm soilevapo = tot_evapo_soil;