An analysis of Lake Victoria vegetation and shoreline ...€¦  · Web viewEffect of hydrologic regime on target species occurrence 54. Effect of hydrologic regime on target species

Analysis of Lake Victoria vegetation and shoreline monitoring data

Final report

Prepared by Angus Webb, Elizabeth Martin, Joe Greet and David Kennedy on behalf of the Murray–Darling Basin Authority.

November 2012

Acknowledgements

This project was funded by the Murray–Darling Basin Authority through Research Contract MD2006. We thank Hugo Bowman, Terry Korodaj and Katie Davis for their support throughout the project. Lake Victoria Scientific Reference Panel members Jane Roberts and John Magee provided careful review of proposed approaches and analysis outputs. Wayne Stephenson and Ian Sluiter provided valuable local knowledge of the Lake Victoria monitoring environment.

Contact for further information

Dr Angus WebbDepartment of Resource Management and GeographyBuilding 379 (221 Bouverie Street, Carlton)The University of Melbourne, 3010, [email protected]

Publication information

Cover image: Revegetation, regeneration and burial protection works at Lake Victoria. Photo: Hugo Bowman

Published by Murray–Darling Basin AuthorityPrint OnlineMDBA Publication No 87/12 MDBA Publication No 88/12ISBN 978 1 922177 24 7 ISBN 978 1 922177 25 4

© Murray–Darling Basin Authority for and on behalf of the Commonwealth of Australia, 2012

With the exception of the Commonwealth Coat of Arms, the MDBA logo, all photographs, graphics and trade marks, this publication is provided under a Creative Commons Attribution 3.0 Australia License, (http://creativecommons.org/licenses/by/3.0au).

Please attribute this publication (and any material sourced from it) using the following wording:

Title: An analysis of Lake Victoria vegetation and shoreline monitoring dataSource: Licensed from the Murray–Darling Basin Authority, under a Creative Commons

Attribution 3.0 Australia LicenseAuthors: Webb JA, Martin EH, Greet J, Kennedy DMPublished: November, 2012Publisher: Murray–Darling Basin Authority, Canberra

The MDBA provides this information in good faith but to the extent permitted by law, the MDBA and the Commonwealth exclude all liability for adverse consequences arising directly or indirectly from using any information or material contained within this publication.

i

Executive Summary

Background and analyses undertaken

Permission to operate Lake Victoria as an important water storage in the southern Murray-Darling Basin is granted through a Section 90 Consent permit. Under the conditions of the Consent, monitoring programs were established for shoreline erosion and deposition and shoreline vegetation. These monitoring programs had the stated aim of investigating the mechanisms driving shoreline erosion and deposition. A major assumption stated in the Consent is that vegetation can stabilize shoreline sediments, thereby reducing erosion and promoting deposition. However, while large data sets exist for both shoreline erosion and deposition and shoreline vegetation, there has been no comprehensive attempt to synthesize these data and test this assumption.

Using a Bayesian statistical modelling approach, we undertook statistical analyses of the monitoring data. These analyses were designed to: detect any overall changes in erosion and deposition or vegetation over time and space; test the effect of hydrologic regime on five target species of shoreline vegetation and on shoreline erosion and deposition; and to test whether the presence of vegetation reduces erosion and promotes deposition. The specific analyses conducted were:

Spatial and temporal trend analysis of:o Vegetation abundance quantified as:

Total vegetation live cover Live cover of annuals and perennials Live cover of different growth forms (grasses, herbs, etc.)

o Shoreline erosion and deposition Effects of hydrologic regime on:

o Presence or absence, and live cover of five target plant specieso Shoreline erosion and deposition

Effects of vegetation live cover on shoreline erosion and deposition

Results and interpretation

We found clear patterns of increasing vegetation live cover over time (between 1998 and 2011) on the eastern and western shore of the lake. We believe this pattern is explained by the destocking of properties in these areas over the last decade, and also by lake operations during the drought (i.e. on-average reduced inundation facilitating plant growth). There were no changes in overall live cover over time on the southern shore, but the cover of perennial species increased while the cover of annual species reduced. This result is explicable in terms of lake operations during the drought. We were unable to statistically analyze the trends for different growth forms, but plots of the data showed most groups increasing over time.

ii

Total vegetation live cover increased with elevation on the eastern and western lake shores; we believe this pattern results from zonation of the different growth forms caused by hydrologic regimes. In particular, large woody species are mostly found above the lake’s full supply level (>27 m AHD). Vegetation live cover peaked at middle elevations (i.e. ~26 m AHD) on the southern shore, with patterns being driven by the more complex geomorphology of this area compared to the other two sectors. In particular, vegetation was most abundant on the ‘islands’ and barriers that characterize the southern shore. Overall, the southern shore supports far higher vegetation cover than the other areas of the lake.

Analyses of spatial and temporal trends in shoreline erosion and deposition revealed a highly stable system that does not appear to be undergoing sustained erosion. There is evidence of sediment re-distribution occurring on the eastern beach, with sediments being moved from the lower beach face to the upper, most likely caused by a combination of wave action and prevailing winds. Other sectors of the lake are stable over different elevations. The eastern and western beaches show no evidence of any change in erosion or deposition over time, with average rates very close to zero. On the southern shore, net deposition at the beginning of the sampling period dropped to zero by approximately 2004, with net erosion occurring after this. We believe this pattern is the result of management works undertaken in the late 1990s to protect exposed burials (sandbags, shade cloth, sand nourishment). The trend analysis suggests that these practices were effective for some years, but lost the capacity to promote sediment deposition as sandbags, etc. began to break down.

Analysis of the effect of three indicators of hydrologic regime (optimal conditions, submergence, exposure) on the occurrence of five key plant species (Spiny Sedge, Common Reed, Common Couch, Rat’s-tail Couch, Spiny Mudgrass) found different results for different species. However in general, history of exposure and submergence were both found to be important predictors of species occurrence and live cover. In contrast, ‘optimal conditions’ (conditions considered optimal for a species’ growth, defined through a review of the literature), proved a good predictor of live cover but was uninformative in predicting the occurrence of these species. We conclude that we were able to model the hydrologic conditions under which the target species can persist and the conditions ideal for their growth, but that our model did not consider all factors important for recruitment. Currently, data are lacking for such an analysis.

Analysis of the effect of exposure to potential wave action on shoreline erosion and deposition found the predicted negative relationship between deposition and ‘wave zone days’ on the eastern beach. In this environment, exposure to wave action mobilizes and erodes sediment from the beach face, with those sediments being moved further up the beach face in line with the findings from the trend analysis. These results provide support for the current LVOS practice of rapidly raising and lowering lake levels when the water surface is near the elevation of Historically Undisturbed Sediment (HUS). In contrast to the eastern shore, on the western and southern shores, number of ‘wave zone days’ was positively correlated with deposition. Prevailing winds at Lake Victoria mean that the western and southern shores are rarely exposed to wave action. Moreover, these environments are physically complex, and could trap sediments mobilized by inundation.

iii

Analysis of the effect of vegetation cover on rates of erosion and deposition produced counter-intuitive findings, seemingly in contradiction of the assumption in the Section 90 Consent. However, we believe this to be a spurious result caused by inadequacies of the combined monitoring programs to address this important question. The scales of data collection, mismatch between the shoreline and vegetation quadrats, and the ways in which data are collected all act to prevent a meaningful analysis. At least some of the future monitoring should be explicitly designed to address this question. For the present analysis, there was no association between vegetation live cover and erosion/deposition for the southern and western shorelines. On the eastern shoreline increased vegetation was associated with increased erosion.

Summary and recommendations

Vegetation on the Lake Victoria shore has increased over time through de-stocking of properties and on-average lower lake-levels caused by drought and altered lake operations. The analyses indicate that it should be possible to tailor lake operations to promote the occurrence and growth of a number of key plant species on the shoreline. Erosion and deposition have been stable over time. Management actions to protect exposed burials appear to have been effective, but may need to be renewed in the near future. Exposure to potential wave action does not appear to have a detrimental effect on erosion in the southern zone where large numbers of burials are located. At this time, the monitoring data are insufficient to address the question of whether vegetation can stabilize sediments and promote deposition. We stress that this is a failure of the data to answer the question rather than any indication that vegetation cannot stabilize sediments.

The existing monitoring programs are a rich source of information and the current monitoring should be retained in some form. However, future monitoring of the Lake Victoria shoreline should consider including:

1. Transects or plots that are surveyed for both vegetation and change in surface profile, with multiple randomly chosen points surveyed within a larger quadrat to allow better assessment of the ability of vegetation to trap sediments.

2. Transects or plots positioned specifically to monitor Historically Undisturbed Sediments (HUS). The shoreline monitoring data collected to date are insufficient to say anything about whether there is sustained erosion of these sediments.

3. Monitoring to quantify the rates of longshore sediment transport. This would better elucidate the mechanisms leading to sediment deposition in the southern and western sectors and sediment erosion in the eastern sector.

4. Targeted research or monitoring to further investigate the apparent positive effect of ‘wave zone days’ on deposition in the culturally-important southern shore area.

5. Extending the definition of vegetation cover to include dormant (or even dead) vegetation foliage and also root and rhizome masses. These will all act to trap sediments.

6. Monitoring of vegetation reproduction and recruitment, rather than only live cover.7. More temporally-frequent monitoring of annual species. Yearly sampling probably

misses much of the dynamics of these species.

iv

8. Improved data management through a central database or similar to reduce the amount of pre-processing required for data analysis.

v

Table of ContentsIntroduction....................................................................................................................................1

Background and motivation for project......................................................................................1

Bayesian Statistical Modelling.....................................................................................................2

Methods..........................................................................................................................................4

Data.............................................................................................................................................4

Vegetation...............................................................................................................................4

Shoreline.................................................................................................................................5

Lake level.................................................................................................................................6

Statistical models........................................................................................................................6

General....................................................................................................................................6

Vegetation trends....................................................................................................................8

Shoreline trends....................................................................................................................10

Effects of hydrologic regime on target species......................................................................10

Effects of hydrologic regime on shoreline erosion and deposition.......................................13

Effect of vegetation live cover on erosion and deposition....................................................14

Results...........................................................................................................................................16

Vegetation trends.....................................................................................................................16

Total live cover......................................................................................................................16

Annuals.................................................................................................................................17

Perennials..............................................................................................................................19

Growth forms........................................................................................................................20

Shoreline trend analysis............................................................................................................21

Effect of hydrologic regime on target species...........................................................................23

Spiny Sedge...........................................................................................................................23

Common Reed, Common Couch, Rat’s-tail Couch.................................................................26

Spiny Mudgrass.....................................................................................................................28

Effects of hydrologic regime on shoreline erosion and deposition...........................................30

Effects of live vegetation cover on shoreline erosion and deposition.......................................32

Discussion.....................................................................................................................................34

Vegetation.................................................................................................................................34

Spatial and temporal trends..................................................................................................34

Effects of hydrologic regime on target species......................................................................36

vi

Vegetation data.....................................................................................................................38

Shoreline erosion and deposition..............................................................................................39

Spatial and temporal trends..................................................................................................39

Exposure to potential wave action........................................................................................40

Effects of vegetation cover....................................................................................................42

Historically undisturbed sediments.......................................................................................45

Data processing and management............................................................................................45

Recommendations for future monitoring.................................................................................46

References cited............................................................................................................................48

Appendix A....................................................................................................................................51

OpenBUGS code........................................................................................................................51

Vegetation trend model........................................................................................................51

Shoreline erosion and deposition trend mod........................................................................53

Effect of hydrologic regime on target species occurrence....................................................54

Effect of hydrologic regime on target species cover, when present......................................55

Effect of hydrologic regime on shoreline erosion and deposition.........................................56

Effect of vegetation cover on shoreline erosion and deposition...........................................57

Appendix B....................................................................................................................................58

Literature used to develop the vegetation conceptual models.................................................58

Appendix C....................................................................................................................................60

Data plots for trend analysis of growth forms...........................................................................60

Appendix D....................................................................................................................................64

Wind rose plots for Mildura......................................................................................................64

vii

Introduction

Background and motivation for project

Lake Victoria is a natural lake near the Murray River in south-western New South Wales near South Australian border, and is the largest natural lake connected to the Murray River (Bowman & Davis 2012). The lake is connected to the Murray by Frenchman’s Creek, which drains from the southern end. The lake outlet was dammed and the lake flooded to present levels in 1928, and it has since been operated as a water storage (MDBC 2008). Flooding resulted in a permanent increase in the lake’s water level, and increased the surface area of the lake, mostly by inundating low-lying lands on the western shore. At full-supply level (~27 m AHD), the lake measures approximately 12 x 9 km and holds 675 GL. The lake is drawn down on a yearly basis to approximately 24 m AHD and 345 GL, delivering water to: guarantee South Australia’s water supply, mitigate and augment flood peaks, manage salinity of the lower Murray River, and store unregulated flows and spills from upstream (Bowman & Davis 2012).

Lake Victoria is also a site of considerable cultural importance to indigenous Australians. Archaeological evidence suggests continuous habitation of the lake shore by the Barkindji and Maraura nations for at least 18,000 years, but possibly much longer (MDBC 2008). Of particular cultural significance are the many burial sites around the lake perimeter, but concentrated on the southern shore. The operation of Lake Victoria as a water storage has been implicated as the cause of shore erosion leading to exposure of many burial sites (Bowman & Davis 2012), concentrated at approximately 23 – 26 m AHD elevation (Carlile 2010). Since 1994, there have been considerable efforts to stabilize sediments around exposed burials, with these works concentrated at around 25 – 26 m AHD elevation (Stephenson et al. 2009; Stephenson & Kennedy 2011). These include surrounding individual burial plots with borders of sandbags, and either using shade cloth or further sandbags to cover the actual burial sites. Other more general anti-erosion works include barriers of sandbags and ‘sand sausages’ at points of perceived erosion risk. These structures are designed not only to provide immediate protection to the shore, but also to promote longer term sediment deposition and eventual recruitment of vegetation to the stabilized sediments (MDBC 2008).

Lake shore vegetation also has cultural value, but is currently perceived as being degraded and in need of restoration (MDBC 2008). Lake operations that led to the lake being held at full supply level for long periods of time, coupled with pastoral grazing around much of the perimeter of the lake and grazing by feral and native species, led to the virtual denudation of the lake shore. Under these conditions, re-working of exposed lake sediments by waves that can reach up to 0.5 m in height occurs (Bowman 2011). Following drawdown of the lake in 2011, such re-working was immediately apparent during a site visit by the Lake Victoria Scientific Reference Panel (Bowman 2011).

In recognition of the importance of Lake Victoria as a water storage, but also of its cultural heritage value, Lake Victoria is operated by SA Water under a Section 90 Consent and Section 87 Permission to Operate (MDBC 2008). These permits require that lake operations be designed to minimize any damage to the cultural heritage values of the lake. The single major assumption of the Section 90 Consent relevant to this project is that re-vegetation of lake shores will stabilize

1

sediments and hence protect cultural heritage values (MDBC 2008). A further assumption is the lake operations can be optimized to promote recruitment and growth of native vegetation. The Lake Victoria Operating Strategy (MDBC 2002), aims to balance the operation of the lake as a water storage with conditions conducive to native vegetation.

An adaptive management experiment with lake operations conducted during the 2010-11 water year was designed to test predictions concerning conditions for Spiny Sedge (Cyperus gymnocaulos) growth and recruitment (Bowman 2011). Spiny Sedge is a native perennial, and the dominant species of shoreline vegetation at Lake Victoria. Its dense rhizomatous root structure is also thought to be ideal for helping stabilize sediments. The experiment with lake operations involved holding the lake at full supply level for the minimum time possible and commencing drawdown as early as possible. Following this experimental lake operation, large scale recruitment of Spiny Sedge was observed in May 2011 (Bowman 2011).

Conditions 29 – 34 of the Section 90 Consent require the establishment of ongoing monitoring programs to investigate the mechanisms driving erosion and deposition of sediments on the lake shore, and those driving change in native vegetation (MDBC 2008). Extensive data sets now exist; shoreline monitoring data have been collected since 1995, and vegetation data since 1998. The consultants collecting these data provide annual reports to SA Water and the Murray–Darling Basin Authority (e.g. Sluiter 2011; Stephenson & Kennedy 2011), but these reports are generally ‘snap shots’ of the year’s monitoring data. Carlile (2010) drew together much of the monitoring data and qualitatively described spatial and temporal patterns. He concluded that lake operations influenced vegetation patterns, with higher vegetation covers observed in years after the lake had been filled for smaller numbers of days. However, this conclusion was challenged on the basis that it was drawn from too little evidence, and from qualitative data summaries only (J. Roberts, Lake Victoria Scientific Reference Panel, unpublished comments). To date, there has been no comprehensive attempt to synthesize the shoreline and vegetation monitoring data.

This project aimed to fill this gap by conducting statistical analyses on the Lake Victoria monitoring data. Using Bayesian modelling techniques (see below), we set out to test for the presence of spatial and temporal trends in vegetation and in shoreline erosion and deposition. We also tested for associations between hydrologic regime and five key plant species, and between hydrologic regime and erosion and deposition. Finally, we tested the key assumption of the Section 90 Consent, whether there was an association between shoreline vegetation and rates of erosion and deposition, and whether the presence of vegetation reduced erosion.

Bayesian Statistical Modelling

We chose Bayesian statistical modelling as the analytical approach for this project because of its inherently flexibility (Clark 2005). Models can be formulated to conform to the requirements of the data, whereas standard statistical approaches must force the data to comply with the requirements of a relatively small number of model types (McCarthy 2007). Bayesian models are finding increasing use in ecology (e.g. Ver Hoef & Frost 2003; Martin et al. 2005), including for questions of ecological responses to water regimes (Webb et al. 2010b). They allow the construction of far more complex models than possible with traditional statistical approaches,

2

and are particularly suited to dealing with the complexities of spatiotemporal aspects in ecology (Clark 2005). A particular example of the type of complexity considered for this project is the presence of semi-quantifiable sampling uncertainty for the vegetation sampling, caused by using the Braun-Blanquet scale (Braun-Blanquet 1964) for estimating vegetation live cover. The Braun-Blanquet scale divides 0-100% cover of a species into 6 intervals of uneven width, with interval width being smaller for lower percentage covers. To analyze such data as percentage cover requires that models account for the fact that, for example, a Braun-Blanquet cover category of 2 indicates a percentage cover of anywhere between 5 and 25% (Table 1). Here, we were able to introduce an extra variable into the models that added random ‘noise’ around each data point to account for this uncertainty.

Bayesian modelling revolves around Bayes’ Theorem (Bayes 1763, reprinted in Barnard 1958). In simple terms, the theorem provides a mathematical expression for updating our estimate of some statistical parameter, say , given a set of observations, y. The theorem states that:

In words, the ‘posterior’ probability distribution (left hand term) of a parameter () given (signified by “|”) the data at hand (y) is proportional to the product of the ‘likelihood function’ (the probability of y given ), and the prior probability distribution for (our estimate of likely values of before collecting any data) (Gelman et al. 2004). In an actual analysis, in the representation above is replaced by the full set of model parameters and their relationships to one another. The presence of the prior distribution is the source of much controversy concerning Bayesian modelling, as it can be seen as making the analyses overly subjective (McCarthy 2007). However, when little information exists concerning a parameter, one is able to assign a so-called minimally-informative or ‘vague’ prior distribution. Such a prior has only slight effects on the posterior distribution, and indeed a familiar analysis (e.g. ANOVA or regression) carried out using Bayesian methods and minimally-informative priors for the parameters (e.g. regression slope) will usually come up with a similar distribution for that parameter as the almost universally used ‘frequentist’ model (see McCarthy 2007 for an explanation of frequentism). In this project, we used minimally-informative prior distributions.

Bayesian modelling requires the integration of complex probability density functions. Until the mid-1990s, models had to be solved analytically (i.e. using calculus). This meant that only relatively simple models could be used. The development of Markov Chain Monte Carlo estimation techniques (MCMC; Andrieu et al. 2003) meant that much more complex probability density functions, including those for which no analytically-solvable integral exists, could be used. This development, more than anything else, fuelled the recent growth in Bayesian statistical modelling of complex problems. In this project, we use the OpenBUGS (Lunn et al. 2009) software package, which uses MCMC estimation to parameterize Bayesian models.

3

Methods

Data

For both vegetation and shoreline data, we only considered data collected at < 30 m AHD elevation. This was the vast majority (~95%) of data collected and is where beaches occur, and ensured that analyses were not skewed by the small number of data points collected at elevations up to 40 m AHD. Moreover, 30 m AHD coincides with the boundary of the Lake Victoria Cultural Landscape Plan of Management (MDBC 2008). It should be noted that this includes areas inundated on a yearly basis (up to 27 m AHD) as well as areas beyond this that are seldom or never inundated (although the majority of data were collected from < 27 m AHD). The inclusion of the higher elevations in an analysis designed to investigate effects of hydrology was a deliberate attempt to measure variation known to not be caused by changes in hydrology. This meant that larger changes observed in inundated areas could be more confidently ascribed to hydrology.

Vegetation

Vegetation data analyzed were collected by Dr Ian Sluiter (Ogyris Ecological Research Pty. Ltd.) over the period 1998-2011. For detailed methods see Sluiter and Robertson (1999), but briefly, vegetation data were collected as live foliar projective cover scores (sensu Walker & Tunstall 1981) using the Braun-Blanquet scale (Braun-Blanquet 1964) as modified for use in the Victorian Mallee (Cheal & Parkes 1989) (Table 1). Plants were identified to species or to the maximum taxonomic resolution possible in the field. Data were also available as biomass estimates for individual species, but because the biomass data are derived directly from the estimates of live cover (by multiplying cover with previously-derived density estimates; I. Sluiter, pers. comm.), we decided to analyze the cover data.

Table 1. Modified Braun-Blanquet cover classes used for quantifying vegetation cover, along with the percentage cover mid-points and lower and upper bounds used when converting these scores to percentage covers in the statistical analysis.

Category Definition Mid-point (%) Lower bound (%) Upper bound (%)+ Cover less than 1% 0.5 0 11 Cover 1- 5% 3 1 52 Cover 5-25% 15 5 253 Cover 25-50% 37.5 25 504 Cover 50-75% 62.5 50 755 Cover greater than 75% 87.5 75 100

Data were collected from permanently-marked 4 x 1 m quadrats distributed along each of 12 transects around Lake Victoria, perpendicular to the shoreline (Table 2). The vertical elevation of each vegetation quadrat was surveyed once, in June 2010 for all quadrats with the exception of

4

points TG111 to TG117 (Talgarry Barrier), which were surveyed in July 2011 because of excessive water depth at the time of the first survey (I. Sluiter, pers. comm.). Data from the transect at Frenchman’s Creek (KFM) were excluded from analysis because the transect was perpendicular to the creek, rather than to the lake. Additionally, transects at East Nanya (KEN) and West Moon (KWM) were excluded from analyses because they were parallel to the lake rather than perpendicular, and were specifically established to monitor vegetation changes following sand nourishment (Sluiter 2011).

Shoreline

Shoreline data analyzed were collected by Dr Wayne Stephenson (University of Otago, formerly University of Melbourne) and Dr David Kennedy (University of Melbourne) over the period 1995-2011. Surveys used a TOPCON 220 total station with data either manually recorded in the field (2001-2010) or downloaded and analyzed using CivilCad surveying software (Version 7). For detailed methods see Stephenson and Kennedy (2011), but briefly, data were collected from 34 transects around Lake Victoria, perpendicular to the shoreline (Table 2). With the exception of data from 1995-1998, and with some variability between 1999-2000, data were collected using standard surveying techniques (Kennedy 2012), with a survey elevation being recorded at each change in gradient along the transect and at regular intervals to maximize data density. The earlier data were collected using fixed intervals between survey points. However, such an approach is likely to miss some detail in the shoreline profile (Kennedy 2012).

For analysis, we converted shoreline survey data to a ‘vertical difference’ between surveys (Figure 1). Specifically, each vertical difference was calculated as the positive (deposition) or negative (erosion) change (m) between subsequent surveys of the same transect for chainage points separated by 10 m.

24

25

26

27

0 10 20 30 40 50 60 70 80 90Chainage (m)

2nd survey

deposition

Elev

ation

(m A

HD)

erosion

1st survey

Figure 1: Hypothetical calculation of vertical difference for shoreline profile analyses. Erosion and deposition are recorded as positive and negative vertical changes, respectively, at evenly-spaced chainage points.

We used geometric and trigonometric rules to calculate the elevation of each chainage point along the transect for each survey, and quantified vertical difference as the difference between the two calculated elevations. The vertical difference metric was designed to investigate the dynamic behavior of the shoreline over time and space, and under different hydrologic regimes.

5

It should be interpreted as a rate of erosion or deposition (e.g. m yr-1 for the trend analysis). We chose to use this method, rather than the method employed by Stephenson and Kennedy (2011) of Excursion Distance Analysis (Winton et al. 1981). We felt that a numerical rate of accretion or erosion is a more intuitively simple concept than having to infer accretion or erosion through the horizontal movement of contours, which is how such information would need to be conveyed by EDA. However, we note that the two techniques are essentially measuring the same thing.

Lake level

Lake water level were sourced from the MDBA web-site (riverdata.mdba.gov.au/sitereports/a4261093/mdba_a4261093_site_report.html), with daily water levels collected at monitoring station A4261093 located near the centre of Lake Victoria.

Statistical models

General

We divided Lake Victoria into three ‘sectors’, on the expectation that vegetation and shoreline processes may differ between different areas of the lake (Figure 2, Table 2).

West

South

East

Figure 2. The three sectors into which the Lake Victoria shoreline was divided for analysis. Imagery ©2012 Cnes/Spot Digital Globe.

6

The eastern sector is characterized by simple, sloped beach faces, with little structural complexity. The southern sector consists of the lake outlet areas including Frenchman’s Creek and Rufus River, and associated delta. The beach faces are shallower and more complex, with formations like Snake Island and Talgarry Barrier providing structural complexity. The southern sector also supports considerably more vegetation than the other sectors, and is the focus of cultural heritage management efforts. The western sector consists of areas flooded by the regulation of Lake Victoria in 1928. The beaches are shallower, and contain large amounts of coarse woody debris, a relic from previous vegetation killed when the lake was flooded.

Table 2. Lake ‘sectors’ used to stratify data in the analyses.

Zone Management units Vegetation transect codes Shoreline transect codes

East Nulla BeachTalgarry Beach

KNU, KTN, KTW tgw, ps28, ktn, ps25, knu, ps21

South Talgarry BarrierSoutheastern BeachSnake IslandSouthern Lake & Frenchman’s Island

KT1, KT2, KSN, KEM, KGK, KWN

wny, midn, eny1, eny2, gck1, gck2, mon1, mon2, mon3, mon4, mon5, mon6, emn1, emn2, ksn, snk1, snk2, tgb1, tgb2, tgb3, tgb4

West Noola BeachSouthwestern Beach

KNL, KVB, KVW ps17, knl, ps13, ps9, kvw, ps5, ps2

The southern sector had around twice as many transects as the other sectors for both the vegetation and shoreline analyses. This was a function of the data available, and is the ultimate product of the management focus on preserving the cultural heritage value of the southern area of lake shore.

Statistical models used various types of continuous relationships (i.e. regressions) to quantify the effect of independent variables on either vegetation or shoreline processes. Continuous variables considered as either main drivers or covariates consisted of the effects of: elevation of the lake shore (vegetation & shoreline), year of sampling (vegetation), date of sampling (shoreline), submergence (vegetation), optimal conditions (vegetation), exposure (vegetation), and number of days in the wave zone (shoreline). Independent variables were standardized to a mean of zero and standard deviation of 1 prior to analysis, as this improves model convergence and stability (see below). Models also employed various categorical random factors to explain additional variation in the data. Random factors considered consisted of: transect (vegetation), quadrat within transect (vegetation), chainage point within transect (shoreline), and sampling year (when temporal trends were not a primary focus of the analysis; shoreline). Within each sector, the data for all transects were pooled; this smoothes the relationships with the independent variables. Quadrat-level data within transects are likely to be extremely uncertain because their size of 4 m2 is small compared to the scale over which clumps of vegetation vary (J. Roberts, pers. comm.). The categorical random factors are able to quantify differences among

7

groups of data points (e.g. transects) while still assuming the same overall relationship of sector-level data with the independent variables.

Specific analyses conducted are summarized below. Overall, the models tested for spatial and temporal trends in the vegetation and shoreline monitoring data, the effect of hydrologic regime on these parameters, and the interaction between vegetation cover and shoreline erosion or deposition. Within the vegetation trend analysis, dividing the data by life-span (annual or perennial) and growth form (grass, herb, sedge/rush or woody) was designed to assess whether any particular types of vegetation were driving vegetation trends, and hence whether management needed to target such a plant type.

Spatial and temporal trend analysis of:o Vegetation abundance quantified as:

Total live vegetation cover Live cover of annuals and perennials Live cover of different growth forms

o Shoreline erosion and deposition Effects of hydrologic regime on:

o Occurrence and live cover of target specieso Shoreline erosion and deposition

Effects of vegetation abundance (total live cover) on shoreline erosion and deposition

For data processing and statistical analysis, we used a combination of the R statistical language (R Development Core Team 2010) and OpenBUGS Bayesian analysis software (Lunn et al. 2009). We used predictive sampling to generate values for graphing, and hypothesis tests to assess the importance of model parameters. Bayesian hypothesis tests differ from the more widely-used frequentist null hypothesis test. For the tests used in this report, a probability value near 1 implies support for the parameter (e.g. a regression slope) being positive, a value near 0 provides support for the parameter being negative, and a value near 0.5 provides no evidence in either direction. We adopted a ‘significance’ level of p < 0.05 or p > 0.95 to identify important variables. The code for all models is included in Appendix A. All models used three separate Markov chains, and we monitored convergence of the chains to the posterior distribution using the Brooks-Gelman Rubin diagnostic (Brooks & Gelman 1998). This led to varying lengths of ‘burn-in’ (non-monitored iterations that ensure that the Markov chains have converged to the range of the posterior distribution before starting sampling). Chains were also monitored for varying numbers of iterations depending on how quickly stable and smooth posterior distributions were produced (Table 3).

Vegetation trends

We assessed temporal and spatial trends in total vegetation live cover, cover of annual species, and cover of perennial species. We attempted to analyze cover of growth form groups, but the data contained too many zeroes to use the model developed for total live cover.

8

Table 3. Implementation details for the different statistical models. All models had three independent Markov chains.

Model Burn-in Iterations sampled per chain

Total sample size for parameter estimation

Vegetation trends 100,000 10,000 30,000Shoreline trends 100,000 10,000 30,000Target species – p/a 5,000 10,000 30,000Target species – cover 5,000 5,000 15,000Shoreline inundation 5,000 10,000 30,000Shoreline vegetation 15,000 15,000 45,000

Live cover data were quantified as the sum of the mid-points of the cover classes (Table 1) of all species belonging to the vegetation type being analyzed (total cover, annuals, perennials). This frequently resulted in ‘cover’ figures of > 100% for individual quadrats, but does not affect the validity of the resulting analysis. These data were positively skewed, and so were square-root transformed for analysis. We tested for the presence of linear trends in vegetation cover over time (year of sampling) and space (elevation of the quadrat), with separate analyses for each sector of the lake shore. Effect of transect and of individual quadrats were included as categorical random factors with zero mean, and variance calculated across all data for each sector. We incorporated the effect of the uncertainty of Braun-Blanquet mid-point data by including a uniformly distributed random variable that sampled from the interval bounded by the summed lower and upper bounds of the Braun-Blanquet classes (Table 1). This variable was not quantified by the analysis, as there were no data to do so; it simply added ‘noise’ to the analysis to acknowledge the uncertainty inherent in Braun-Blanquet cover classes. The trend model was thus specified as:

where the data point mp is the sum of mid-points of the Braun-Blanquet cover classes for species within that sample, and is collected from quadrat k within transect j within sector i, and in year l and at elevation m. Transformed mp is modeled as a normally distributed variable, conditional upon regression parameters and random factors. The main regression parameters (int, eff.year, eff.elev) were assigned separate minimally-informative normal prior distributions N(0,102), and overall model uncertainty was assigned a minimally-informative uniform prior distribution U(0,10). The categorical random factors (eff.trans, eff.quad) were normally distributed random variables with a fixed mean of zero and variance drawn from a uniformly distributed hyperparameter for each sector. Thus, for the effect of transects:

9

with an identical construction for the effect of quadrats within transects. Lastly, the ‘noise’ effect of Braun-Blanquet cover uncertainty was calculated as:

where lb and ub are the lower and upper bounds for the cover class of the data point (with the sub-scripts omitted to aid readability).

Shoreline trends

We assessed spatial and temporal trends in shoreline vertical difference using a similar model structure as employed for the vegetation trend analysis. The vertical difference data (v.diff) were normally distributed, and so did not require transformation prior to analysis. We were unable to include the random categorical effect eff.trans because of ‘non-identifiability’ problems. Non-identifiability occurs when two parameters in the model are essentially describing the same variation within the data. The Markov chains for the two variables become very highly autocorrelated (i.e. it is possible to predict the value of one directly from the other), and do not converge to stable posterior distributions. In this case, within each sector the int parameter was autocorrelated with the eff.trans parameters for transects within that sector. Attempting to run the model without eff.trans, but with a random effect equivalent to the eff.quad variable from the vegetation trend model (but describing variation among the evenly spaced chainage points for which v.diff was calculated) resulted in the same problems. Thus the final model was simply:

where v.diff is the rate of vertical change per year, modeled as normally distributed variable contingent upon the regression parameters. The parameters are as described for the vegetation trend model above, with the exception that date is the actual date of sampling rather than simply year (as there were two samples per year for most chainage points from 2009 onwards Prior distributions for regression parameters and uncertainty were as described for the vegetation trend model.

Effects of hydrologic regime on target species

We tested for the effects of hydrologic regime on the plant live cover of five target species: Spiny Sedge (Cyperus gymnocaulos), Common Reed (Phragmites australis), Common Couch (Cynodon dactylon), Rat’s-tail Couch (Sporobolus mitchellii), and Spiny Mudgrass (Pseudoraphis spinescens) – hereafter referred to by their common names. These species were chosen by the Lake Victoria Scientific Reference Panel, with the exception of Spiny Mudgrass, which was suggested by Dr Ian Sluiter. The species chosen by the SRP were chosen because they were perennial, hardy, and

10

have clonal growth form or are mat-forming. We developed a common conceptual model describing the effects of exposure, submergence, and optimal conditions on the growth of each species (Figure 3, Table 4). The model does not consider a specific growing season, as growth at Lake Victoria is unlikely to be limited by temperature (J. Roberts, pers. comm.). Recruitment of new individuals (as opposed to clonal growth) was not specifically considered using this model as sexual reproduction tends to occur at particular times of year.

Wat

er le

vel r

elati

ve to

qua

drat

Shore level

Critical depth and duration of submergence

Lake surface relative to ground level

Optimal conditions water level range Critical depth and

duration of exposure

Time

Figure 3. Conceptual model relating exposure, submergence and optimal conditions for growth of the five target plant species. Different depths and durations were chosen for each of the five target species (Table 4).

Exposure and submergence are both defined by critical water depths; if water depth is below (exposure) or above (submergence) the critical depth for sustained periods, we expect negative impacts. These variables are quantified in terms of multiples of critical periods (if either occurs for more than 1.0 critical durations, it is expected to have negative effects).

Table 4. Parameter values for the common conceptual model for each of the five target species.

Species Critical depth submergence

Critical duration of submergence

Optimal conditions depth

Critical depth of exposure

Critical duration of exposure

Spiny Sedge 30 cm 11 months ± 10 cm -20 cm 11 monthsCommon Reed 1 m 11 months +50 cm

-20 cm-50 cm 11 months

Common Couch

10 cm 8 months 0 cm-20 cm

-30 cm 10 months

Rat’s tail Couch 15cm 6 months 0 cm-20 cm

-20 cm 10 months

Spiny Mudgrass 1 m 7 months +50 cm-10 cm

-10 cm 7 months

11

Exposure and submergence can occur over more than one year, so values could be well above 1.0. Optimal conditions are quantified as the number of days that water levels were within the defined optimal range for growth during the period between successive vegetation samples. The model used these same three parameters for each species, and they were calculated in the same way, but they had different values. Parameter values (Table 4) for the different species were drawn from a review of the literature (See Appendix B for sources consulted). These values were based on reports of the tolerances to flooding and drying of the target species. In general, we adopted conservative values rather than some of the more extreme values reported in the literature.

The live cover data for the five target species were dominated by 0% cover values. Moreover, the second most common cover category was then the 0-1% category, and so on. Thus the data were not normally distributed, and could not be made to approximate normality by any transformation. We used a log-Poisson regression (Kéry 2010) to model change in cover with hydrologic regime. This type of model is normally used for count data, but was equally applicable to the percentage cover data. However, the predominance of zeroes in the data was such that they could not be adequately modeled even with a Poisson distribution. To accommodate this, we performed a two-stage analysis. First, we used a logistic regression to assess the effects of submergence, exposure, and optimal conditions on the presence or absence of the five target species. Second, we performed the log-Poisson analysis for quadrats where cover was > 0%. As was the case with the shoreline trend analysis, we were unable to include categorical random effects in the target species cover analyses because of problems with non-identifiability of parameters, similar to those encountered for the shoreline trend analysis. The final models thus simply included the main effects, with the cover model also including the effect of Braun-Blanquet mid-point imprecision as described for the vegetation trend analysis above. The models were:

where pa is presence or absence of the target species, modeled as a Bernoulli-distributed random variable with probability p. For the regression, p is logit-transformed to linear parameter space, allowing linear regression parameters to be fitted. For the live cover analysis, cover is modeled as a Poisson-distributed random variable with mean lambda. Lambda is log-transformed to allow linear regression parameters. All of the hydrologic regime variables were calculated at the level of the individual quadrat measurements, but the sub-scripts have

12

been omitted from the equations above to improve readability. As with the previous models, regression parameters for each sector were assigned minimally-informative normal prior distributions.

Effects of hydrologic regime on shoreline erosion and deposition

We tested for the effect of hydrologic regime on erosion and deposition on the shoreline. We hypothesized that maximal erosion (and potentially deposition) is likely when that portion of the shoreline is within the ‘wave zone’ – from the lake surface up to the crest height of waves. Moreover, the bulk of erosion / deposition is likely to be caused by the largest waves.

We calculated wave heights based on the wind speed calculations from Mildura as reported in Stephenson (2012), which calculates trigger velocities for wind events for aeolian sediment transport on the shore as a combination of wind speed and duration. Wave calculations were based on equations for the US Army Corps of Engineers Shore Protection Manual using the Young and Verhagen method (US Army Corps of Engineers 1984). Calculations were performed using the Coastal and River Engineering Support System, Netherlands (www.cress.nl).

Based on the bathymetry of Hudson and Bowler (Hudson & Bowler 1997), we assumed an average water depth of the lake of 4.6 m at full supply level (27m AHD) and 1.2 m for low lake levels (23.6m AHD). We used the minimum trigger value for high-energy wind events reported in Stephenson (2012) – maximum wind speed of 50 km h-1 (13.9 m s-1) for 2 h duration. The most common winds are 10-20 km h-1, so we also calculated waves under a 24 h event of 20 h-1 (5.56 m s-1). The shallowness of Lake Victoria causes significant dissipation of wave energy and inhibits the growth of large waves. At low water level the lake is depth limited, which means waves cannot attain their maximum height. The wave calculation method used cannot predict wave heights that are depth limited as this requires significant modeling as well as a much more detailed knowledge of the bathymetry than is currently available. Thus our calculated wave heights are likely to be higher than actual at low water levels.

Waves were modeled from the North, South, East and West, with a maximum fetch distance of 12.36 km (north – south at full lake levels) used. Low lake levels in an East – West direction) have the minimum fetch of 9.09 km. Waves caused by winds blowing for 2 h are duration limited, while waves generated at low water levels are depth limited. The maximum modeled wave height occurred for the 24 hr x 20 km h-1 wind event. This wind event was above the high energy wind event trigger of 12 h duration for this wind speed (Stephenson 2012). The maximum calculated wave height was 0.32 m with a period of 2.16 s. For average winds, wind velocity at Mildura is below 3.5 m s-1 55% of the time. Using a fetch distance of 12 km (North – South), and wind speed of 3 m s-1, a ‘typical’ wave height of 0.14 m (water depth limited to 3m) was also calculated. We calculated ‘wave zone days’ for both the 0.32 and 0.14 m wave heights as the number of days between shoreline surveys that a chainage point (for which v.diff was calculated) was within the wave zone (i.e. at lake surface level or within the wave height).

v.diff was modeled as a normally distributed random variable conditional upon the number of wave zone days during the period for which v. diff was calculated. We considered multiple

13

model structures with different additional random factors and covariates. As with the shoreline trend analysis, there were issues of non-identifiability between the intercept and categorical random factors. The final model, chosen on the basis of the width of prediction intervals and the level of significance of the hypothesis tests was:

As with the vegetation hydrologic regime model, wave zone days (inund) was calculated at the level of the individual sampling point, but the subscripts have been omitted from the equation for clarity. This model does not contain an intercept term. In assessing the fit of different models, we found that including the random effect of year of sampling explained more variation than including the intercept term for each sector. Effect of year was pooled across the three sectors, as it seemed reasonable to assume that year to year variation in hydrological effect would operate at the whole-lake level. We were unable to include both terms because of non-identifiability. Including elevation of the sampling point as a continuous covariate also improved model performance. Prior distributions for the regression parameters, categorical factor and overall model uncertainty were as described above.

Effect of vegetation live cover on erosion and deposition

The final model combined the two processes hitherto considered separately, and treated vegetation live cover as the hypothesized primary driver of shoreline erosion and deposition. This required that we identify ‘pairs’ of vegetation and shoreline survey transects. We used GIS layers supplied by the MDBA to identify shoreline transects that were sufficiently close to vegetation transects to make such pairs. There were 11 such matching pairs, but the pair from Frenchman’s creek was excluded from analysis for the reasons outlined above, resulting in 10 transect pairs for analysis (Table 5).

Table 5. Matching pairs of shoreline and vegetation transects for testing the effect of vegetation cover on shoreline erosion.

Sector Vegetation Transect Shoreline TransectEast Nulla (KNU) knu

Talgarry North (KTN) ktnSouth East Moon (KEM) emn2

Gecko (KGK) gck2Snake (KSN) ksnTalgarry 1 (KT1) tgb1Talgarry 2 (KT2) tgb2West Nanya (KWN) wny

West Lake Victoria West (KVW) KvwNoola (KNL) Knl

14

We used the surveyed elevations of the vegetation quadrats to identify chainage points of equivalent elevation on the adjacent shoreline transects (using the surveyed shoreline elevation that was closest in date to the date of the vegetation quadrat elevation survey). Such chainage points had to be interpolated between pairs of surveyed elevations on the shoreline transect; this was done using geometric rules. We then calculated vertical difference values for these chainage points across subsequent shoreline surveys using the methods described for shoreline trend analyses.

Ideally vegetation surveys would take place at a similar time to the shoreline surveys. However, this was rarely the case and vegetation and shoreline surveys had to be temporally matched. We matched v.diff figures calculated between a pair of shoreline surveys to the previous vegetation survey, with two caveats. First, if the previous vegetation survey was more than 12 months prior to the date at which v.diff was calculated (the second of the two shoreline surveys), we excluded both data points. Second, if the two shoreline surveys used to calculate v.diff were more than 15 months apart, we excluded the v.diff data point (and consequently did not match to any vegetation surveys occurring during that time). These cutoffs were adopted because we did not believe we could reasonably assume that vegetation live cover would be constant enough over such periods to relate vegetation cover to v.diff.

The statistical model treated vegetation cover (square-root transformed) as the primary driving variable. We were unable to include the effects of Braun-Blanquet mid-point imprecision as described above, but were able to scale the model residuals to include this effect. We considered various other covariates and categorical random factors, with the final model structure being:

where v.diff is the change in vertical height as described above, and is modeled as normally-distributed variable conditional upon the vegetation cover in the matched quadrat (mp), with the number of days between the two shoreline surveys (daysl2-l1) being a covariate and year of survey being a categorical random factor. As with the model for effect of hydrological regime on erosion and deposition, the random effect of year of sampling was pooled across the whole lake. To incorporate the Braun-Blanquet uncertainty in the driving mp variable, a scaled residual was calculated for each data point (the other models estimate a single distribution of residuals for each sector – by multiplying a sector-scale estimate of standard deviation (s.sdi)by the Braun-Blanquet range of plausible values for the mp value related to that datum [range(mp)].

15

Results

Vegetation trends

Total live cover

There were increases in vegetation live cover over time and with increasing elevation for both eastern and western sectors (Figure 4a,b,d,e). Median predicted cover was extremely low for early years and lower elevations for both these sectors. The eastern sector predictions were characterized by very high uncertainty (Figure 4a,b), with the lower prediction limit frequently being below 0% cover (not plotted). For the southern sector, there was no evidence of any change in total vegetation live cover over time (Figure 4d, Figure 5b). There was a slight decrease with elevation (Figure 4c), but this effect was marginally non-significant according to the criteria specified above (Figure 5a, Table 6).

0

20

40

60

80

100 (a) (b)

0

20

40

60

80

100 (c) (d)

24 25 26 27 28 29 300

20

40

60

80

100 (e)

1998 2001 2004 2007 2010

(f)

Mean

Cove

r (%)

Elevation Year

16

Figure 4. Vegetation trend analysis results for total cover. Left column shows trends with increasing elevation on the lake shore, and right column shows trends over time. Rows are eastern, southern and western sectors, respectively. Solid line is the median predicted cover, holding all other variables constant, and dotted lines are the 2.5% and 97.5% credible limits for predicted cover.

East South West-1.0

-0.5

0.0

0.5

1.0

1.5

(a)

East South West

0.0

0.5

1.0

1.5

(b)Eff

ect of

Eleva

tion

Effect

of Ye

arFigure 5. Distributions of the regression parameters. Solid dots are median estimate, with error bars extending to the 2.5% and 97.5% limits of the credible interval. Horizontal dotted line is that of no effect.

The random effects accounted for substantial variation within the data. The pooled standard deviations attributable to differences among transects and differences among quadrats within individual transects were as large (or larger) than the overall model residual standard deviation (Table 6). The random effects explained the most variation in the eastern sector.

Table 6. Results of hypothesis tests and magnitude of random effects for the vegetation trend model for total cover. Hypothesis test results are for the regression parameters. Probabilities shown are p (x > 0) – i.e. a positive effect. Magnitude of random effects is shown as the summary statistics for the distribution of standard deviation (mean ± 1.0 SD) attributable to transect and quadrat, and the overall model residual.

Sector → East South WestEffect of elevation 0.984 0.061 0.995Effect of time >0.999 0.821 >0.999standard deviation among transects 2.31±2.18 1.75±0.92 1.45±1.71standard deviation among quadrats 2.16±0.49 1.61±0.19 1.59±0.37residual standard deviation 1.19±0.09 1.49±0.06 1.59±0.11

Annuals

Predicted live cover of annual species was low, with the highest covers predicted in the western sector near the top of the elevation range (Figure 6e). Despite these low covers, and the fact that predicted covers often fell below 0%, there were significant changes over time for all sectors, with increases in the east and west, and decreases in south (Figure 6b,d,f, Figure 7b, Table 7). There were also significant decreases with elevation in the south, and increases with elevation in the west, but no evidence of any change with elevation in the east. (Figure 6a,c,e, Figure 7a, Table 7). As with the data for total cover, the random effects of transect and quadrat explained a considerable amount of variation in the data. Overall, standard deviations

17

attributable to transect and quadrat were of the same order of size as overall model residual variation (Table 7)

0

10

20

30

40 (a) (b)

0

10

20

30

40 (c) (d)

24 25 26 27 28 29 300

10

20

30

40 (e)

1998 2001 2004 2007 2010

(f)

Mean

Cove

r (%)

Elevation YearFigure 6. Vegetation trend analysis results for cover of annual species. See Figure 4 for explanation of plot arrangement and interpretation.

East South West

-0.5

0.0

0.5

1.0 (a)

East South West

-0.2

0.0

0.2

0.4

(b)

Effect

of El

evatio

n

Effect

of Ye

ar

Figure 7. Distributions of the regression parameters for the trend analysis of cover of annual species. See Figure 5 for explanation of plot arrangement and interpretation.

18

Table 7. Results of hypothesis tests and magnitude of random effects for the vegetation trend model for annual species. Hypothesis test results are for the regression parameters. Probabilities shown are p (x > 0) – i.e. a positive effect. Magnitude of random effects is shown as the summary statistics for the distribution of standard deviation (mean ± 1.0 SD) attributable to transect and quadrat, and the overall model residual.

Sector → East South WestEffect of elevation 0.911 0.002 0.999Effect of time >0.999 0.003 0.985standard deviation among transects 1.53±1.68 0.79±0.45 1.15±1.47standard deviation among quadrats 1.32±0.31 0.75±0.10 0.64±0.21residual standard deviation 1.00±0.06 1.16±0.04 1.33±0.09

Perennials

Perennial species were in higher abundances than annuals. There were significant increases in live cover of perennials with elevation for the eastern and western sectors, but no evidence of any change with elevation for the southern sector (Figure 8a,c,e, Figure 9a, Table 8).

0

10

20

30

40

50

60 (a) (b)

0

10

20

30

40

50

60 (c) (d)

24 25 26 27 28 29 300

10

20

30

40

50

60 (e)

1998 2001 2004 2007 2010

(f)

Mean

Cove

r (%)

Elevation Year

19

Figure 8. Vegetation trend analysis results for cover of perennial species. See Figure 4 for explanation of plot arrangement and interpretation.All sectors showed strong evidence of increase in cover of perennial species over time (Figure 8b,d,e, Figure 9b, Table 8).The random effects of transect and quadrat explained a similar amount of model variation as was the case for analyses of total cover and of annual species (Table 8).

East South West

-0.5

0.0

0.5

1.0

1.5 (a)

East South West

0.5

1.0

1.5

(b)Eff

ect of

Eleva

tion

Effect

of Ye

arFigure 9. Distributions of the regression parameters for the trend analysis of cover of perennial species. See Figure 5 for explanation of plot arrangement and interpretation.

Table 8. Results of hypothesis tests and magnitude of random effects for the vegetation trend model for perennial species. Hypothesis test results are for the regression parameters. Probabilities shown are p (x > 0) – i.e. a positive effect. Magnitude of random effects is shown as the summary statistics for the distribution of standard deviation (mean ± 1.0 SD) attributable to transect and quadrat, and the overall model residual.

Sector → East South WestEffect of elevation 0.964 0.225 0.974Effect of time >0.999 >0.999 >0.999standard deviation among transects 1.79±1.86 1.94±1.02 1.30±1.60standard deviation among quadrats 1.57±0.37 1.64±0.20 1.61±0.36residual standard deviation 1.32±1.08 1.59±0.06 1.45±0.10

Growth forms

As mentioned in the Methods, the high number of 0% cover values for individual growth forms meant that the linear model developed for total cover could not adequately represent the growth form data (and in fact would not run at all). We plotted the raw data against elevation and sampling year to provide some form of comparison to Figure 4,Figure 6, andFigure 8 (Appendix C). These plots show that with the exception of woody species, the live covers of the different vegetation growth forms are greatest in the southern sector. All growth forms showed some increase in cover over time in the eastern and western sectors (with the exception of grasses in the west). There was some evidence of an increase in cover of rushes and sedges in the southern sector over time, but the cover of other growth forms remained relatively constant or was variable. Changes in cover of the different growth forms with elevation were mixed. In the south, grasses and rushes and sedges showed an apparent increase in cover to around 26 m AHD, but then rapidly dropped beyond this. Also in the south, herbs appeared to reduce in cover with elevation. The only other apparent pattern was for woody vegetation in the west, which was virtually absent below 27 m AHD, but for which high values were recorded above this.

20

Vegetation at Lake Victoria was numerically dominated by herbs; this growth form had between approximately 5-10 times as many species as the other growth forms (Table 9). The five target species used in the analysis of hydrological effects were all among the most numerically dominant species for their respective growth forms. Within the different life history strategies, there were nearly twice as many annual species as perennial, but no species was overwhelmingly numerically dominant within the annual species (Table 9).

Table 9. Summary of different growth forms and life history strategies. Table shows the number of species within each group and the dominant species in each group, along with the frequency of occurrence of those species within that group.

Growth Form No. of Species Dominant Species

Grass 22 Spiny Mud-grass (Pseudoraphis spinescens) 33.5%Common Couch (Cynodon dactylon) 13.3%Rat’s-tail Couch (Sporobolus mitchellii) 12.8%

Herb 143 Common Sneezeweed (Centipeda cunninghamii) 5.4%Blue Rod (Stemodia florulenta) 5.3%

Rush/Sedge 10 Spiny Flat-sedge (Cyperus gymnocaulos) 71.2%Common Reed (Phragmites australis) 12.9%

Woody 18 Ruby Saltbush (Enchylaena tomentosa) 32.7%Sago Saltbush (Maireana pyramidata) 23.0%

Life HistoryAnnual 123 Spreading Sneezeweed (Centipeda minima) 5.3%

Lesser Joyweed (Alternanthera denticulata) 5.2%Little Medic (Medicago minima) 5.2%

Perennial 71 Spiny Flat-sedge (Cyperus gymnocaulos) 17.8%Spiny Mud-grass (Pseudoraphis spinescens) 14.5%

Shoreline trend analysis

Changes in the rate of shoreline erosion/deposition varied among the three sectors. There was no evidence of any change with elevation for the southern and western sectors (Figure 10a, Table 10). Moreover, these results show no evidence for net erosion or deposition in these sectors (i.e. the 0 m yr-1 rate is near the middle of the prediction intervals; Figure 11c,e). In contrast, there was strong evidence of an increase in rate of deposition with elevation in the eastern sector (Figure 10a, Table 10). Figure 11a shows that low-shore elevations in the eastern sector experience net erosion, with the rate being greater at lower elevations, but that higher elevations (> ~26 m AHD) experience net deposition, with the rate increasing with elevation.

There was no evidence to suggest any change in the rate of erosion/deposition over time for the eastern and western sectors, but southern sector has experienced a decrease over time (Figure 10b, Table 10). Figure 11d shows that the southern sector has changed from experiencing net deposition in the early sampling years to net erosion by the end of the sampling period. In other

21

words, over sampling period, this sector first gained sediment, but has now lost that sediment and is continuing to lose more.

East South West-0.010

-0.005

0.000

0.005

0.010

0.015

(a)

East South West-0.010

-0.005

0.000

0.005

0.010

0.015

(b)

Effect

of El

evatio

n

Effect

of Da

te

Figure 10. Distributions of the regression parameters for the shoreline trend analysis. Solid dots are median estimate, with error bars extending to the 2.5% and 97.5% limits of the credible interval. Horizontal dotted line is that of no effect.

-0.10

-0.05

0.00

0.05

0.10 (a) (b)

-0.10

-0.05

0.00

0.05

0.10 (c) (d)

23 25 27 29-0.10

-0.05

0.00

0.05

0.10 (e)

1998 2001 2004 2007 2010

(f)

Vertic

al Dif

feren

ce (m

)

Elevation Year

Figure 11. Prediction plots for the shoreline trend analysis. Left column shows changes with elevation; right column shows changes over time. Rows are results for eastern, southern, and western sectors, respectively.

22

Table 10. Results of hypothesis tests for the regression parameters in the shoreline trend analysis. Probabilities shown are p (x > 0) – i.e. a positive effect.

Sector → East South WestEffect of elevation 0.988 0.561 0.208Effect of time 0.570 0.011 0.652

Effect of hydrologic regime on target species

Spiny Sedge

There was no strong evidence of a beneficial effect of ‘optimal conditions’ on occurrence of Spiny Sedge (Figure 12a,d,g).

0.0

0.2

0.4

0.6

0.8

1.0 (a)

0.0

0.2

0.4

0.6

0.8

1.0 (d)

0 15 30 45 60 75 900.0

0.2

0.4

0.6

0.8

1.0 (g)

(b)

(e)

0.0 0.2 0.4 0.6 0.8 1.0

(h)

(c)

(f)

0 3 6 9 12

(i)

Proba

bility

of Oc

curre

nce

Optimal Conditions Submergence ExposureFigure 12. Effects of hydrologic regime on occurrence of Spiny Sedge. The left column shows effects of optimal conditions, the middle column shows effects of submergence, and the right column shows effects of exposure. The three rows are for eastern, southern, and western sectors, respectively. Solid line is the median estimate of the regression parameter, with the dotted lines being the 2.5% and 97.5% limits of the credible interval.

23

The effect for the eastern sector was positive, but marginally non-significant, and there was no effect for the southern or western sectors (Figure 13a, Table 11). There were on-average negative associations between Spiny Sedge occurrence and submergence (Figure 12b,e,h), but only the effect for the western sector was significant (Figure 13b, Table 11). Similarly, there were on-average negative effects of exposure on occurrence (Figure 12c,f,i), but the effect for the eastern quadrat was not significant (Figure 13c, Table 11).

East South West-0.5

0.0

0.5

1.0

(a)

East South West-1.0

-0.8

-0.6

-0.4

-0.2

0.0

0.2

(b)

East South West

-1.5

-1.0

-0.5

0.0

(c)

Effect

of Op

timal

Cond

itions

Effect

of Su

bmerg

ence

Effect

of Ex

posur

e

Figure 13. Distributions of the regression parameters for the analyses of hydrologic regime effects on occurrence of Spiny Sedge. Solid dots are median estimate, with error bars extending to the 2.5% and 97.5% limits of the credible interval. Horizontal dotted line is that of no effect.

Table 11. Results of hypothesis tests for the regression parameters in the analyses of hydrologic regime effects on occurrence of Spiny Sedge. Probabilities shown are p (x > 0) – i.e. a positive effect.

Sector → East South WestEffect of optimal conditions 0.949 0.634 0.293Effect of submergence 0.255 0.156 0.002Effect of exposure 0.226 <0.001 <0.001

The analysis of effects of hydrologic regime on live cover of Spiny Sedge produced extremely precise values of regression parameters, and hence very tight predictions of cover under different inundation regimes (Figure 15) and extreme probability values in the hypothesis tests (Table 12).

East South West-0.6

-0.5

-0.4

-0.3

-0.2

-0.1

0.0

0.1 (a)

East South West

-1.2

-1.0

-0.8

-0.6

-0.4

(b)

East South West-0.6

-0.4

-0.2

0.0

0.2 (c)

Effect

of Op

timal

Cond

itions

Effect

of Su

bmerg

ence

Effect

of Ex

posur

e

Figure 14. Distributions of the regression parameters for the analyses of hydrologic regime effects on cover of Spiny Sedge, when present. Interpretation of plots is as for Figure 13.

24

0

10

20

30

40 (a)

0

10

20

30

40 (d)

0 15 30 45 60 75 900

10

20

30

40 (g)

(b)

(e)

0.0 0.2 0.4 0.6 0.8 1.0

(h)

(c)

(f)

0 3 6 9 12

(i)

Perce

ntage

Cove

r

Optimal Conditions Submergence ExposureFigure 15. Effects of hydrologic regime on cover of Spiny Sedge, when present. Arrangement of panels and interpretation of content is as for Figure 12.

Of the regression parameters, only the effect of optimal conditions in the eastern sector was not significant. All other parameters showed significant negative effects of the hydrologic regime component on cover, with the exception of the effect of exposure in the southern sector (Table 12). Parameter distributions were mostly well away from the zero line of no effect (Figure 14).

Table 12. Results of hypothesis tests for the regression parameters in analyses of hydrologic regime effects on cover of Spiny Sedge, when present. Probabilities shown are p (x > 0) – i.e. a positive effect.

Sector → East South WestEffect of optimal conditions 0.249 <0.001 <0.001Effect of submergence <0.001 <0.001 <0.001Effect of exposure <0.001 >0.999 <0.001

25

Common Reed, Common Couch, Rat’s-tail Couch

There were insufficient occurrences of Common Reed, Common Couch and Rat’s-tail Couch in the eastern and western sectors to be able to make any inference concerning sensitivity of these species to hydrologic regime in these sectors. All three species occurred more frequently in the southern sector. For Common Reed, there was a significant negative effect of submergence, and a significant positive effect of exposure, but no effect of optimal conditions, on occurrence (Figure 16a,b,c, Figure 19a,c,e, Table 13). In contrast, live cover of Common Reed was significantly reduced by exposure and submergence, but increased by optimal conditions (Figure 16d,e,f, Figure 19b,d,f, Table 13)

0.0

0.1

0.2

0.3

0.4 (a) (b) (c)

0 60 120 180 240 3000

5

10

15

20 (d)

0.0 0.2 0.4 0.6 0.8

(e)

0 4 8 12

(f)

Proba

bility

of Oc

curren

ce

Optimal Conditions Submergence Exposure

Perce

ntage

Cove

r

Figure 16. Effects of hydrologic regime on occurrence (top row) and cover (bottom row) of Common Reed for the southern sector. Interpretation of content is as for Figure 12.

For Common Couch, there were strong negative impacts of submergence and exposure on both occurrence and live cover (Figure 17b,c,e,f, Figure 19a,b,c,d, Table 13). There was no evidence of any effect of optimal conditions on occurrence, but a strong positive effect of optimal conditions on cover (Figure 17a,d, Figure 19e,f, Table 13).

Finally, for Rat’s-tail Couch, there were negative effects of exposure and optimal conditions on occurrence, but no evidence of any effect of submergence (Figure 18a,b,c, Figure 19a,c, e, Table 13). Conversely, live cover of Rat’s-tail Couch was negatively affected by both submergence and exposure, but there was a marginally non-significant increase with optimal conditions (Figure 18d,e,f, Figure 19b,d,f, Table 13).

26

0.0

0.1

0.2

0.3

0.4

0.5

0.6 (a) (b) (c)

0 20 40 60 80 1000

10

20

30

40

50

60

70 (d)

0.0 0.4 0.8 1.2

(e)

0 4 8 12 16

(f)

Proba

bility

of Oc

curren

ce

Optimal Conditions Submergence Exposure

Perce

ntage

Cove

r

Figure 17. Effects of hydrologic regime on occurrence (top row) and cover (bottom row) of Common Couch for the southern sector. Interpretation of content is as for Figure 12.

0.0

0.1

0.2

0.3

0.4

0.5

0.6 (a) (b) (c)

0 20 40 60 80 1000

2

4

6

8

10 (d)

0.0 0.3 0.6 0.9 1.2 1.5 1.8

(e)

0 4 8 12 16

(f)

Proba

bility

of Oc

curre

nce

Optimal Conditions Submergence Exposure

Perce

ntage

Cove

r

Figure 18. Effects of hydrologic regime on occurrence (top row) and cover (bottom row) of Rat’s-tail Couch for the southern sector. Interpretation of content is as for Figure 12

27

-0.6

-0.4

-0.2

0.0

0.2

(a)

-3.0-2.5-2.0-1.5-1.0-0.50.0

(c)

CommonReed

CommonCouch

Rat-tailsCouch

-6-5-4-3-2-10

(e)

0.000.050.100.150.200.250.30

(b)

-0.6

-0.5

-0.4

-0.3

-0.2

-0.1 (d)

CommonReed

CommonCouch

Rat-tailsCouch

-1.0

-0.8

-0.6

-0.4

-0.2

(f)

Effect

of Op

timal

Cond

itions

Effect

of Su

bmerg

ence

Effect

of Ex

posur

e

Figure 19. Distributions of the regression parameters for the analyses of hydrologic regime effects on occurrence (left column) and cover (right column) of Common Reed, Common Couch and Rat’s-tail couch, when present, for the southern sector. The rows are the different hydrologic regime components Interpretation of plots is as for Figure 13.

Table 13. Results of hypothesis tests for the regression parameters in analyses of hydrologic regime effects on occurrence and cover of Common Reed, Common Couch and Rat’s-tail Couch in the southern sector. ‘p/a’ = presence/absence. Probabilities shown are p (x > 0) – i.e. a positive effect.

Sector → Common Reed Common Couch Rat’s-tail CouchAnalysis → p/a cover p/a cover p/a coverEffect of optimal conditions

0.581 0.964 0.702 >0.999 0.001 0.932

Effect of submergence

<0.001 0.001 <0.001 <0.001 0.253 <0.001

Effect of exposure >0.995 <0.001 <0.001 0.001 <0.001 <0.001

Spiny Mudgrass

Spiny Mudgrass occurred too infrequently in the eastern sector for any valid inferences. In the southern and western sectors, there was a strong negative effect of exposure on occurrence (Figure 20c,f, Figure 21c, Table 14). There was a strong negative effect of submergence on occurrence in the western sector, but the equivalent effect for the southern sector was

28

marginally non-significant (Figure 20b,e, Figure 21b, Table 14). There were on-average negative effects of optimal conditions on occurrence for both sectors, but neither effect was significant (Figure 20a,d, Figure 21a, Table 14). For analysis of live cover, there were sufficient data only for the southern sector. For this sector, there was a positive effect of optimal conditions, and negative effects of submergence and exposure on the cover of Spiny Mudgrass (Figure 20g,h,i, Figure 21a,b,c, Table 14).

0.0

0.2

0.4

0.6

0.8 (a) (b) (c)

0.0

0.2

0.4

0.6

0.8 (d) (e) (f)

0 50 100 150 200 2500

10

20

30

40 (g)

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

(h)

0 5 10 15 20

(i)

Proba

bility

of Oc

curre

nce

Proba

bility

of Oc

curre

nce

Percen

tage C

over

Optimal Conditions Submergence Exposure

Figure 20. Effects of hydrologic regime on occurrence and cover of Spiny Mudgrass. The top two rows are for occurrence in the southern and western sectors, respectively. The third row is for cover in the southern sector. Interpretation of content is as for Figure 12.

29

PA-Sth PA-West Cover-Sth

-0.8

-0.6

-0.4

-0.2

0.0

(a)

PA-Sth PA-West Cover-Sth

-1.5

-1.0

-0.5

0.0 (b)

PA-Sth PA-West Cover-Sth

-5

-4

-3

-2

-1

(c)

Effect

of Op

timal

Cond

itions

Effect

of Su

bmerg

ence

Effect

of Ex

posur

e

Figure 21. Distributions of the regression parameters for the analyses of hydrologic regime effects on occurrence of Spiny Mudgrass in the southern and western sectors, and on cover of Spiny Mudgrass in the southern sector. Within panels, ‘PA’ and ‘Cover’ refer to parameters for occurrence and cover, respectively. Interpretation of plots is as for Figure 13.

Table 14. Results of hypothesis tests for the regression parameters in analyses of hydrologic regime effects on occurrence of Spiny Mudgrass for the southern and western sectors, and cover of Spiny Mudgrass for the southern sector. Probabilities shown are p (x > 0) – i.e. a positive effect.

Analysis → Occurrence CoverSector → South West SouthEffect of optimal conditions 0.052 0.060 >0.999Effect of submergence 0.012 <0.001 <0.001Effect of exposure <0.001 <0.001 <0.001

Effects of hydrologic regime on shoreline erosion and deposition

For the peak wave height of 0.32 m, there were strong, but differing effects of number of days in the wave zone among the three sectors. There was a significant negative effect of exposure in the eastern sector – i.e. the greater the number of days of wave exposure, the greater the net erosion from that sector (Figure 22a, Figure 23a, Table 15). In the western sector, the effect was opposite – i.e. greater number of days of wave exposure resulted in an increase in net deposition. A similar trend was observed in the southern sector, but the effect was not significant (Figure 22b,c, Figure 23a, Table 15). Very similar effects were seen for the typical wave height of 0.14 m. The only difference was that the positive association between deposition and wave-zone days was significant for the southern sector, as well as the western (Figure 22d,e,f, Figure 23b, Table 15).

30

0 40 80 120 160-0.10

-0.05

0.00

0.05

0.10 (a)

0 40 80 120 160

(b)

0 40 80 120 160

(c)

0 20 40 60 80-0.10

-0.05

0.00

0.05

0.10 (d)

0 20 40 60 80

(e)

0 20 40 60 80

(f)

Vertic

al Dif

feren

ce (m

)

Wave Zone Days Wave Zone Days Wave Zone Days

Figure 22. Effects of wave-zone days on erosion and deposition. The two rows are for the maximal wave height of 0.32 m and typical wave height of 0.0.14 m, respectively. The three columns are eastern, southern and western sectors, respectively. Solid line is the median effect of number of wave-zone days. Dotted lines are the 2.5% and 97.5% limits of the credible interval for this parameter. The horizontal line is that of zero effect (and no erosion or deposition).

East South West

-0.010

-0.005

0.000

0.005

0.010 (a)

East South West

-0.010

-0.005

0.000

0.005

0.010 (b)

Effect

of In

unda

tion

Figure 23. Distributions of the effect of hydrologic regime on shoreline erosion and deposition. Panels are for 0.32 m waves and 0.14 m waves, respectively. Interpretation of plots is as for Figure 13.

The variation in the model attributable to the random effect of year of sampling was small compared to overall residual standard deviation for the model (Table 15). However, its inclusion did improve model performance slightly compared to other model structures tests (comparative results not shown).

31

Table 15. Results of hypothesis tests and magnitude of random effects for the model of effects of hydrologic regime (calculated for two different wave heights) on erosion and deposition. Probabilities shown are p (x > 0) – i.e. a positive effect. Magnitude of random effects is shown as the summary statistics for the distribution of standard deviation (mean ± 1.0 SD) attributable to sampling year (pooled across all three sectors), and the overall model residual for each sector.

Sector → East South WestMaximum wave height (0.32 m)

p(x>0) 0.010 0.756 0.992standard deviation among years 0.021±0.004

residual standard deviation 0.125±0.002 0.084±0.001 0.072±0.001Typical wave height (0.14 m)

p(x>0) 0.015 0.994 0.993standard deviation among years 0.021±0.004

residual standard deviation 0.125±0.003 0.084±0.001 0.072±0.002

Effects of live vegetation cover on shoreline erosion and deposition

There was a significant negative effect of vegetation live cover on rate of erosion and deposition in the eastern sector – i.e. increased vegetation cover was associated with decreased deposition up to around 20% cover, and thereafter with increasing erosion (Figure 24a, Figure 25, Table 16). The effect for both southern and western sectors was not significantly different from zero (Figure 24b,c, Figure 25, Table 16).

0 20 40 60 80 100-0.10

-0.05

0.00

0.05

0.10 (a)

0 20 40 60 80 100

(b)

0 20 40 60 80 100

(c)

Percentage Cover Percentage Cover Percentage Cover

Vertic

al Dif

feren

ce (m

)

Figure 24. Effect of vegetation cover on erosion and deposition for the eastern, southern and western sectors, respectively. Solid line is median effect, with dotted lines being the 2.5% and 97.5% limits of the credible interval.

32

East South West

-0.020

-0.015

-0.010

-0.005

0.000

0.005

0.010

Effec

t of V

egeta

tion

Figure 25. Distributions of the effect of vegetation cover on shoreline erosion and deposition. Interpretation of plots is as for Figure 13.

As with the model of hydrologic regime on erosion and deposition, the random effect of year of sampling explained relatively little of the variation in the data (Table 16). It is more difficult to compare the variation explained by year of sampling to overall model residual because of the scaling of data-level residuals with vegetation cover that was included in the model (see methods). However, the overall low magnitude of the figure suggests that it was relatively unimportant.

Table 16. Results of hypothesis tests and magnitude of random effects for the model of effects of vegetation cover on erosion and deposition. Probabilities shown are p (x > 0) – i.e. a positive effect. Magnitude of random effects is shown as the summary statistics for the distribution of standard deviation (mean ± 1.0 SD) attributable to sampling year (pooled across all three sectors), and the overall model residual for each sector. Note that model residual standard deviation was scaled by the uncertainty caused by Braun-Blanquet sampling for each data point in the analysis, and that this is not reflected in these overall figures.

Sector → East South Westp(x>0) 0.003 0.445 0.344

standard deviation among years 0.012±0.001residual standard deviation 0.093±0.004 0.027±0.001 0.024±0.002

33

Discussion

Vegetation

Spatial and temporal trends

We identified some clear spatial and temporal trends in the shoreline and dune vegetation of Lake Victoria. For the eastern and western sectors, there were increases in total vegetation live cover (including the cover of both annuals and perennials) with elevation. However, patterns for the southern sector were more variable with the cover of annuals decreasing with elevation and the cover of perennials peaking at around 26 m AHD (Appendix C). For the eastern and western sectors, there were also clear increases in total vegetation live cover (including the cover of annuals, perennials and the various growth forms) over time. In the southern sector, the cover of annuals decreased, while the cover of perennials (and sedges) increased, over the same period. These trends are likely to be driven by differences in the shoreline morphology and substrate of the different sectors of the lake, and changes to the hydrologic regime and stocking rates over the sampling period.

A variable hydrologic regime (such as that which results from the Lake Victoria Operating Strategy) creates a zonation of vegetation along an elevation gradient reflecting individual plants’ ability to cope with flooding stress (Blom & Voesenek 1996). Beyond full supply level, terrestrial species are likely to dominate. At Lake Victoria, plant growth at the lowest elevations of the lake shore (i.e. 24 m AHD) is likely to be restricted by near continual inundation. With increasing elevation, more opportunities for plant growth would occur, with flood-dependent and flood-tolerant plants expected at frequently-inundated lower elevations (~ 24-27 m AHD) and terrestrial species at elevations beyond the full supply level of Lake Victoria (> 27 m AHD). At Lake Victoria, flood-tolerant sedges, rushes and grasses tend to dominate these lower elevations, while terrestrial herb and woody species are in greatest abundance at higher elevations (Appendix C). The positive relationships of total vegetation live cover with elevation (between 24-30 m AHD) on the shorelines of the eastern and western sectors appear to be driven by the occurrence of these terrestrial species at higher elevations. For the western sector in particular, high cover values for woody vegetation above 27.5 m AHD (e.g. for Maireana pyramidata, Sago Bush, and Enchylaena tomentosa, Ruby Saltbush) appear to be driving the pattern in total live cover.

The differing trends with elevation of the eastern and western sectors compared to the southern sector are likely to be a result of distinct geomorphologies of these areas. The shorelines of the eastern and western sectors have relatively steeper gradients, consisting of beach, cliff and dune formations. In contrast, the southern sector consists of lower-lying, complex and undulating beach habitat, with ‘islands’ and barriers interspersed; habitat more suitable for flood-dependent and flood-tolerant species. Indeed, many of the transects in the southern sector did not extend beyond 27 m AHD – i.e. the amount of habitat for terrestrial species in this sector is much less. The peak in vegetation live cover at 25 – 26 m AHD in the southern sector may reflect

34

peak vegetation abundances on the barriers and islands. The linear regression used to model spatial vegetation patterns may be inappropriate for the southern sector; arguably a quadratic (or even more complex) model would have provided a better representation of changes in total cover with elevation.

There were clear increases in total vegetation live cover (including the cover of annuals and perennials and the various growth forms) over time for the eastern and western sectors. However, there was no such change for total cover in the south. The increases occurred from very low bases – i.e. cover was extremely low at the beginning of the sampling period. This may reflect the effects of stock grazing, and its subsequent removal, more so than any effect of lake operations or drought. Information on the stocking history of the Lake Victoria shoreline is patchy, but stock were excluded from the western sector transects in 2003, with a more recent exclusion (2009) occurring at Nulla Nulla station in the eastern sector (Sluiter 2011). In contrast, grazing pressure in the southern sector has been lower throughout the sampling period, and this sector has supported much higher average vegetation live cover. Much of the southern sector borders the Lake Victoria State Forest, which is not stocked, and Duncan’s Corner at the eastern end of the southern sector was destocked in 2000 (Carlile 2010). The substantial temporal increases in total vegetation live cover in the west is consistent with the 2003 exclusion of stock. The smaller increase in cover over time in the east may reflect the more recent exclusion of stock. Indeed, Sluiter (2011) observed the early stages of vegetation recovery along the KNU transect within this sector.

However, our results suggest that altered hydrologic regime due to drought is also partly responsible for the temporal patterns in vegetation live cover over time, as has been suggested previously (Bowman 2011). This would have resulted in changes in the relative suitability of habitat for groups of plants with different life history strategies (i.e. annuals and perennials). For example, over the sampling period, the cover of perennials increased in the southern sector despite no major changes in grazing pressure. During the sampling period, drought resulted in lake levels being drawn down from 27 m AHD earlier, and for longer, than previously. This would have resulted in shorter periods of inundation for shoreline elevations 24 – 27 m AHD, allowing more opportunities for growth for flood-tolerant perennial species. Indeed, many of the flood-tolerant perennial sedge and grass species dominant at these elevations are sensitive to prolonged submergence (see ‘Target species’ section below). Also, the cover of sedges appeared to increase in the southern sector over the sampling period. Conversely, the decrease in the cover of annual species over the same period probably also reflects the progressively drier conditions. Annual species may colonise moist sediments exposed by declining lake levels or be promoted by rainfall. While annual herbs were prominent during the early years of the drought (1998-2006), replacing a diverse lakeshore assemblage with a virtual monoculture in some quadrats (Sluiter 2011), the abundance of annuals is likely to decrease with continuing dry conditions (Stromberg et al. 2009). We suggest that the increase in vegetation live cover (of both annual and perennial species) in the eastern and western sectors over time is a combined result of reduced grazing and flooding stress, while the more modest increase in the cover of perennial species (and decrease in the cover of annuals) in the southern sector is more likely due to the lower stress caused by lack of flooding stress during the drought.

35

It should be noted however, that the statistical results for annual species are probably not as robust as those for total cover or perennials. Once the data were divided into annual and perennial species, there were many 0% cover readings for annual species. The linear model developed for total cover becomes less appropriate with an increasing proportion of 0% cover values (and indeed would not run at all for the different growth forms for this reason). We used the same model for total live cover and for annuals and perennials so as to maintain maximum comparability of the results. However the presence of 0% cover values means that the results for annual species may be less informative. Moreover, the value of yearly sampling for cover of annual species is questionable. By their nature, cover of annuals can vary considerably from year to year. We would expect annuals to colonise the shoreline as water levels recede, and to be abundant above lake water levels in wet years because of rainfall. With such variation, it is difficult to interpret long-term trends, and concentrating on annual variation may be more enlightening.

The random effects of transect and quadrat within transect explained a considerable amount of variation within the trend models. The models were structured to pool data from different transects to calculate spatial and temporal trends, with the random effects then able to explain extra variation in the data around this pooled trend line. This is best interpreted as overall cover in a transect or quadrat being on-average higher or lower than expected by the fitted trend line. The variation being explained by these random factors reflects small (quadrat) and medium (transect) scale variation in vegetation cover caused by (for example) differing soil quality or micro-topography. The inclusion of such variables in the trend models would have substantially increased their ability to discern trends in the data.

Effects of hydrologic regime on target species

Two of the three hydrologic regime parameters developed proved useful for predicting both the occurrence and liver of the five target species, with exposure and submergence being important negative influences. In contrast, while there was no indication that the ‘optimal conditions’ parameter, as developed from our literature search, had any beneficial effect on target species occurrence, it did have positive effects on live cover for all of the target species except Spiny Sedge. For management of these species, it may be that working to avoid excessive exposure and/or submergence of the elevations in which they occur (24-27 m AHD) will provide conditions sufficient to increase the occurrence of target species, but then providing optimal conditions will promote their growth.

Submergence reduced probability of occurrence for four out of the five target species, and live cover of all target species. For the two species where multiple sectors were tested (Spiny Sedge, Spiny Mudgrass), all trends for occurrence were negative, but only one was significantly negative. Only for Rat’s-tail Couch was there no indication of a negative effect of submergence on occurrence. However, live cover was still significantly negatively affected by submergence. Thus, submergence as defined was a robust indicator of both occurrence and live cover of the target species. While all of the target species are generally tolerant of prolonged periods of inundation (see Appendix B for refs), such conditions are likely to reduce their growth and

36

prevent their clonal spread or recruitment of new individuals. For example, new shoots and seedlings of Common Reed experience reduced growth rates when subjected to prolonged inundation (Roberts & Marston 2011). Intolerance of the target species to submergence concurs with our findings that vegetation live cover typically increases with elevation between (24-26 m AHD) and that the cover of perennial species has increased over a sampling period that has seen on-average lower water levels in the Lake due to extended drought (see ‘Spatial and temporal trends’ above).

Exposure was also significant predictor of reduced probability of occurrence for four out of the five target species, and reduced live cover for all target species. The effect was non-significant for the eastern sector analysis of occurrence of Spiny Sedge, but was still negative. There was a also a significant increase in cover of Spiny Sedge with exposure in the southern sector, contradicting results for the other two sectors. There was a non-significant increase in occurrence for Common Reed. Both these species have considerable tolerance to both flooding and drying (Blanch et al. 1999). Common Reed is extremely hardy. Once established, they can survive long periods of exposure (i.e. many years), with their rhizomes able to extend to 1.5 m in depth (and roots extending beyond this) to access groundwater (Rogers & Ralph 2011). Notably, both Common Reed and Spiny Sedge were commonly found at elevations above that at which the other target species occur (e.g. 27 – 28 m AHD). For Common Reed, we conservatively assessed exposure as occurring when the lake surface level was 0.5 m below the plant. This decision was taken because when assessing exposure, while we can assume that lake water penetrates into the soil at the same level as the lake surface for shallow depths, this assumption becomes increasingly invalid at greater depths because the plant is increasingly distant from the lake shore. It is possible had we defined exposure less conservatively, an effect of exposure on the occurrence of Common Reed may have been observed, but lake operations would have meant that the durations of such exposure periods would have been very short. The results suggest that this would have been a more realistic model structure for this species! Nonetheless, for the other target species, Common Couch, Rat’s-tail Couch and Spiny Mudgrass, exposure was a robust indicator of occurrence.

The ‘optimal conditions’ parameter is useful for assessing the adequacy of different hydrologic regimes for the growth of the target species (with the exception of Spiny Sedge). Optimal conditions was a robust indicator of increased live cover (which can be treated as an analogue for growth) for the other four target species. Optimal conditions for growth were considered to be those that would provide the plant with moist or waterlogged substrate conditions (i.e. water levels not below the bulk of their root mass) and not inundate the plant above its capacity for internal gas exchange. For example, the limit of ‘optimal depth’ for Common Reed and Spiny Mudgrass was 50 cm as they efficiently ventilate their submerged portions via tall or buoyant stems respectively. In contrast, the limit of ‘optimal depth’ for Common Couch and Rat’s Tail Couch was 0 cm (or at the level of the substrate). While both can survive long periods of inundation, they will only grow when water levels subside leaving their aerial parts exposed, but their roots wet. The failure of the ‘optimal conditions’ parameter to predict the live cover of Spiny Sedge suggests our selected values were not appropriate. In retrospect, a lower limit to optimal conditions of 20cm below surface level might be more appropriate for this species. The

37

fact that submergence and exposure were also good predictors of cover (as well as occurrence) is unsurprising. If a species is not able to persist under particular hydrologic conditions, then there will certainly be no ‘growth’ of that species under such conditions. It is noteworthy that while submergence and exposure were not prohibitive to the occurrence of Common Reed and Rat’s Tail Couch, respectively, these parameters were good indicators of limits to the hydrologic conditions amenable to their growth.

In contrast, the optimal conditions variable was unable to predict occurrence of any of the target species. Occurrence is reflective of past recruitment events, and it is likely that our model does not include all factors important for the promotion of recruitment events for the target species. For instance, recruitment events may be most sensitive to the timing of water level fluctuations, which our model does not consider. For example, an abundance of Spiny Sedge recruits were observed ~25 m AHD when water levels were held relatively constant at this level during its period of bulbil formation (Jan-Mar) (Bowman, 2011). However, because recruitment of new individuals has not been recorded previously, we had no data to test a recruitment model for any of the target species. Our conceptual model was originally designed to assess effects of hydrology on growth of the target species, and so the failure of one of the derived variables to successfully model recruitment processes should not be surprising.

Lastly, the common conceptual model adopted for the analyses of the five target species in relation to hydrological regime excludes some important processes known to be important for some of the species. For example, our review of the literature suggested that the timing of lake drawdown would be an important predictor of the growth of Spiny Mudgrass (Ward 1991). However, to maintain comparability among the results for the different species, a consistent model was adopted. Nevertheless, the results suggest that the model captured many of the important conditions affecting occurrence and growth of the target species.

Vegetation data

Collecting vegetation data at Lake Victoria only as live foliar projective covers fails to capture some important aspects of vegetation dynamics. Firstly, for species like Spiny Sedge, the canopy comprises thin erect cylindrical stems or culms, rather than flat layers of leaves. Estimating the cover of such structures is more difficult. Indeed the ‘low’ covers we discuss above may only be low because of the types of structures being surveyed. Second, the target species all form rhizomatous mats that will be of more importance for sediment stability on the Lake Victoria shoreline than will their foliage. Third, upon emersion from prolonged inundation (i.e. many months) some species may take considerable time to recover and produce new foliage. If plants are surveyed too soon after lake drawdown, such plants are not recorded at all (they are not even recorded as ‘dead’). Given the primary assumption of the Section 90 consent that vegetation can promote shoreline stability and sediment deposition (MDBC 2008), such data fails to capture the important contribution of rhizome masses temporarily devoid of live foliage.

38

Shoreline erosion and deposition

Spatial and temporal trends

The single most important result from the trend analysis of shoreline erosion and deposition is the spatial and temporal stability of sediments at Lake Victoria. Although individual dune-shoreline transects show dynamic behavior between surveys (Carlile 2010), the overall spatial and temporal trends described a very stable system within the survey period. Thus, it is reasonable to infer that the way in which Lake Victoria has been operated, in combination with existing smaller-scale management actions, is not causing excessive and sustained erosion of the Lake Victoria shoreline.

The trend analyses identified two exceptions to this general result. There was a positive effect of elevation on the rate of erosion/deposition for the eastern sector; lower shore elevations have experienced erosion over the period of record, but upper shore elevations have experienced deposition. This result is explained by the predominant wind direction (and hence wave direction) at Lake Victoria. A qualitative synthesis of wind rose data from the Mildura weather station (the nearest station with a sufficiently long-term data set; Appendix D) found predominantly southerly winds for the summer-autumn period and west-south-westerly winds for winter-spring. Thus shoreline transects in the eastern sector will be exposed to wind and wave energy to a far greater extent than transects in the other sectors.

The positive effect of elevation on rate of deposition is therefore consistent with wind (and to a lesser extent wave) energy pushing sediments up the beach face into the extensive dune system that exists along the eastern shore of Lake Victoria. The trend result provides no indication that there is a net gain or loss of sediment from the eastern sector shoreline transects as a whole. That is, the ‘area under curve’ for Figure 11a is approximately zero when the positive (deposition) and negative (erosion) areas are summed. Rather, the plot indicates net sediment re-distribution within this sector related to the transport of sand from the beach face landward into the sand dunes. However, we cannot be certain that longshore sediment transport is not occurring as part of this redistribution. The pattern is typical of wide beaches with strong onshore winds where vegetation is concentrated at the rear of the beach (Hesp 1988, 2002). The beach face will dry out during low lake levels, which then allows the sand to be reworked by the prevailing winds. Such transport could potentially be leading to sustained erosion of historically undisturbed sediments (HUS) on the lower beach face in the eastern sector.

A similar pattern was seen when we conducted the same analysis on all data (i.e. including those data beyond 30 m AHD elevation) except that the positive relationship with elevation for the eastern sector was weaker, and overall non-significant (results not shown). We suspect this occurs because at higher elevations, the shoreline transects extend into the dune system, with some transects going through blowout areas (D.M. Kennedy, pers. obs.). These areas do not experience net deposition because they are typically transport and deflation zones within the dune system (Hesp 2002).

39

The other exception to the generally observed spatial and temporal stability concerned temporal trends in the southern sector. The trend analysis identified net deposition at the beginning of the sampling period, reducing to net erosion by the end of it. We believe this result is caused by the active management of exposed burial sites in this sector. In the late 1990s and early 2000s, a large number of exposed burial sites were protected with sandbags and shade cloth. ‘Sand sausages’ (very long sand bags) were also used to prevent and reverse perceived erosion of the beach face. Such works were also complemented by sand nourishment in some places. Nearly all the transects in the southern sector cross these shore protection measures. The sediment transport processes on those profiles that do not cross these areas are also likely to be affected by them, though the precise influence would be difficult to quantify. Thus, the ‘net deposition’ observed in the early part of the time series of data may partly be sand nourishment. Sandbags, shadecloth and sand sausages are also designed to act as sediment traps. The trend result is consistent with these structures either trapping sediment in large amounts, or retaining the material placed on the sites, during the first years after their installation. This effect then diminished over time as the sediment trapping capacity of the structures was exhausted and the structures began to deteriorate. Net deposition was replaced with net erosion around 2005, and by the end of the sampling period the southern shore would have lost most of the sediment gained during the earlier years. It is impossible to be sure of this interpretation, because there were few data collected before active erosion management began, but the results certainly suggest that the management practices designed to protect burials and prevent erosion were effective, but probably now need to be renewed to prevent further erosion.

Exposure to potential wave action

Our a priori hypothesis was that areas of shore exposed to potential wave actions (i.e. wave-zone days) for larger number of days would experience net erosion. This hypothesis is also implicit in management actions that seek to minimize the number of days that deposits of historically undisturbed sediment (HUS) are near the lake surface – i.e. the practice of quickly filling and emptying the lake when its surface is near these elevations.

This hypothesis was supported by results for the eastern sector, with a strong negative relationship between rate of deposition and number of wave-zone days experienced (i.e. a positive relationship between net erosion and wave-zone days). It is worth noting that elevations beyond the full supply level of ~ 27 m AHD were exposed to no wave action. Thus the result for the analysis is consistent with, and provides a potential mechanistic explanation for, the elevation trend in erosion/deposition observed for the eastern sector. That is, wave action could be mobilizing sediments, which are then re-distributed further up the beach face (and beyond the maximal wave zone) by wind action if the beach is wide enough to dry out and is not covered in stabilizing vegetation. These results suggest that HUS on the eastern beach is being affected by exposure to wave actions, and provide support for the current LVOS practice of raising and lowering lake levels rapidly near HUS deposits.

However, the hypothesis was not supported by results from the southern and western sectors. These sectors experienced increased deposition with increasing number of wave-zone days. Why

40

might this be the case? The predominant wind directions at Lake Victoria mean that the southern and western sectors will experience little actual wave action – i.e. the waves travel north and east across the lake to impact upon the eastern sector. Fringing woody vegetation (mostly River Red Gum) will further act to reduce the impact of winds on the southern and western beaches by acting as a windbreak for the prevailing winds. Thus, while shores on the southern and western sectors will experience inundation, the amount of actual wave action will be minimal compared to the eastern sector, and erosive forces will be minor. The net deposition observed in Figure 22 cannot be directly explained by this lowered erosive force, and must therefore be caused by something correlated with the wave zone days variable.

The beaches of the western and southern sectors are more physically complex than the eastern beach. The western beach is characterized by large amounts of woody debris, a relic from when this area was dry land that was flooded by the flooding of Lake Victoria in 1928. Woody debris plays an important role in trapping sediment on beaches, and for sandy systems is critically important at the upper limit of the swash zone (Kennedy & Woods 2012). Its presence on the western shore likely contributes to the profile’s stability. The western beach may also experience a net influx of sediment from gullying of nearby alluvial fans. A small amount of aeolian transport from upwind dunes is also possible. The southern sector, as noted above, supports much higher densities of vegetation than the other sectors, and has been actively armored against erosion by sandbagging. This extra complexity would act to dissipate wave energy, and would also act as roughness elements to trap sediment. Such sediment may partly be derived from the sand nourishment works that have taken place in this sector, but is probably mostly derived from longshore transport from the western sector. In conclusion, inundation in the southern and western sectors may be sufficient to mobilize sediments, but the physical complexity of the environments may then act to trap sediment rather than re-distribute them further up the beach face.

This interpretation is consistent with our knowledge of processes and the environments of the western and southern sectors, and would suggest that rapid filling and emptying of the lake near the level of exposed burials may actually reduce the amount of deposition that occurs. However, exposed burials are primarily managed by point-scale management (i.e. sandbags and shadecloth) and the LVOS was not designed to prevent or reduce erosion around burials (J. Magee, Lake Victoria Scientific Reference Panel, pers. comm.).

More in-depth analysis and further targeted monitoring would be useful to better elucidate these mechanisms. For example, we note that the results were very similar for calculations of both 0.32 m waves (the maximum expectation) and 0.14 m waves (‘typical’ height). It must be remembered that both these metrics assume (inaccurately) a constant wave height throughout the year. Thus, these are not metrics describing the occurrence of waves at these heights, and the two metrics are highly correlated (explaining the similarity in results). Also, as noted in the methods, depth limitation in Lake Victoria also makes accurate modelling of peak wave heights difficult, and our calculations are likely to be overestimating peak wave heights, particularly at low water levels. It would be possible to use more rigorous modelling, along with data from the recently-installed weather station at Lake Victoria to more accurately model wave heights

41

through time. However, the strong results obtained from the present analysis would suggest that similar results are likely to be found.

Also, better characterization of the sediments in the lake shore may provide further information on the mechanisms driving the observed effects. Grain sizes are generally smaller with increasing elevation – a result of water transport and aeolian sorting. Sediment grain size information may tell us about the sources of deposited sediment in the southern and western segments; whether they are sourced primarily from the off-lake alluvial fans or being redistributed by long-shore sediment transport. We have recently obtained sediment samples from Lake Victoria and will include the analysis of those results in a journal paper examining the shoreline monitoring data.

Effects of vegetation cover

The Section 90 Consent to operate Lake Victoria as a water storage relies on the assumption that vegetation consolidates shoreline sediments, reducing erosion; and also acts as a sediment trap, promoting deposition. The results of the analysis of effects of vegetation live cover on shoreline erosion and deposition seem to contradict this assumption. We found no association between vegetation and erosion/deposition in the southern and western sectors. Even more surprising, there was a negative association in the eastern sector; the results suggest that increasing vegetation cover actually increases rates of erosion. However, we are confident that these are spurious results caused by the ways in which the data are collected. We have several potential explanations of the counterintuitive results for the eastern sector, but cannot explain why the result for this sector would be different to the other two. The limitations of the data set used in this analysis also need to be acknowledged; they prevent a meaningful analysis of the effect of vegetation on shoreline erosion and deposition.

Most seriously, the analysis had to assume that the vegetation live cover for a sampled quadrat is representative of a point of equivalent vertical elevation on a shoreline transect some distance away. Shoreline vegetation is clumped, and the 4 m2 size of the quadrats will be insufficient to capture the small-scale variability in vegetation data. Thus, vegetation cover on the nearby shoreline transect will only be loosely related to that of the sampled quadrat.

The data are also collected at fundamentally different scales. The vegetation live cover figure for a quadrat is an estimated average across the 4 m2. Shoreline transect data points are collected at the scale of the base of the survey staff – perhaps 0.001 m2. At this small scale, vegetation is likely to be either present (100%) or absent (0%), rather than occurring at the quadrat-scale average.

The analysis had an order of magnitude fewer data points – some 520 points – than did either of the other two analyses of erosion and deposition. This was a result of the need to match chainage points to vegetation quadrats, and led to the exclusion of many shoreline transects, plus the consideration of far fewer points along each transect than was the case for the other analyses. A lack of data may partly explain our inability to identify any associations for the southern and western sectors, particularly the western sector, for which there were only 64

42

relevant data points. However, this does not explain the counter-intuitive result for the eastern sector.

The analysis for the eastern sector was informed by 212 data points. The great majority of these points came from a single transect – Talgarry North (KTN). Thus the results cannot even be considered to be broadly representative of the eastern sector. Nevertheless, we suggest three potential explanations for the negative association observed.

First, vegetation acting as a highly efficient sediment trap may become largely covered by sediment between shoreline samples. We associated the vegetation sample prior to the second of the two survey points. Thus, if present vegetation had trapped a great deal of sediment – leading to a high deposition value, it might be recorded as having low vegetation! This effect would be reduced for species with erect foliage, such as Spiny Sedge, for which the culms will project through accumulated deposited sediments, and live cover figures will be minimally affected (Figure 26).

Figure 26. Talgarry North (KTN) vegetation quadrat 7, 2003, showing apparently recently-deposited sediments around established Spiny Sedge plants (Photo Ian Sluiter).

Nevertheless, Figure 27 shows the raw data used in this analysis, and it is apparent that many of the highest recorded deposition figures (high v.diff) are associated with extremely low vegetation cover near (or at) 0%.

Second, vegetation data are collected as ‘live’ foliar projective covers. Dormant or dead vegetation is not recorded, despite the role it may play in stabilizing sediments. In some cases, live and ‘dead’ vegetation cover will be negatively correlated. For example, Spiny Sedge rhizome masses that have been immersed for more than two weeks will normally have low cover of live culms shortly after emersion from the water (regrowth will begin shortly before or after emersion). However, the dead foliage may have been trapping sediments in the manner

43

hypothesized for the southern and western sector analyses of effects of hydrologic regime on erosion and deposition.

0 40 80 120-1.0

-0.6

-0.2

0.2

0.6 (a)

0 40 80 120

(b)

0 40 80 120

(c)

Vertic

al Diffe

rence

(m)

Percentage Cover Percentage Cover Percentage CoverFigure 27. Calculated v.diff amounts versus total vegetation cover in matched vegetation quadrats. The three panels are for eastern, southern, and western sectors, respectively.

Finally, and related to the mismatch of scales of the two data types, it is possible that localized sediment trapping by clumps of vegetation may be actively avoided by shoreline surveys (D.M. Kennedy, pers. obs.). Some operators may consider vegetation clumps and associated trapped sediment to be small-scale features not worthy of inclusion in a shoreline transect hundreds of meters long. Vegetation clumps of less than approximately 0.25 m2 may be ignored in shoreline surveys. However, this is exactly the type of sediment trapping that the assumption in the Section 90 consent is based on, and is visually apparent for many photopoints (e.g. Figure 28).

Figure 28. Talgarry North (KTN) vegetation quadrat 8, 2006. ‘Shadow dunes’ are apparent on the downwind side of several of the clumps of spiny sedge, as are sediments trapped within the clump in the foreground (Photo Ian Sluiter).

It is even possible that in actively avoiding these local features of the shoreline transect, operators may be taking survey points in bare areas more likely to be experiencing erosion. This effect would be worst in areas with high vegetation cover. Any or all of these mechanisms may

44

be responsible for the negative association observed between vegetation live cover and shoreline erosion / deposition in the eastern sector. What is apparent is that much-better targeted monitoring is required to adequately address this important question.

Historically undisturbed sediments

We have presented neither methods nor results for the analysis of historically-undisturbed sediments (HUS). This is because the data simply did not exist to support such an analysis. We initially suggested that we would identify chainage points identified as consisting of HUS and look at cumulative erosion and deposition of these points. However, we abandoned even this aim when it became apparent just how patchy the identification of HUS points were from year to year, and also how uncertain the identification of HUS is (W.J. Stephenson, pers. comm.). Any points identified in a data file as HUS in one year have a strong chance of being in error. Reporting erosion of such sediments may be deceptive. The targeted project undertaken by the University of Melbourne to identify HUS layers could form the basis of a new monitoring program designed to better track this culturally-important sediment layer.

Data processing and management

It was always an expectation that this project would expend considerable effort on data processing. It is also worth noting that the quality of the data provided by the MDBA, via the consultants who collected them, compares very favourably with other large monitoring data sets with which we are familiar. The data for lake level, in particular, were excellent. Nevertheless, it is worth documenting several aspects of the data that, if changed, could make future analyses of these data more straightforward.

By far the biggest difficulty encountered was inferring the presence of 0% cover in vegetation quadrats. Each row on the spreadsheet is the cover of one species in one quadrat at one time. With such a system, if a quadrat is completely empty, it does not appear on the data spreadsheet. However, some quadrats were not sampled every year. Reasons for not sampling included the quadrat being under water or being otherwise inaccessible at the time of sampling, or simply being lost because the marker stake had been buried by sand (I. Sluiter, pers. comm.). From the data provided it was difficult to tell whether a non-reported quadrat was empty (0% cover) or had not been sampled. There are two immediately-obvious ways of dealing with this. First, using the current spreadsheet structure, record a ‘species’ of ‘no plants’ in quadrats that have been sampled, but are empty. Second, replace the current data format with a matrix of samples (rows) x species (columns). Each sample consumes one row, with the cover of each species found being recorded in the appropriate column. If a quadrat is empty, all the cover cells are left empty. It is worth noting that this data recording issue is not confined to the Lake Victoria. For another major monitoring program (the Victorian Environmental Flows Monitoring and Assessment Program), one of us (JAW) has faced identical difficulties with assessing whether a vegetation quadrat was empty or was not sampled.

Both the vegetation and the shoreline data both also arrived in multiple versions. For shoreline data, there were three different versions: a single sheet containing all data, files that had a single

45

page for each year, and files that had a separate page for every transect sampled). These files did not always match perfectly. For example, we discovered shoreline monitoring transects that were included in the yearly files, but that had not been copied across to the ‘all data’ file. Unfortunately, we did not have the capacity to thoroughly check and correct such inconsistencies. It is likely that we missed some data in the analyses. Managing data through a database would prevent such ‘version control’ problems, as there would be one central repository.

Finally, we note that for processing data using a programming language such as R, the best format for data is to have a single row for every data point in the set, with redundant information (e.g. quadrat number, sampling date) repeated on every row. This was the format that we converted data to for the huge data processing tasks undertaken within this project (e.g. calculating ‘wave zone days’, ‘optimal conditions’, etc.). Such a format for data could be easily achieved using a basic query from a central database.

Recommendations for future monitoring

The existing monitoring data have proved to be a rich source of data for assessing spatial and temporal trends in shoreline vegetation cover and erosion and deposition. Moreover, they provided data for the analyses of the effects of hydrologic regime on these variables, which also produced strong results. Any redesign of the monitoring program should allow for the continued analysis of temporal trends back to the late 1990s. This could be achieved by maintaining at least a sub-set of transects within the existing monitoring programs. It may also be possible to convert data collected using new methods and transects to the scales used in the existing program, and therefore allow temporal trend analysis of the entire data set. This would require the establishment of ‘double sampling’ protocols to establish the relationship between data collected under the new program and the methods of collection used currently.

Shoreline monitoring should continue to include the twice-yearly monitoring of transects. This is attempting to ascertain whether shoreline processes vary seasonally, but we believe there are as yet insufficient data to make this determination.

The monitoring program as currently structured cannot address more detailed hypotheses concerning the mechanisms underlying changes in shoreline profiles or vegetation. Most importantly, the programs cannot address the key question of whether vegetation can stabilize lakeshore sediments and promote deposition. Below, we provide some suggestions for future monitoring programs that may better be able to elucidate these mechanisms.

1. Establish transects or plots that are surveyed for both vegetation and change in surface profile. Shoreline profiles within these plots should be done at multiple randomly located points within the larger quadrat, and the cover of vegetation at that point should be recorded. This close matching of data points, would much better allow us to determine whether vegetation can promote shoreline stability and deposition.

46

2. Establish transects or plots designed to target the erosion of Historically Undisturbed Sediments (HUS). The shoreline monitoring data collected to date are insufficient to say anything about whether there is sustained erosion of these sediments. The concurrent project being undertaken by the University of Melbourne could be used to design these surveys. Of most importance is repeated surveys of the same points on the HUS face, so that it can be confidently inferred whether erosion is occurring.

3. Undertake monitoring to quantify longshore sediment transport. This would allow us to determine whether the trend in erosion with elevation and with exposure to wave zone days in the eastern sector is predominantly a vertical redistribution of sediments, or whether sediments are moving along the shore. It would also allow us to identify the sources of sediment deposited during inundation in the western and southern sectors.

4. Consider a targeted research project and/or monitoring to further investigate the apparent positive effect of ‘wave zone days’ on sediment deposition in the culturally-important southern shore area.

5. Future vegetation monitoring, including at existing transects should record the cover of dead or dormant foliage, and of any exposed rhizome clumps, in addition to the live foliar projective covers currently being recorded. These should be recorded as different types of cover in the data sheets, but all will be important for trapping sediments.

6. Consider monitoring of vegetation dynamics to allow better statistical modelling of the mechanisms driving changes in vegetation cover. In particular, there is currently no monitoring of reproduction and recruitment, despite the fact that we know these will be affected by lake operations (Roberts & Marston 2011).

7. If it is important to monitor short-lived (i.e. annual) species, then seasonal monitoring of would be more informative than annual monitoring.

8. Lastly, improved data management through a central database or similar would reduce the amount of pre-processing required for data analysis and reduce the likelihood of errors. Data preparation and cleaning is an acknowledged part of any analysis, but can be reduced with a little care when the data are first compiled. The database design and implementation undertaken for the Victorian Environmental Flows Monitoring and Assessment Program (VEFMAP) may be a suitable template to use (Webb et al. 2010a).

47

References cited

Andrieu C., de Freitas N., Doucet A. & Jordan M.I. (2003). An introduction to MCMC for machine learning. Machine Learning, 50, 5-43.

Barnard G.A. (1958). Studies in the history of probability and statistics .IX. Thomas Bayes' essay towards solving a problem in the doctrine of chances. Biometrika, 45, 293-295.

Blanch S.J., Ganf G.G. & Walker K.F. (1999). Tolerance of riverine plants to flooding and exposure indicated by water regime. Regulated Rivers-Research & Management, 15, 43-62.

Blom C. & Voesenek L. (1996). Flooding: the survival strategies of plants. Trends Ecol. Evol., 11, 290-295.

Bowman H. (2011). Observations of outcomes of the targeted operation of Lake Victoria for the benefit of foreshore vegetation in 2010-11, and recommendations for 2011-12 water year. Murray-Darling Basin Authority, Canberra.

Bowman H. & Davis K. (2012). Operating Lake Victoria storage (western NSW) to establish a foreshore resilient to extreme conditions by establishing a self sustaining bio-physical system. In: Proceedings of the 6th Australian Stream Management Conference, Managing for Extremes (eds. Grove JR & Rutherfurd ID). River Basin Management Society, Canberra, Australia, pp. 425-432.

Braun-Blanquet J. (1964). Pflanzensoziologie: Grundzuge der Vegetation Skunde. 3rd edn. Springer, Vienna.

Brooks S.P. & Gelman A. (1998). General methods for monitoring convergence of iterative solutions. Journal of Computational and Graphical Statistics, 7, 434-455.

Carlile P. (2010). Review of Lake Victoria vegetation and shore-change monitoring data. Murray-Darling Basin Authority, Canberra.

Cheal D.C. & Parkes D.M. (1989). Mallee vegetation in Victoria. In: Mediterranean Landscapes in Australia — Mallee Ecosystems and their Management (eds. Noble JC & Bradstock RA). CSIRO, East Melbourne.

Clark J.S. (2005). Why environmental scientists are becoming Bayesians. Ecology Letters, 8, 2-14.

Gelman A., Rubin J.B., Stern H.S. & Rubin D.B. (2004). Bayesian Data Analysis. Chapman & Hall/CRC, Boca Raton, FL.

Hesp P. (1988). Morphology, dynamics and internal stratification of some established foredunes in southeast Australia. Sedimentary Geology, 55, 17-41.

Hesp P. (2002). Foredunes and blowouts: initiation, geomorphology and dynamics. Geomorphology, 48, 245-268.

Hudson J. & Bowler J. (1997). The cultural heritage of Lake Victoria. 7. Geomorphology, stratigraphy and sand resources. Murray-Darling Basin Commission, Canberra.

48

Kennedy D.M. (2012). Topographic field surveying in geomorphology. In: Methods in Geomorphology: Treatise of Geomorphology (eds. Switzer AD & Kennedy DM). Elsevier.

Kennedy D.M. & Woods J.L.D. (2012). The influence of coarse woody debris on gravel beach geomorphology. Geomorphology, 159-160, 106-115.

Kéry M. (2010). Introduction to WinBUGS for Ecologists. Elsevier, Chennai, India.

Lunn D., Spiegelhalter D., Thomas A. & Best N. (2009). The BUGS project: evolution, critique and future directions (with discussion). Statistics in Medicine, 28, 3049-3082.

Martin T.G., Kuhnert P.M., Mengersen K. & Possingham H.P. (2005). The power of expert opinion in ecological models using Bayesian methods: Impact of grazing on birds. Ecological Applications, 15, 266-280.

McCarthy M.A. (2007). Bayesian methods for ecology. Cambridge University Press, Cambridge, UK; New York.

MDBC (2002). Lake Victoria Operating Strategy. Murray-Darling Basin Ministerial Council, Canberra.

MDBC (2008). Lake Victoria cultural landscape plan of management. Murray-Darling Basin Commission, Canberra.

R Development Core Team (2010). R: A language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria.

Roberts J. & Marston F. (2011). Water regime for wetland and floodplain plants: a source book for the Murray–Darling Basin. National Water Commission, Canberra.

Rogers K. & Ralph T.J. (2011). Floodplain wetland biota in the Murray-Darling basin: water and habitat requirements CSIRO Publishing, Collingwood, VIC. Australia.

Sluiter I.R.K. (2011). Flora and fauna of the Lake Victoria area, southwest New South Wales: 14. Annual vegetation monitoring at Lake Victoria, winter 2010. Part 1: main report. Ogyris Ecological Research, Merbein.

Sluiter I.R.K. & Robertson P. (1999). Flora and fauna of the Lake Victoria area, southwest New South Wales. 2. Establishment and analysis of permanent vegetation and waterbird monitoring sites. Ogyris Ecological Research, Merbein.

Stephenson W.J. (2012). Review of high energy wind events trigger and profile monitoring at Lake Victoria, New South Wales. Melbourne Consulting and Custom Programs, Melbourne.

Stephenson W.J. & Kennedy D.M. (2011). Lake Victoria monitoring programme: assessment of shoreline change 2011. University of Melbourne, Melbourne.

Stephenson W.J., Thornton L. & Marshall R. (2009). Lake Victoria Monitoring Programme: Assessment of shoreline change 2009. Melbourne Consulting and Custom Programs & S.A. Water, Melbourne.

49

Stromberg J.C., Hazelton A.F. & White M.S. (2009). Plant species richness in ephemeral and perennial reaches of a dryland river. Biodiversity and Conservation, 18, 663-677.

US Army Corps of Engineers (1984). Shore protection manual. 4th edn. Deptartment of the Army, Waterways Experiment Station, Corps of Engineers, Coastal Engineering Research Center, Washington, DC.

Ver Hoef J.M. & Frost K.J. (2003). A Bayesian hierarchical model for monitoring harbor seal changes in Prince William Sound, Alaska. Environmental and Ecological Statistics, 10, 201-219.

Walker J. & Tunstall B.R. (1981). Field estimation of foliage cover in Australian woody vegetation. CSIRO.

Ward K. (1991). Investigation of the flood requirements of the Moira grass plains in Barmah Forest, Victoria. Department of Conservation and Environment, Victoria, Melbourne.

Webb J.A., Stewardson M.J., Chee Y.E., Schreiber E.S.G., Sharpe A.K. & Jensz M.C. (2010a). Negotiating the turbulent boundary: the challenges of building a science-management collaboration for landscape-scale monitoring of environmental flows. Marine and Freshwater Research, 61, 798-807.

Webb J.A., Stewardson M.J. & Koster W.M. (2010b). Detecting ecological responses to flow variation using Bayesian hierarchical models. Freshwater Biology, 55, 108-126.

Winton T.C., Chou I.B., Powell G.M. & Crane J.D. (1981). Analysis of coastal sediment transport processes from Wrightsville Beach to Fort Fisher, North Carolina. Army Engineer Waterways Experiment Station, Coastal Engineering Research Center.

50

Appendix A

OpenBUGS code

Vegetation trend model

Main Data Frameyear[] sect[] trans[] quad[] elev[] mp[] lb[] ub[]7 1 1 1 28.48 0.0 0 08 1 1 1 28.48 0.0 0 0…1090 rows

Additional Data. Index points for the different sectors in the analysis. They tell the model where to start counting among quadrats and transects for the different sectors.

list(start=1,finish=1090,q.start=c(1,37,81),q.finish=c(36,81,98),t.start=c(1,6,11),t.finish=c(5,10,13),n.sect=3,min.elev=24.1,n.elev.g.points=c(30,30,30),res.elev=0.2,n.year.g.points=27,res.year=0.5)

Model{for (i in start:finish){

sqrt.mp[i] <- sqrt(mp[i]) #transform mid-point, lower bound, and upper bound datasqrt.lb[i] <- sqrt(lb[i])sqrt.ub[i] <- sqrt(ub[i])

c.elev[i] <- (elev[i]-mean(elev[]))/sd(elev[]) #center and standardize covariatesc.year[i] <- (year[i]-mean(year[]))/sd(year[])

sqrt.mp[i] ~ dnorm(mu[i],tau[sect[i]])mu[i] <- int[sect[i]] + eff.elev[sect[i]] * c.elev[i] + eff.year[sect[i]] * c.year[i] + eff.bb[i] + eff.quad[sect[i],quad[i]] +

eff.trans[sect[i],trans[i]]#cont effects of elev and year, separate within each sector#semi-random effect of braun-blanquet imprecision#random effect of quadrat#random effect of transect

neg.range[i] <- sqrt.lb[i] - sqrt.mp[i] #uneven range for the bb effect in transformed spacepos.range[i] <- sqrt.ub[i] - sqrt.mp[i]eff.bb[i] ~ dunif(neg.range[i],pos.range[i]) #random effect for braun-blanquet imprecision

}

for (i in 1:n.sect){ for (j in q.start[i]:q.finish[i]){ #count through quadrat numbers for each sector

eff.quad[i,j] ~ dnorm(0,q.tau[i]) #just estimate variance from quadrat - zero mean}for (j in t.start[i]:t.finish[i]){

eff.trans[i,j] ~ dnorm(0,t.tau[i]) #same approach as for eff.quad[]}

#independent priors for each sector; no hierarchicy

int[i] ~ dnorm(0,0.01) #vague normal priors for regression parameterseff.elev[i] ~ dnorm(0,0.01)eff.year[i] ~ dnorm(0,0.01)

tau[i] <- 1/(sd[i]*sd[i]) #transform precision to standard deviationsd[i] ~ dunif(0,10) #prior sd = 10 for possible data range of 10; pretty vague

q.tau[i] <- 1/(q.sd[i]*q.sd[i])q.sd[i] ~ dunif(0,10)

t.tau[i] <- 1/(t.sd[i]*t.sd[i])t.sd[i] ~ dunif(0,10)

p.eff.elev[i] <- step(eff.elev[i]) #probability of an increase in cover with elevationp.eff.year[i] <- step(eff.year[i]) #probability of an increase in cover over time

for (j in 1:n.elev.g.points[i]){ #predictions to graph changes with elevation

g.elev[i,j] <- min.elev + (j - 1) * res.elev #specification of elevations for graphing

51

c.g.elev[i,j] <- (g.elev[i,j]-mean(elev[]))/sd(elev[]) #center and standardize these elevationse.fitted[i,j] <- int[i] + eff.elev[i] * c.g.elev[i,j] #fitted value at each elevation point (at mean year)

#back-transform percentile values in R and truncate at 0#doing it in BUGs doesn't work because of negative values

}

for (j in 1:n.year.g.points){ #predications to graph changes over time at mean elevationg.year[i,j] <- 1 + (j - 1) * res.year #don't need to do this three times, but easier to codec.g.year[i,j] <- (g.year[i,j]-mean(year[]))/sd(year[]) y.fitted[i,j] <- int[i] + eff.year[i] * c.g.year[i,j]

}}

}

52

Shoreline erosion and deposition trend mod

Main Data Framesect[] s.date[] s.elev[] s.v.diff[]1 3773 28.02 -0.1341 4094 27.90 0.261…7139 rows

Additional Data. Index points

list(start=1, finish=6972, n.sect=3, min.elev=22.5, n.elev.g.points=c(16,16,16), res.elev=0.5, min.date=c(1300,0,1500), n.date.g.points=c(45,58,43), res.date=100)

Model{for (i in start:finish){

c.s.elev[i] <- (s.elev[i] - mean(s.elev[]))/sd(s.elev[]) #center and standardize covariatesc.s.date[i] <- (s.date[i] - mean(s.date[]))/sd(s.date[])

s.v.diff[i] ~ dnorm(mu[i],tau[sect[i]])mu[i] <- int[sect[i]] + eff.date[sect[i]]*c.s.date[i] + eff.elev[sect[i]]*c.s.elev[i]

#cont effects of elev and date, separate within each sector}

for (i in 1:n.sect){int[i] ~ dnorm(0,0.001) #priors on regression parameterseff.elev[i] ~ dnorm(0,0.001)eff.date[i] ~ dnorm(0,0.001)

tau[i] <- 1/(sd[i]*sd[i]) #prior sdsd[i] ~ dunif(0,100)

p.eff.elev[i] <- step(eff.elev[i]) #probability of an increase in cover with elevationp.eff.date[i] <- step(eff.date[i]) #probability of an increase in cover with time

for (j in 1:n.elev.g.points[i]){ #predictions to graph changes with elevationg.elev[i,j] <- min.elev + (j - 1) * res.elev #specification of elevations for graphingc.g.elev[i,j] <- (g.elev[i,j]-mean(s.elev[]))/sd(s.elev[]) #center and standardize these elevationse.fitted[i,j] <- int[i] + eff.elev[i] * c.g.elev[i,j] #fitted value at each elevation point (at mean year)

#no back-transformation required for this one}

for (j in 1:n.date.g.points[i]){ #predictions to graph changes with timeg.date[i,j] <- min.date[i] + (j - 1) * res.datec.g.date[i,j] <- (g.date[i,j]-mean(s.date[]))/sd(s.date[])y.fitted[i,j] <- int[i] + eff.date[i] * c.g.date[i,j]

}}

}

53

Effect of hydrologic regime on target species occurrence

Main Data Framesect[] pa[] opt[] inund[] expos[]1 0 0 0.00 10.211 0 0 0.00 11.53…1090 rows

Additional Data. Index pointslist(N=1024,n.sect=3,n.g.points=50)

Model{for (i in 1:N){

c.opt[i] <- (opt[i] - mean(opt[]))/sd(opt[]) #centre and standardize independent variablesc.inund[i] <- (inund[i] - mean(inund[]))/sd(inund[]) #same as for veg trend modelsc.expos[i] <- (expos[i] - mean(expos[]))/sd(expos[])

pa[i] ~ dbern(p[i])logit(p[i]) <- int[sect[i]] + eff.opt[sect[i]] * c.opt[i] + eff.inund[sect[i]] * c.inund[i] + eff.expos[sect[i]] * c.expos[i]

#logistic regression for presence/absence#separate models for each sector

}

for (i in 1:n.g.points){ #for graphing effects of hydrologic variablesg.opt[i] <- ranked(opt[],1) + (((ranked(opt[],N) - ranked(opt[],1)) / (n.g.points-1))) * (i-1)

#creates 50 evenly spaced points from min to maxc.g.opt[i] <- (g.opt[i]-mean(opt[]))/sd(opt[]) #put this onto the same scale as used in the regression

g.inund[i] <- ranked(inund[],1) + (((ranked(inund[],N) - ranked(inund[],1)) / (n.g.points-1))) * (i-1)c.g.inund[i] <- (g.inund[i]-mean(inund[]))/sd(inund[])

g.expos[i] <- ranked(expos[],1) + (((ranked(expos[],N) - ranked(expos[],1)) / (n.g.points-1))) * (i-1)c.g.expos[i] <- (g.expos[i]-mean(expos[]))/sd(expos[])

for(j in 1:n.sect){logit(g.p.opt[j,i]) <- int[j] +eff.opt[j] * c.g.opt[i] #graph the change in p with variable, holds all else constant

#for graphing, do g.p v g. variableslogit(g.p.inund[j,i]) <- int[j] +eff.inund[j] * c.g.inund[i]logit(g.p.expos[j,i]) <- int[j] +eff.expos[j] * c.g.expos[i]

}}

for(i in 1:n.sect){ #priors for regression slope parametersint[i] ~ dnorm(0,0.01)eff.opt[i] ~ dnorm(0,0.01)eff.inund[i] ~ dnorm(0,0.01)eff.expos[i] ~ dnorm(0,0.01)

p.eff.opt[i] <- step(eff.opt[i]) #hypothesis test for the effect of the hydrologic variablesp.eff.inund[i] <- step(eff.inund[i])p.eff.expos[i] <- step(eff.expos[i])

}}

54

Effect of hydrologic regime on target species cover, when present

Main Data Framesect[] mp[] lb[] ub[] opt[] inund[] expos[]1 0.5 0 1 37 0.16 0.401 0.5 0 1 0 0.00 2.73…differing number of rows for each species

Additional Data. Index pointslist(N=533,n.sect=3,n.g.points=50)

Model{ #ended up running 1 or 2 sectors for all but spiny sedgefor(i in 1:N){

c.opt[i] <- (opt[i] - mean(opt[]))/sd(opt[]) #centre and standardize independent variablesc.inund[i] <- (inund[i] - mean(inund[]))/sd(inund[])c.expos[i] <- (expos[i] - mean(expos[]))/sd(expos[])

t.lb[i] <- max(lb[i],0.25) #need to fudge the zeros a bit because of log-scaleneg.range[i] <- log(t.lb[i]) - log(mp[i]) #braun-blanquet effect as done for veg trend modelpos.range[i] <- log(ub[i]) - log(mp[i]) #but on a different scaleeff.bb[i] ~ dunif(neg.range[i],pos.range[i])

mp[i] ~ dpois(lambda[i])log(lambda[i]) <- int[sect[i]] + eff.opt[sect[i]] * c.opt[i] + eff.inund[sect[i]] * c.inund[i] + eff.expos[sect[i]] * c.expos[i] +

eff.bb[i] #log-poisson regression for data with cover > 0#same driving variables as pa plus the bb random effect

}

for(i in 1:n.g.points){ #for graphing effects of hydrologic variables#y axis on these ones is cover

g.opt[i] <- ranked(opt[],1) + (((ranked(opt[],N) - ranked(opt[],1)) / (n.g.points-1))) * (i-1)#creates 50 evenly spaced points from min to max

c.g.opt[i] <- (g.opt[i]-mean(opt[]))/sd(opt[]) #put this onto the same scale as used in the regression

g.inund[i] <- ranked(inund[],1) + (((ranked(inund[],N) - ranked(inund[],1)) / (n.g.points-1))) * (i-1)c.g.inund[i] <- (g.inund[i]-mean(inund[]))/sd(inund[])

g.expos[i] <- ranked(expos[],1) + (((ranked(expos[],N) - ranked(expos[],1)) / (n.g.points-1))) * (i-1)c.g.expos[i] <- (g.expos[i]-mean(expos[]))/sd(expos[])

for(j in 1:n.sect){log(g.c.opt[j,i]) <- int[j] +eff.opt[j] * c.g.opt[i] #graph the change in cover with variable, all else constant

#for graphing, do g.p v g. variableslog(g.c.inund[j,i]) <- int[j] +eff.inund[j] * c.g.inund[i]log(g.c.expos[j,i]) <- int[j] +eff.expos[j] * c.g.expos[i]

}}

for(i in 1:n.sect){ #priors for regression slope parametersint[i] ~ dnorm(0,0.01)eff.opt[i] ~ dnorm(0,0.01)eff.inund[i] ~ dnorm(0,0.01)eff.expos[i] ~ dnorm(0,0.01)

p.eff.opt[i] <- step(eff.opt[i]) #hypothesis test for the effect of the hydrologic variablesp.eff.inund[i] <- step(eff.inund[i])p.eff.expos[i] <- step(eff.expos[i])

}}

55

Effect of hydrologic regime on shoreline erosion and deposition

Main Data Frame Additional data frames for 14 and 32 cm inund variable; load 1

14cm 32cmsect[] s.elev[] v.diff[] end.year[] inund[] inund[]1 26.91 -0.098 12 35 451 26.81 -0.028 13 0 0…6978 rows

Additional Data. Index pointslist(start=1, finish=6978, n.sect=3, n.year=17) < 30 mlist(n.inund.g.points=17) for graphing effects of 14 cm inundation variablelist(n.inund.g.points=33) for graphing effects of 32 cm inundation variable

Model{for (i in start:finish){

c.s.elev[i] <- (s.elev[i] - mean(s.elev[]))/sd(s.elev[]) #standardize elevation covariatec.inund[i] <- (inund[i] - mean(inund[]))/sd(inund[]) #standardize the inundation predictor variables

v.diff[i] ~ dnorm(mu[i],tau[sect[i]])mu[i] <- eff.inund[sect[i]] * c.inund[i] + eff.year[end.year[i]] + eff.elev[sect[i]]*c.s.elev[i] # int[sect[i]] +

#cont. effects of inundation and elevation, cat. effect of year#int not included because it was correlated with eff.year

}

for (i in 1:n.sect){int[i] ~ dnorm(0,0.001) #priors for continuous parameters - 1 per sectoreff.inund[i] ~ dnorm(0,0.001)eff.elev[i] ~ dnorm(0,0.001)

tau[i] <- 1/(sd[i]*sd[i]) #prior for model uncertainty - 1 per sectorsd[i] ~ dunif(0,100)

p.eff.inund[i] <- step(eff.inund[i]) #hypothesis test on effect of inundation variable

for (j in 1:n.inund.g.points){ #predictions to graph changes with inundationg.inund[i,j] <- (j - 1) * 5 #specification of 5 day intervals for graphingc.g.inund[i,j] <- (g.inund[i,j]-mean(inund[]))/sd(inund[])

#center and standardize the inundation graphing variablei.fitted[i,j] <- eff.inund[i] * c.g.inund[i,j] + mean(eff.year[])

#mean effect of year takes place of intercept in prediction#fitted value at each inundation point (at mean year)

}}

for (i in 1:n.year){ #priors for categorical random varibleseff.year[i] ~ dnorm(0,y.tau)

}y.tau <- 1/ (y.sd*y.sd) #hyperpriors on random effect of yeary.sd ~ dunif(0,100)

}

56

Effect of vegetation cover on shoreline erosion and deposition

Main Data Framesect[] mp[] lb[] ub[] end.yr[] n.days[] v.diff[]1 0.5 0 1 11 167 0.3501 0.5 0 1 12 190 0.217…533 rows

Additional data. Index pointslist(n.sect=3,n.year=14,N=533,n.g.points=51)

Model{for (i in 1:N){

sqrt.mp[i] <- sqrt(mp[i]) #transform mid-point, lower bound, and upper bound datasqrt.lb[i] <- sqrt(lb[i])sqrt.ub[i] <- sqrt(ub[i])

c.mp[i] <- (sqrt.mp[i]-mean(sqrt.mp[]))/sd(sqrt.mp[]) #centre and scale transformed midpointc.days[i] <- (n.days[i]-mean(n.days[]))/sd(n.days[]) #center and standardize covariate

v.diff[i] ~ dnorm(mu[i], tau[i]) #precision calculated separately for each data pointmu[i] <- eff.veg[sect[i]] * c.mp[i] + eff.days[sect[i]] * c.days[i] + eff.year[end.yr[i]]

#linear model with random effect of sampling year

tau[i] <- 1/(sd[i] * sd[i]) #scale residuals by braun-blanquet imprecisionbb.range[i] <- max((sqrt.ub[i] - sqrt.lb[i]),0.5) #need to use 0.5 for 0% coversd[i] <- s.sd[sect[i]] * bb.range[i]

}

for (i in 1:n.sect){eff.veg[i] ~ dnorm(0,0.01) #regression parameter priorseff.days[i] ~ dnorm(0,0.01)

s.sd[i] ~ dunif(0,10) #prior on sector-level standard deviation

p.eff.veg[i] <- step(eff.veg[i]) #hypothesis test of main effect of interest}

for (i in 1:n.year){eff.year[i] ~ dnorm(0,y.tau) #priors on random effect

}y.tau <- 1/(y.sd * y.sd) #hyperprior for random effect precisiony.sd ~ dunif(0,10)

for (i in 1:n.g.points){g.mp[i] <- sqrt((i-1) * 2) #graphing point each two percent from 0 to 100% coverc.g.mp[i] <- (g.mp[i]-mean(sqrt.mp[]))/sd(sqrt.mp[]) #put onto same scale as analysis

for (j in 1:n.sect){v.d.fitted[j,i] <- eff.veg[j] * c.g.mp[i] + mean(eff.year[])

#fitted vertical differencess.v.d.fitted[j,i] <- v.d.fitted[j,i] / mean(n.days[]) * 365

#scaled to v.diff per year}

}}

57

Appendix B

Literature used to develop the vegetation conceptual models

Blanch S.J., Ganf G.G. & Walker K.F. (1999). Tolerance of riverine plants to flooding and exposure indicated by water regime. Regulated Rivers-Research & Management, 15, 43-62.

Blanch S.J., Walker K.F. & Ganf G.G. (2000). Water regimes and littoral plants in four weir pools of the River Murray, Australia. Regulated Rivers-Research & Management, 16, 445-456.

Bowman H. (2011). Observations of outcomes of the targeted operation of Lake Victoria for the benefit of foreshore vegetation in 2010-11, and recommendations for 2011-12 water year. Murray–Darling Basin Authority, Canberra.

Cook G.D., Setterfield S.A. & Maddison J.P. (1996). Shrub invasion of a tropical wetland: Implications for weed management. Ecological Applications, 6, 531-537.

Ekstam B., Johannesson R. & Milberg P. (1999). The effect of light and number of diurnal temperature fluctuations on germination of Phragmites australis. Seed Science Research, 9, 165-170.

Finlayson C.M. (1991). Production and major nutrient composition of 3 grass species on the Magela floodplain, Northern Territory, Australia. Aquatic Botany, 41, 263-280.

Finlayson C.M., Cowie I.D. & Bailey B.J. (1990). Sediment seedbanks in grassland on the Magela Creek floodplain, Northern Australia. Aquatic Botany, 38, 163-176.

Furness H.D. & Breen C.M. (1985). Interactions between period of exposure, grazing and crop growth-rate of Cynodon dactylon (L) Pers in seasonally flooded areas of the Pongolo River floodplain. Hydrobiologia, 126, 65-73.

Greenwood M.E. & MacFarlane G.R. (2006). Effects of salinity and temperature on the germination of Phragmites australis, Juncus kraussii, and Juncus acutus: implications for estuarine restoration initiatives. Wetlands, 26, 854-861.

Lenssen J.P.M., ten Dolle G.E. & Blom C. (1998). The effect of flooding on the recruitment of reed marsh and tall forb plant species. Plant Ecology, 139, 13-23.

MDBC (2002). Lake Victoria Operating Strategy. Murray-Darling Basin Ministerial Council, Canberra.

Roberts J. (2002). Species-level knowledge of riverine and riparian plants: a constraint for determining flow requirements in the future. Australian Journal of Water Resources, 5, 21-31.

58

Roberts J. & Marston F. (2011). Water regime for wetland and floodplain plants: a source book for the Murray–Darling Basin. National Water Commission, Canberra.

Rogers K. & Ralph T.J. (2011). Floodplain wetland biota in the Murray-Darling basin: water and habitat requirements CSIRO Publishing, Collingwood, VIC. Australia.

Schooler S.S., Cook T., Prichard G. & Yeates A.G. (2010). Disturbance-mediated competition: the interacting roles of inundation regime and mechanical and herbicidal control in determining native and invasive plant abundance. Biological Invasions, 12, 3289-3298.

Semple W.S., Cole I.A. & Koen T.B. (2004). Native couch grasses for revegetating severely salinised sites on the inland slopes of NSW. Rangeland Journal, 26, 88-101.

Semple W.S., Cole I.A., Koen T.B., Costello D. & Stringer D. (2006). Native couch grasses for revegetating severely salinised sites on the inland slopes of NSW. Part 2. Rangeland Journal, 28, 163-173.

Siebentritt M.A., Ganf G.G. & Walker K.F. (2004). Effects of an enhanced flood on riparian plants of the River Murray, South Australia. River Research and Applications, 20, 765-774.

Sluiter I.R.K. (2001). Catalogue of common plant taxa around the lakeshore of Lake Victoria. Ogyris Ecological Research, Merbein.

Stromberg J.C., Richter B.D., Patten D.T. & Wolden L.G. (1993). Response of a Sonoran riparian forest to a 10-year return flood. Great Basin Naturalist, 53, 118-130.

Taylor B. & Ganf G.G. (2005). Comparative ecology of two co-occurring floodplain plants: the native Sporobolus mitchellii and the exotic Phyla canescens. Marine and Freshwater Research, 56, 431-440.

Tenten N., Bo Z. & Kazda M. (2010). Soil stabilizing capability of three plant species growing on the Three Gorges Reservoir riverside. Journal of Earth Science, 21, 888-896.

Walker K.F., Boulton A.J., Thoms M.C. & Sheldon F. (1994). Effects of water-level changes induced by weirs on the distribution of littoral plants along the River Murray, South Australia. Australian Journal of Marine and Freshwater Research, 45, 1421-1438.

59

Appendix C

Data plots for trend analysis of growth forms

0

10

20

30

40

50

60 (a) (b)

0

10

20

30

40

50

60 (c) (d)

24 25 26 27 28 29 300

10

20

30

40

50

60 (e)

1998 2001 2004 2007 2010

(f)

Mean

Cove

r (%)

Elevation Year

60

Figure 29. Plots of changes in cover of Grasses with elevation and time. Error bars are 1.0 SE of the cover scores at that elevation or year.

0

10

20

30

40

50

60 (a) (b)

0

10

20

30

40

50

60 (c) (d)

24 25 26 27 28 29 300

10

20

30

40

50

60 (e)

1998 2001 2004 2007 2010

(f)

Mean

Cove

r (%)

Elevation YearFigure 30. Plots of changes in cover of Herbs with elevation and time. Error bars are 1.0 SE of the cover scores at that elevation or year.

61

0

10

20

30

40

50

60 (a) (b)

0

10

20

30

40

50

60 (c) (d)

24 25 26 27 28 29 300

10

20

30

40

50

60 (e)

1998 2001 2004 2007 2010

(f)

Mean

Cove

r (%)

Elevation YearFigure 31. Plots of changes in cover of Rushes and Sedges with elevation and time. Error bars are 1.0 SE of the cover scores at that elevation or year.

62

0

10

20

30

40

50

60 (a) (b)

0

10

20

30

40

50

60 (c) (d)

24 25 26 27 28 29 300

10

20

30

40

50

60 (e)

1998 2001 2004 2007 2010

(f)

Mean

Cove

r (%)

Elevation YearFigure 32. Plots of changes in cover of Woody species with elevation and time. Error bars are 1.0 SE of the cover scores at that elevation or year.

63

Appendix D

Wind rose plots for Mildura

Jan AprMarFeb

May AugJulJun

Sep DecNovOct

Figure 33. Monthly wind roses for the Mildura weather station (1946 – 2010). Compiled from Bureau of Meteorology wind rose reports.

In addition, Jane Roberts (pers. comm.) checked maximum wind speed graphs to assess whether any months were significantly windier than others. There was a smooth relationship among months for maximum gust velocity, with September – February generally having higher gusts than other months.

64