Toward river health assessment using species distributional modeling
Introduction
Effective and sustainable management of water resources for conservation purposes requires adequate assessment of the condition of a water resource, followed by an evaluation of the response of the water resource to intervention programs (ANZECC and ARMCANZ, 2000). The most commonly used approach to river health assessment is to compare the condition of a test site with that of a range of “reference” sites in order to estimate the biota that could be expected to occur at the test site in its natural state. The condition of the test site is evaluated by determining the degree of dissimilarity to the reference sites. This approach has been called the reference condition approach (Reynoldson et al., 1997, Bailey et al., 2004). However, the approach has a number of limitations. For example, undisturbed sites similar to the test site may not exist (e.g. there are few large, undisturbed rivers). To compensate for this, some authors select reference sites that are “minimally disturbed” or “least affected” by human disturbance (Bailey et al., 2004), but this can lead to a subjective and inconsistent definition of expected condition. Another possible limitation to the reference condition approach is that the number of reference sites available to represent the range of habitat types and environmental conditions likely to be encountered at tests sites, and to describe natural variability among reference sites, may be small. Finally, if a large number of adequate reference sites do exist, the cost of accessing these, sampling them and then processing the data collected would be a major investment of resources.
There are some alternative approaches to establishing reference condition if references sites do not physically exist. The first is to use expert opinion to define the undisturbed nature of a site using biotic metrics such as the Index of Biotic Integrity (Karr, 1981, Fausch et al., 1984, Harris and Silveira, 1999). Expert opinion has been used in Australia to define the expected fish assemblages in river reaches of the Murray–Darling Basin for the Sustainable Rivers Audit, and is termed the Reference Condition for Fish (Davies et al., 2010). While the use of expert opinion can be appealing in natural resource management as it is seen as a cost-effective way of making predictions, it is by its nature subjective and can potentially introduce biases or errors into river health assessments (Lele and Allen, 2006). Another approach to predict the reference state of fish assemblages is through the use of environmental “filters”, a concept introduced by Tonn et al. (1990). The concept represents the processes (filters) that sequentially reduce the regional species pool to those that occur locally (Angermeier and Winston, 1998). The use of the environmental filters approach involves a comparison between the expected potential taxa derived from applying environmental filters, and the taxa actually collected at a test site, to provide an assessment of disturbance based on decreased taxonomic richness. The environmental filters approach for river health assessment was piloted by Chessman and Royal (2004) for aquatic macroinvertebrates and Chessman (2006) for freshwater fish. However, this approach is yet to be widely adopted and requires further refinement in the areas of accumulation of information on the distributions of aquatic taxa to enable specification of taxon traits and environmental tolerances with greater detail and accuracy.
Species distributional modeling (SDM) is a more recently developed alternative to the reference site approach. Species distributional modeling can be defined as any statistical/analytical algorithm that predicts the potential distribution of a species, given field observations (e.g. museum records) and environmental predictors (Hengl et al., 2009). Species distributional modeling has emerged as a tool for exploring niche theory and for producing geographical distribution maps for management purposes. One application of species distributional modeling is to evaluate environmental variables that may influence the distribution of the study species. These can then be used to map the species’ expected distribution in a disturbed area (such as a regulated river) to determine the impact of the disturbance on the species’ actual distribution. Dams and weirs are an example of an environmental disturbance that can have a substantial affect on fish communities through modified flow regimes, coldwater pollution and reduced access to spawning grounds (Beasley and Hightower, 2000, Sherman et al., 2007, Flowers et al., 2009). Fish species diversity in regulated coastal rivers is often lower upstream of major dams and weirs than downstream reaches, largely due to the absence of diadromous species that were otherwise known or expected to occur (Gehrke et al., 1999, Gehrke et al., 2002, Rolls, 2011).
The Hawkesbury–Nepean River system of New South Wales (NSW) central coast is one of the most highly modified river systems in Australia (Harris, 1984a). A total of 147 barriers to fish passage have been identified in the Hawkesbury–Nepean river system including water supply dams up to 94 m in height that provide the city of Sydney with 97% of its drinking water and a series of low weirs less than six meters high used mainly for water supply to smaller towns and irrigation (Thorncraft and Harris, 2000). The large dams completely obstruct upstream fish passage. The smaller weirs downstream of the large dams, however, are occasionally inundated during periods of high flows, potentially allowing intermittent fish passage. These barriers were estimated to obstruct fish movement to approximately 50% of the total catchment area (Harris, 1984b). Recent studies have shown that fish species diversity declines from the lower reaches to the upper reaches, despite some species being expected to be able to utilize habitat further upstream (Gehrke et al., 1999, Baumgartner and Reynoldson, 2007). Therefore, despite the occasional inundation of the weirs, not all migratory species are successfully traveling upstream to the full extent of their natural range. This may be because the higher flows occur outside their peak migration period, or because water velocities passing over the weirs exceed their maximum swimming ability.
In this study, we use SDM to determine the species composition at reference sites using the highly regulated Hawkesbury–Nepean River system as our study system. First, we predict the natural distribution of a range of fish species prior to the installation of dams and weirs. We then compare the predicted distribution to the present documented distribution within the Hawkesbury–Nepean River system to determine the observed to expected species ratio and thus gain an understanding of how species are affected by dams and weirs. We predict that diadromous species will be absent upstream of major water supply dams, resulting in a low observed to expected fish species ratio at these sites. We also predict that the observed to expected fish species ratio will be similar in unregulated sites downstream of barriers, but the ratio will decline upstream through a series of low weirs.
Section snippets
Study area
The study area encompasses seven coastal catchments with a total area of approximately 81,000 km2 (Fig. 1a). The rivers originate in the Great Dividing Range at altitudes of up to 1200 m Six model training catchments on the mid north coast of NSW were used to generate SDMs: Bellinger, Macleay, Hastings, Manning, Port Stephens/Wallis Lake and Hunter (Fig. 1b). These catchments were chosen as they lie close to the test catchment, the Hawkesbury–Nepean (Fig. 1b) and are predominately unregulated.
Results
Geographic predictive maps were produced for 21 fish species (Table 1). A total of 2904 fish records from 584 unique locations were used for model predictions, although the number of sites varied between species. Model accuracy for the training catchments was high with AUC greater than 0.80 for the majority species, indicating reliable fitting of the training models. Catchment area and elevation were the two most important environmental variables contributing to species distributions, with
Discussion
The results of this study corroborate our prediction that O/E fish species ratios would decrease longitudinally along the disturbance gradient, clearly demonstrating the usefulness of the distributional model approach to riverine health assessment in situations, such as the Hawkesbury–Nepean, where reference sites do not exist. It also indicates that weirs affect the fish assemblages. In addition, the O/E ratio was extremely low above major dams (∼0.45), suggesting several migratory species may
Conclusion
In summary, the primary benefit of the SDM approach to river health assessment is that the natural fish assemblage can be described for any given site. The method uses existing data so no additional resources are required to develop this approach for other biotic groups and geographical areas. However, further research is required into the ability of this approach to detect other environmental disturbances; to assess the benefits of the exclusion of rare species from the analysis; and to
Acknowledgements
The authors wish to thank Bob Creese and Julie Bindokas for their valuable comments on the manuscript.
References (51)
- et al.
Spatial prediction of species’ distributions from occurrence-only records: combining point pattern analysis, ENFA and regression-kriging
Ecol. Mod.
(2009) - et al.
Maximum entropy modelling of species geographic distributions
Ecol. Model.
(2006) The role of life-history and location of barriers to migration in the spatial distribution and conservation of fish assemblages in a coastal river system
Biol. Conserv.
(2011)Australian and New Zealand Water Quality Guidelines
(2000)- et al.
Local vs. regional influences on local diversity in stream fish communities of Virginia
Ecology
(1998) - et al.
Bioasseessment of Freshwater Ecosystems Using the Reference Condition Approach
(2004) - et al.
Fish communities of the Nepean River in the vicinity of Pheasants Nest Weir
(2007) - et al.
Effects of a low-head dam on the distribution and characteristics of spawning habitat used by striped Bass and American Shad
Trans. Am. Fish. Soc.
(2000) - et al.
Rare species are important for bioassessment–reply to Marchant's comments
Limnol. Oceanogr.
(1999) - et al.
How important are rare species in aquatic community ecology and bioassessment?
Limnol. Oceanogr.
(1998)