Geographic extension of benthic Invertebrate RCA bioassessments: how far can we go?

Abstract

The management of aquatic ecosystems is important to preserve the ecosystem services provided to humanity. The development of environmental assessment has allowed the management and therefore protection of these important resources. Reference Condition Approach (RCA) bioassessments using benthic macroinvertebrates as indicators are common tools that provide a means of determining impairment of stream sites by comparing exposed test sites to relatively unexposed reference sites. RCA predictive models are commonly developed at the scale of drainage basin, ecoregion, or political region (i.e. United Kingdom or Australian state), and test site assessment is restricted within the spatial boundaries of the model. If test site assessment can be applied outside the spatial scope of the model, insofar that the environmental characteristics are similar, it would reduce extensive sampling i.e. remote northern locations and time-consuming development of numerous models. The overall goal of my study was to assess whether a predictive model applied across a larger spatial extent, and therefore encompassing a greater area for test sites to be assessed, is as effective as models generally developed within smaller geographic regions such as within a basin or watershed. Benthic invertebrates and habitat data from three areas in Canada were examined: the Attawapiskat River basin in northern Ontario, the Fraser River basin in British Columbia and the Yukon River basin. The RCA predictive model method was used in this study that determines the relationships between benthic community groups and the environmental descriptors that explain them and the Benthic Assessment of Sediment (BEAST) assessment method to compare test sites with a physical similar group of reference sites. The performance of the bioassessment was assessed using a common set of simulated impact (“simpacted”) sites with known responses iv of taxa to disturbance. Models for each basin and a multi-basin model were compared on prediction performance, parsimony, and sensitivity. The multi-basin model had comparable prediction performance (65% correctly classified) to single basin models (56-72%) but lacked the sensitivity that models for single basins possessed. The Attawapiskat was the most parsimonious with only 2 predictors but the Fraser and multi-basin models explained the most variance with more predictors (Wilks’ λ = 0.06 and 0.1 for the Fraser and multi-basin models, respectively). The results of this study showed that sites can be assessed outside the range of their reference data insofar that the test site is within the range of environmental characteristics within the model. A test site assessed as disturbed for the multi-basin model will in fact be disturbed but disturbed sites are less frequently detected compared to single basin models. Therefore as with any bioassessment, users need to be aware of the chance of committing type 1 and 2 errors. Developing models that target a single stressor of concern to increase the pool of available candidate predictors is recommended; such a model may possess greater prediction performance and sensitivity.