Development of a prediction system for lake stony-bottom littoral macro invertebrate communities

Assessing the biological integrity of aquatic habitats often requires a comparison of observed conditions with those conditions expected to occur in the absence of human-generated stress. Using data from least impaired lakes, models were developed to predict the macroinvertebrate community composition of stony littoral habitats of relatively small (median surface area = 0.26 km(2)) boreal lakes. In brief, three steps were used in model calibration: (i) identification of biological groups by clustering, (ii) selection of environmental variables that explain among-group variance using ordination, and (iii) model calibration using discriminant function analysis. RIVPACS-type predictive models were individually developed for the three major regions of Sweden; namely, the mixed forest, coniferous forest and arctic/alpine regions. Geographic position (latitude, longitude, altitude) and variables indicative of substratum (e.g. cobble and vegetation type) and water color were often found to discriminate among-group variance and were used in model calibrations. The lowest errors associated with observed to expected (O/E) scores for taxon richness were found for models developed for the mixed forest region (mean O/E score = 0.997 +/- 0.254, CV = 25.5%), and the highest errors were associated with models developed for the arctic/alpine region ((mean O/E score = 1.011 +/- 0.337, CV = 33.3%). The use of RIVPACS-type models in ecological assessment is increasing, principally because the site-specific predictions that are produced are intuitive and rather straightforward. This study shows that the RIVPACS approach may be used for assessing and monitoring the ecological quality of boreal lake ecosystems using littoral macroinvertebrate communities.