A new index of habitat alteration and a comparison of approaches to predict stream habitat conditions

Abstract

1. Stream habitat quality assessment complements biological assessment by providing a mechanism for ruling out habitat degradation as a potential stressor and provides reference targets for the physical aspects of stream restoration projects. This study analysed five approaches for predicting habitat conditions based on discriminant function, linear regressions, ordination and nearest neighbour analyses. 2. Quantitative physical and chemical habitat and riparian conditions in minimally-impacted streams in New Hampshire were estimated using United States Environmental Protection Agency's Environmental Monitoring and Assessment Program protocols. Catchment-scale descriptors were used to predict segment-scale stream channel and riparian habitat, and the accuracy and precision of the different modelling approaches were compared. 3. A new assessment index comparing and summarizing the degree of correspondence between predicted and observed habitat based on Euclidean distance between the standardized habitat factors is described. Higher index scores (i.e. greater Euclidean distance) would suggest a greater deviation in habitat between observed conditions and expected reference conditions. As in most biotic indices, the range in index scores in reference sites would constitute a situation equivalent to reference conditions. This new index avoids the erroneous prediction of multiple, mutually exclusive habitat conditions that have confounded previous habitat assessment approaches. 4. Separate linear regression models for each habitat descriptor yielded the most accurate and precise prediction of reference conditions, with a coefficient of variation (CV) between predictions and observations for all reference sites of 0.269. However, for a unified implementation in regions where a classification-based approach has already been taken for biological assessment, a discriminant analysis approach, that predicted membership in biotic communities and compared the mean habitat features in the biotic communities with the observed habitat features, was similar in prediction accuracy and precision (CV = 0.293). 5. The best model had an error of 27% of the mean index value for the reference sites, indicating substantial room for improvement. Additional catchment characteristics not readily available for this analysis, such as average rainfall or winter snow-pack, surficial geological characteristics or past land-use history, may improve the precision of the predicted habitat features in the reference streams. Land-use history in New Hampshire and regional environmental impacts have greatly impacted stream habitat conditions even in streams considered minimally-impacted today; thus as regional environmental impacts change and riparian forests mature, reference habitat conditions should be re-evaluated. © 2007 The Authors.