Comparison between Model-Predicted and Field-Measured Stream Habitat Features for Evaluating Fish Assemblage-Habitat Relationships

Abstract

The use of model-predicted, local-scale habitat data as inputs in analyses intended to evaluate multiscale fish assemblage-habitat relationships in streams has become increasingly common as the scale at which such studies are conducted has increased. We used fish assemblage and habitat data from 208 wadeable streams in Wisconsin and Michigan to determine whether model-predicted habitat data would yield results similar to those of field-measured data in multiscale analyses of fish assemblage-habitat relationships. Predictions of local habitat features from landscape variables were generated by means of generalized additive modeling with likelihood-based boosting. Relationships between fish assemblage measures and landscape and local habitat features were studied via partial constrained multivariate ordination analyses. The total variation explained in the fish assemblage data sets was similar for model-predicted local habitat features and field-measured data, as was the proportion of variation explained that was due independently to local and regional (i.e., landscape) habitat features. We observed dissimilar results in the magnitude of ordination scores for local habitat features and the directional relationships between local habitat ordination scores and individual species and assemblage metric scores. Our findings indicate that model-predicted, local-scale habitat data can be useful for evaluating the relative strengths of local and regional habitat features in structuring fish assemblages, but caution may be necessary when evaluating species-habitat or assemblage metric-habitat relationships.