Field-measured versus derived: What are the most effective predictor variables in stream biodiversity models?

Abstract

Statistical models used to predict and map patterns of biodiversity require environmental variables with full coverage across an area of interest. By necessity, these variables are derived from GIS, remote sensing, or via interpolation, and may not be as physiologically relevant to biota or as representative of on-the-ground conditions as field-measured variables. This research used generalized dissimilarity modeling and occurrence data for freshwater fish and benthic invertebrates in Maryland to examine differences in explanatory power, predictive ability, and management inference yielded by derived and field-measured variables. Across the state and for both taxa, models fit with field-measured variables were superior in explanation and prediction, and nearly always more parsimonious. However, there was little difference between the variable sets in ability to predict management-related indices. Results suggest that field-measured variables are preferred over derived variables overall, but their absence from predictive models may not have a large effect on management inference.