Field‐measured variables outperform derived alternatives in Maryland stream biodiversity models

Abstract

AbstractAimIn order to map patterns of biodiversity in support of conservation efforts, statistical models require environmental variables with full coverage across the study area, typically in the form of gridded surfaces derived from GIS, remote sensing or via interpolation. However, derived variables may not be as physiologically relevant or as representative of on‐the‐ground conditions as field‐measured variables. Here, we examine differences in the abilities of derived and field‐measured variables to explain and predict biogeographical patterns of freshwater fish and benthic invertebrate communities.LocationMaryland first‐ through fourth‐order streams, USA.MethodsWe fit generalized dissimilarity models to fish and benthic invertebrate occurrence data (n = 2,165 site‐years sampled over 18 years) using one set of field‐measured predictors collected concurrent and collocated with faunal sampling, two sets of derived predictors (one representing local and one representing upstream conditions) or a combination. We then compared how well models explained and predicted spatial turnover in taxonomic composition (beta diversity).ResultsFor all regions (four physiographic regions and the state as a whole) and for both fish and benthic invertebrates, models fit with field‐measured variables were more explanatory and usually more predictive than models fit with derived variables. Within the category of derived predictors, those accounting for upstream conditions were more explanatory and predictive than local‐scale versions.Main conclusionsAlthough derived variables are most commonly used to describe and map biodiversity, they may be broadly inferior to field‐measured variables as predictors in low‐order stream biodiversity models. Collection of field‐measured data and development of derived data that consider upstream conditions and capture physiologically relevant environmental characteristics are likely to improve our capability to predict and explain spatial patterns of low‐order stream biodiversity.