A model-based assessment of anthropogenic disturbance on lotic macroinvertebrate assemblages.

Traditionally, the effects of anthropogenic disturbance on biological assemblages are elucidated by comparing an assemblage observed at a site to one that represents a minimally disturbed state. Unfortunately, defining a minimally disturbed state is extremely challenging because of the extent of human disturbance. We use a national scale dataset and a model-based approach to assess how benthic macroinvertebrate assemblages at 1748 sites would change if common anthropogenic disturbances were removed from in-stream physiochemical variables. First, we used random forest models and current landscape data to predict physiochemical conditions and then infer abiotic condition in the absence of disturbance. Second, we combined these estimates with joint species distribution models to predict the assemblage that could occur in these undisturbed conditions. Random forest models explained 48%-75% of the variation in log-transformed total nitrogen, phosphorus, sulfate, chloride, and substrate diameter. Generally, nutrient and salinity concentrations were higher, and substrates were finer than predicted to be without disturbances. Using these physiochemical data, joint species distribution models, fitted to nine separate ecoregions, accurately explained genus richness (