Code for Peer Review: Ecological Drought

Drought adaptation planning is hindered by an inability to predict risk to riverine ecosystems due to reduced flows. Our goal was to develop an empirical model that predicts ecological drought risk across a large region using existing biological and physical data. An ensemble of machine-learning models was calibrated with monitoring data collected by two federal agencies in streams and rivers within the Cascade, Blue, and Sierra Nevada ecoregions of the western US. Data collected by the state of California in the Sierra Nevada was used for model validation. The occurrence of degraded invertebrate communities was predicted with 71% overall accuracy in the validation dataset. The most influential predictors were streamflow deficits in the months preceding sample collection, and groundwater influence on streamflow. We illustrate a use case for our models by mapping regional predictions of ecological risk during a recent drought.