Automated Probabilistic Spatial Co-Occurrence Assessments for Aquatic Endangered Species.

Abstract

The US Environmental Protection Agency must evaluate potential impacts on federally listed threatened and endangered species during the course of pesticide registration. However, current deterministic methods for analyzing geospatial co-occurrence between listed species and pesticide applications do not account for spatial and temporal variability. To address this challenge, we developed the Automated Probabilistic Co-Occurrence Assessment Tool (APCOAT). Using APCOAT, we modeled potential co-occurrence between atrazine applied to corn and aquatic habitats across the continental U.S. by developing habitat models for 375 species in flowing waters and 130 species in static waters. The species habitat models showed high predictive power (70-99% accuracy, median 98%) while maintaining parsimony (median 9 environmental variables). Analysis of both local watershed and upstream pesticide transport revealed that 70% of habitat-pesticide combinations had <5% co-occurrence probability, 25% showed 5-10%, and 5% exceeded 10%. The probabilistic approach provides more refined estimates of both species habitat extent and pesticide usage patterns compared to deterministic methods. These spatially explicit models of species distributions and pesticide application patterns provide valuable tools individually, while their combination enables nuanced probabilistic co-occurrence assessment. The methods and results demonstrate how incorporating probability and uncertainty can improve both species conservation planning and regulatory decision-making.