Predicting atrazine concentrations in waterbodies across the contiguous United States: The importance of land use, hydrology, and water physicochemistry

Abstract

Atrazine contamination is ubiquitous in North American surface waters, but the dependency of the herbicide's distribution on landscape and within-lake processes is currently poorly known. We sought to identify novel predictors of atrazine and to build a coherent framework to model its concentration in waterbodies through the development of binomial-gamma hurdle models and LASSO regression models. We constructed models for over 900 waterbodies in the contiguous United States using data from the 2012 U.S. EPA National Lake Assessment, the 2012 U.S. Department of Agriculture CropScape and the Global HydroLAB HydroLAKES databases. Atrazine was detected in 32% of U.S. waterbodies, with a mean concentration of 0.17 μg L−1 when detected. The two-part hurdle model explained as much as 75% of the variance in atrazine across a spatially and temporally heterogeneous landscape. Three predictors explained 31% of the variability in atrazine detection in U.S. waterbodies, where the proportion of corn + soy cultures in the watershed was the most important variable. Once atrazine was detected, our models explained an additional 29% of the variability in atrazine concentrations, where the estimated areal weight of atrazine application (kg atrazine km2) in the watershed was the most important predictor. Spatially, water quality variables associated with eutrophication were linked to increased levels of atrazine contamination while cooler water temperatures and natural lakes and landscapes were associated with decreased levels of contamination. Our results suggest that changes in land-use practices may be the most effective way to mitigate atrazine contamination in waterbodies.