Estimating Daily Nitrate Loads in Iowa Streams Using a Partial Least Squares Regression Framework

Abstract

Agricultural nitrate pollution is a major threat to water quality in Iowa. Iowa uses a majority of its land for row crop agriculture and maintains a large livestock population, which together cause high nitrate loads in streams. High-frequency stream nitrate data can aid policy decisions for reducing nitrate emissions by identifying streams with high nitrate loads, historical trends of improvement or deterioration in nitrate loads, and land use or practice changes that affect water quality. We developed a time series regression model framework to supplement existing sensor data and predict daily nitrate loads in Iowa streams lacking nitrate monitoring. Using nitrate data from statewide and national resources, this framework was trained and validated using 11 study sites of diverse geography and land use in Iowa. Partial least squares regression (PLSR) was used with geographical predictors, including land use, hydrogeology, and meteorology, to predict streamflow-nitrate load relationships across the study sites. The developed PLSR model, combined with daily streamflow data, was then used to predict daily nitrate loads with high accuracy over a three-year study period with a mean Kling–Gupta Efficiency of 0.74. Our framework was then used to estimate mean nitrate concentrations at 34 sites that lack nitrate sensors, demonstrating a low-cost, facile method for the accurate prediction of daily nitrate loads in Iowa streams.