Insights from an Evaluation of Nitrate Load Estimation Methods in the Midwestern United States

Abstract

This study investigated the accuracy and suitability of several methods commonly used to estimate riverine nitrate loads at eight watersheds located southwest of Lake Erie in the Midwestern United States. This study applied various regression methods, including a regression estimator with five, six, and seven parameters, an estimator enhanced by composite, triangular, and rectangular error corrections with residual and proportional adjustment methods, the weighted regressions on time, discharge, and season (WRTDS) method, and a simple linear interpolation (SLI) method. Daily discharge and nitrate concentration data were collected by the National Center for Water Quality Research. The methods were compared with subsampling frequencies of 6, 12, and 24 times per year for daily concentrations, daily loads, and annual loads. The results indicate that combinations of the seven-parameter regression method with composite residual and rectangular residual adjustments provided the best estimates under most of the watershed and sampling frequency conditions. On average, WRTDS was more accurate than the regression models alone, but less accurate than those models enhanced by residual adjustments, except for the most urbanized watershed, Cuyahoga. SLI was the most accurate in the Vermilion and Maumee watersheds. The results also provide some information about the effects of rating curve shape and slope, land use, and record length on model performance.