Paper No. 00139 of the Journal of the American Water Resources Association.Discussions are open until June 1, 2002.
REGRESSION MODELS FOR ESTIMATING HERBICIDE CONCENTRATIONS IN U.S. STREAMS FROM WATERSHED CHARACTERISTICS1
Version of Record online: 8 JUN 2007
JAWRA Journal of the American Water Resources Association
Volume 37, Issue 5, pages 1349–1367, October 2001
How to Cite
Larson, S. J. and Gilliom, R. J. (2001), REGRESSION MODELS FOR ESTIMATING HERBICIDE CONCENTRATIONS IN U.S. STREAMS FROM WATERSHED CHARACTERISTICS. JAWRA Journal of the American Water Resources Association, 37: 1349–1367. doi: 10.1111/j.1752-1688.2001.tb03644.x
- Issue online: 8 JUN 2007
- Version of Record online: 8 JUN 2007
- water quality;
- nonpoint source pollution;
ABSTRACT: Regression models were developed for estimating stream concentrations of the herbicides alachlor, atrazine, cyanazine, metolachior, and trilluralin from use-intensity data and watershed characteristics. Concentrations were determined from samples collected from 45 streams throughout the United States during 1993 to 1995 as part of the U.S. Geological Survey's National Water-Quality Assessment (NAWQA). Separate regression models were developed for each of six percentiles (10th, 25th, 50th, 75th, 90th, 95th) of the annual distribution of stream concentrations and for the annual time-weighted mean concentration. Estimates for the individual percentiles can be combined to provide an estimate of the annual distribution of concentrations for a given stream. Agricultural use of the herbicide in the watershed was a significant predictor in nearly all of the models. Several hydrologic and soil parameters also were useful in explaining the variability in concentrations of herbicides among the streams. Most of the regression models developed for estimation of concentration percentiles and annual mean concentrations accounted for 50 percent to 90 percent of the variability among streams. Predicted concentrations were nearly always within an order of magnitude of the measured concentrations for the model-development streams, and predicted concentration distributions reasonably matched the actual distributions in most cases. Results from application of the models to streams not included in the model development data set are encouraging, but further validation of the regression approach described in this paper is needed.