Simulation of daily pesticide concentrations from watershed characteristics and monthly climatic data

Abstract

A time-series model was developed to simulate daily pesticide concentrations for streams in the coterminous United States. The model was based on readily available information on pesticide use, climatic variability, and watershed charac-teristics and was used to simulate concentrations for four herbicides [atrazine, ethyldipropylthiocarbamate (EPTC), metolachlor, and trifluralin] and three insecticides (carbofuran, ethoprop, and fonofos) that represent a range of physical and chemical properties, application methods, national application amounts, and areas of use in the United States. The time-series model approximates the probability distributions, seasonal variability, and serial correlation characteristics in daily pesticide concentration data from a national network of monitoring stations. The probability distribution of concentrations for a particular pesticide and station was estimated using the Watershed Regressions for Pesticides (WARP) model. The WARP model, which was developed in previous studies to estimate the probability distribution, was based on selected nationally available watershed-characteristics data, such as pesticide use and soil characteristics. Normality transformations were used to ensure that the annual percentiles for the simulated concentrations agree closely with the percentiles estimated from the WARP model. Seasonal variability in the transformed concentrations was maintained by relating the transformed concentration to precipitation and temperature data from the United States Historical Climatology Network. The monthly precipitation and temperature values were estimated for the centroids of each watershed. Highly significant relations existed between the transformed concentrations, concurrent monthly precipitation, and concurrent and lagged monthly temperature. The relations were consistent among the different pesticides and indicated the transformed concentrations generally increased as precipitation increased but the rate of increase depended on a temperature-dependent growing-season effect. Residual variability of the transformed concentrations, after removal of the effects of precipitation and temperature, was partitioned into a signal (systematic variability that is related from one day to the next) and noise (random variability that is not related from one day to the next). Variograms were used to evaluate measurement error, seasonal variability, and serial correlation of the historical data. The variogram analysis indicated substantial noise resulted, at least in part, from measurement errors (the differences between the actual concen-trations and the laboratory concentrations). The variogram analysis also indicated the presence of a strongly correlated signal, with an exponentially decaying serial correlation function and a correlation time scale (the time required for the correlation to decay to e-1 equals 0.37) that ranged from about 18 to 66 days, depending on the pesticide type. Simulated daily pesticide concentrations from the time-series model indicated the simulated concentrations for the stations located in the northeastern quadrant of the United States where most of the monitoring stations are located generally were in good agreement with the data. The model neither consistently overestimated or underestimated concentrations for streams that are located in this quadrant and the magnitude and timing of high or low concentrations generally coincided reasonably well with the data. However, further data collection and model development may be necessary to determine whether the model should be used for areas for which few historical data are available.