Multiple regression analysis of pesticide occurrence in streamflow related to pesticide properties and quantities applied

Monitoring data of 25 pesticides in surface water from a small agricultural catchment in southern Sweden during 1990–1994 were subjected to stepwise multiple linear regression analyses to express concentration, transported amount and loss rate as functions of different pesticide intrinsic properties (or a combination of these) and quantities applied. The single most significant variable was quantities applied, accounting for ca. 50–85% of the variability in concentration and transported amount during individual years. Model performance was slightly improved by adding intrinsic properties of the pesticides as explanatory variables, thus explaining between 70 and 95 percent of the variability of the two dependent variables. Evaluation of loss rate, in percent of applied quantities, using simple linear regression analysis identified log Pow as the most significant variable for pesticide loss to surface water accounting for 26% of the variance when monitoring results from four individual years were pooled. Using the best model equations in a crossvalidation procedure, surface water concentrations, transported amount and loss rates were calculated that compared well to monitoring results for 80–100% of the compounds measured. This study demonstrated that quantities of pesticides used in the catchment area was the most important estimator of level and amount of pesticides occurring in the stream and that log Pow was the most significant intrinsic property estimating relative loss of pesticides from the catchment.