Exploring factors controlling the variability of pesticide concentrations in the Willamette River Basin using tree-based models

We analyzed available concentration data of five commonly used herbicides a nd three pesticides collected from small streams in the Willamette River Ba sin in Oregon to identify factors that affect the variation of their concen trations in the area. The emphasis of this paper is the innovative use of c lassification and regression tree models for exploratory data analysis as w ell as analyzing data with a substantial amount of left-censored values. Am ong variables included in this analysis, land-use pattern in the watershed is the most important for all bur one (simazine) of the eight pesticides st udied, followed by geographic location, intensity of agriculture activities in the watershed (represented by nutrient concentrations in the stream), a nd the size of the watershed. The significant difference between urban site s and agriculture sites is the variability of stream concentrations. While all 16 nonurban watersheds have significantly higher variation than urban s ites, the same is not necessarily true for the mean concentrations. Seasona l variation accounts for only a small fraction of the total variance in all eight pesticides.