Modeling groundwater nitrate concentrations in private wells in Iowa.

摘要: Contamination of drinking water by nitrate is a growing problem in many agricultural areas the country. Ingested can lead to endogenous formation N-nitroso compounds, potent carcinogens. We developed predictive model for concentrations private wells Iowa. Using 34,084 measurements wells, we trained and tested random forest models predict log levels systematically assessing performance 179 variables 36 thematic groups (well depth, distance sinkholes, location, land use, soil characteristics, nitrogen inputs, meteorology, other factors). The final contained 66 17 groups. Some most important were well slope length within 1 km well, year sample, nearest animal feeding operation. correlation between observed estimated was excellent training set (r-square=0.77) acceptable testing (r-square=0.38). had substantially better than traditional linear regression or tree. Our will be used investigate association cancer risk Iowa participants Agricultural Health Study cohort.