Assessing multiple geospatial modeling techniques of assigning pesticide exposure in the California central valley

Abstract

Sixty residences in the central valley of California were sampled for pesticide concentrations in carpet dust during years 2001-2004 in order to validate predictive metrics of pesticide exposure. Pesticide use metrics were calculated based on pesticide use proximate to residences as reported in the Pesticide Use Reporting (PUR) database maintained by California Department of Pesticide Regulation for years 1998-2004. PUR reports pesticide application information according to the public land survey system section (each nominally one square mile). Metrics were tested by including pesticide applications within 300, 500, 750, 1000, and 1250 meters of residence locations and during time periods of six months, one year, two years, and three years prior to the residence sampling date. Additional pesticide use metrics were calculated using land use maps created by California Department of Water Resources for the area surrounding each residence at the time the residence was sampled to refine the spatial locations of the pesticide applications. These metrics were not able to adequately identify detections of dicofol, iprodione, phosmet, or trifluralin in homes (ROC C-statistics < 70%) and were significant, but not adequate, predictors of pesticide concentrations of chlorpyrifos, iprodione, phosmet, simazine, and trifluralin in homes where the pesticide was detected (p-values < 0.05 for all but one metric, R2 ranged between 4.1% and 29.9%). When these metrics were augmented with soil and meteorological data they improved in their prediction of pesticide detection (ROC C-statistics between 70.3% and 73.8%) and prediction of concentration (R2 between 18.6% and 58.4%). Wind information was found to be the most important geographical variable for improvement. Soil information and questionnaire answers about agricultural employment and home pesticide use were not significant predictors in pesticide detection and concentration models.