Use of Logistic Regression and GIS Modeling to Predict Groundwater Vulnerability to Pesticides

Abstract

Soils are considered important components of many pesticide contamination models and are frequently the direct or indirect targets of pesticides applied during agricultural activities. Soil texture is commonly referenced on pesticide labels as an important factor in the selection and application of pesticides and in identifying target areas that are vulnerable to leaching. In general, no guidelines exist for the common interpretation of generic soil texture terms found on pesticide labels, for example, coarse or coarse‐textured soils. In the present study, a significant logistic regression model (P = 0.017) was developed that is based on the soil particle‐size class composition of sections containing wells sampled for DBCP (1,3‐dibromochloropropane). Particle‐size class is a concept used in Soil Taxonomy to describe soil family texture and is a component of soil family names. The model contains terms for the sandy and fine particle‐size classes. The model was validated using data obtained from sources independent of those used to develop the model. Records in the California Soils Map Unit Inventory database that describes the soil map unit composition of >65 000 sections (1600 m2 or 1 mi2) were used to generate probability scores for >15 000 sections located in the San Joaquin Valley, CA. A geographic information system, an information management technique that is becoming an accepted tool for a wide range of regulatory agencies, was used to generate visual images of the probability scores. A map was developed that depicts four distinct probability classes of the DBCP‐contamination status of section‐sized areas and their distribution within the study area.

Groundwater Risk Assessment Model (GRAM): Groundwater Risk Assessment Model for Wellfield Protection

“…Such an approach requires modeling of pollutant transport and fate. Similarly, statistical methods based on the concepts of uncertainty have been developed [20][21][22]. Both methods require an extensive data base, including monitoring and actual measurements of contaminant concentrations to calibrate and validate the models.…”

Probability of Nitrate Contamination of Recently Recharged Groundwaters in the Conterminous United States

“…Logistic regression is well-suited to analysis of nondetects because a threshold value is specified to define the response categories and has been successfully applied in prior studies on the risk of groundwater contamination (13)(14)(15)(16)(17)(18)(19)(20)(21)(22). Readers may consult Hosmer and Lemeshow (23) and Kleinbaum (24) for a detailed discussion of logistic regression.…”

Direct assessment of groundwater vulnerability from single observations of multiple contaminants

“…These approaches represent a simplification of the concentration data as logistic regression predicts the probability of the presence or absence of a contaminant and does not use or predict either the frequency of detection or the concentration of the compound. Teso et al [1996] used a similar approach to link soil particle classes to the risk of groundwater pollution from 1, 3-dibromochloropropane (DBCP). Kolpin [1997] used Spearman's rank correlation coefficient to examine the relationship between nitrate and pesticide concentrations found in monitoring wells and their surrounding catchment.…”