National Water-Quality Assessment (NAWQA) Program
By Bernard T. Nolan, Kerie J. Hitt, and Barbara C. Ruddy[Environmental Science and Technology, v. 36, no. 10, p. 2138-2145.]
This article supersedes ES&T vol. 31, no. 8; WCP vol. 39, no. 12; and FS-092-96.
A new logistic regression (LR) model was used to predict the probability of nitrate contamination exceeding 4 mg/L in predominantly shallow, recently recharged ground waters of the United States. The new model contains variables representing (1) N fertilizer loading (p < 0.001) , (2) percent cropland-pasture (p < 0.001), (3) natural log of human population density (p < 0.001), (4) percent well-drained soils (p < 0.001), (5) depth to the seasonally high water table (p <0.001), and (6) presence or absence of unconsolidated sand and gravel aquifers (p = 0.002). Observed and average predicted probabilities associated with deciles of risk are well correlated (r2 = 0.875), indicating that the LR model fits the data well. The likelihood of nitrate contamination is greater in areas with high N loading and well-drained surficial soils over unconsolidated sand and gravels. The LR model correctly predicted the status of nitrate contamination in 75% of wells in a validation data set. Considering all wells used in both calibration and validation, observed median nitrate concentration increased from 0.24 to 8.30 mg/L as the mapped probability of nitrate exceeding 4 mg/L increased from less than or equal to 0.17 to > 0.83.
Table of Contents
Results and discussion
Recalibration of multivariate logistic regression model
Groundwater nitrate in specific areas follows predicted probabilities
Logistic regression model validation
General verification of logistic regression model
Uses and limitations