Regression model for aquifer vulnerability assessment of nitrate pollution in the Osona region (NE Spain)

Abstract

A multiple linear regression (MLR) model was developed for the Osona region (1260km2, NE Spain) to determine aquifer vulnerability to nitrate pollution in groundwater. The MLR model was based on a data set of nitrate data from 57 private wells sampled in 2010. Up to 70 explanatory variables were screened to statistically identify the key parameters that determine nitrate concentration in groundwater in the study area. Finally, 5 variables were selected for the regression model, which are the following: (1) nitrogen loading as manure; (2) aquifer type; (3) percent of well-drained and deep soils; (4) percent of irrigated crops; and (5) an indicator of the presence of denitrification processes, which together represent the input of nitrogen, transport and attenuation of nitrate in groundwater. The model features three innovations: net nitrogen input was determined throughout the region using a simple mass balance approach and a newly available GIS tool; the model addresses the layering of sedimentary formations distinguishing different aquifer types; and detailed aquifer chemistry at well locations was incorporated into the spatial model by reducing data to composite variables for subsequent interpolation by kriging to represent denitrification processes occurring in the aquifer. All variables were significant in the MLR model (p-values<0.10) and explained 75% of the variation in groundwater nitrate concentrations. A plot of observed vs. predicted nitrate for 25 validation wells yielded a significant (p-value<0.001) and a positive correlation. According to the results obtained in the regression model, vulnerability maps for unconfined, leaky and confined aquifers were developed