Aquifer nitrate vulnerability assessment using positive and negative Weights of Evidence methods, Milan, Italy

Statistical methods are extensively used by hydrogeologists for assessing groundwater vulnerability. Several of these methods require to express the response variable as binary and to select a threshold distinguishing between positive and negative indicators of contamination that are usually identified as occurrences and non-occurrences, respectively. In this study, both occurrences and non-occurrences were alternately used as training points (TPs) in the Weights of Evidence (WofE) for assessing groundwater vulnerability to nitrate contamination of a shallow, unconfined, porous aquifer. This was done to better understand the individual role and the combined effect of explanatory variables in both protecting and exposing groundwater from and to nitrate contamination in the study area. The idea behind this approach is that, for a given aquifer, each explanatory variable should have an unequivocal effect on the physical process of groundwater contamination. As part of this study, a procedure for multi-class generalization was developed. Results showed that an evidential theme, even if it appears to be a statistically significant predictor of occurrences, can show an equivocal spatial relationship with the positive and the negative indicators of contamination due to the presence of a sampling bias between the TPs and the evidential theme. It was demonstrated that, if sampling bias is not recognized and corrected, the use of such evidential theme in the analysis could lead to obtain unreliable groundwater vulnerability maps. In order to deal with this issue, a quantitative methodology to correct the effects of sampling bias was successfully tested. Indeed, once the spatial relationships between the different type of TPs and the considered evidential themes were corrected for the effects of sampling bias, the WofE method was found to be a reliable modeling technique for assessing groundwater vulnerability and proved to be capable of identifying areas characterized by different degrees of vulnerability.