Bayes' Rule and GIS for Evaluating Sensitivity of Groundwater to Contamination

The study deals with the assessment of regional groundwater vulnerability coupling the Weights of Evidence model (Bonham-Carter G.F. et al., 1989; Bates L.E. et al., 1996) with geographical information system. A series of predictor maps, representing the spatial distribution of factors influencing the groundwater vulnerability, are derived for the investigated area and a statistic-probabilistic methodology is adopted and implemented in a GIS for the analysis of the relationships between available data sets. The spatial data analysis for groundwater vulnerability consists of the following steps: ¨ Data collection; ¨ Pre-processing of available data; ¨ Construction of GIS database containing all causal factors (classified themes and other supporting information); ¨ Construction of the prediction model, including the creation of unique condition subareas, computation of probability tables, construction of prediction maps and representation of preliminary prediction results. The model expresses the relative susceptibility of groundwater to contamination in terms of probabilities and combines them by the bayesian combination formulas under the conditional independence assumption. ¨ Validation of prediction results (time robustness and space robustness) by the comparison with spatial distribution of nitrate in different years in the area; ¨ Visualization of prediction results. Potential problems related to conditional independence of the predictor patterns with respect to the occurrence of the response event are also considered.