Author(s): N. Obregon; Fragalaf; O. Garcia

Linked Author(s):

Keywords: Quifer; Vulnerability; Hydroinformatics; Decision trees; Machine learning; SINTACS

Abstract: Vulnerability to pollution is important to groundwater resources management. However, its evaluation is problematic because it depends on several parameters which estimation is difficult. During the last thirty years several qualitative and parametric models have been developed. Among last ones SINTACS methodology is one of the most popular (Civita, 1994; Civita & De Maio, 1997) employing 7 parameters (depth to groundwater, recharge, rock media in the unsaturated zone, soil textures, aquifer material, aquifer hydraulic conductivity and slope ranges), some of them hard to estimate. This work shows the advances in the simplification of SINTACS method via implementation of a decision tree (e.g. Mitchell, 1997). Implemented methodology involved pseudo-random generation of sampling parameters values, which are evaluated following SINTACS approach to determine vulnerability in each case. These input patterns (parameters values, weights and vulnerability degree) are used to train the decision tree. The results suggest that information entropy concept, underlying decision trees, allows to encapsulate the relative importance of factors that affect aquifer vulnerability, besides the inherent simplification of using hydroinformatics tools, such as decision trees.

DOI:

Year: 2003