Delineation of groundwater potential zones by means of ensemble tree supervised classification methods in the Eastern Lake Chad basin

This paper presents a machine learning method to map groundwater potential in crystalline domains. First, spatially-distributed set of explanatory variables for occurrence is compiled into geographic information system. Twenty classifiers are subsequently trained on sample 488 boreholes and excavated wells region eastern Chad. process includes collinearity, cross-validation, feature elimination parameter fitting routines. Random forest extra trees outperformed other algorithms (test score > 0.80, balanced AUC 0.87). Fracture density, slope, SAR coherence (interferometric correlation), topographic wetness index, basement depth, distance channels slope aspect proved the most relevant variables. Three major conclusions stem from this work: (1) using large number supervised classification advisable studies; (2) choice performance metrics constrains relevance variables; (3) seasonal variations satellite images contribute successful mapping.