Application of deep neural networks to groundwater potential mapping in the region of Sidi Bel Abbès, Algeria

Groundwater potential mapping becomes key to several fields due to water shortages and drought impacts. During this work, we apply deep neural networks to delineate the potential zones in the plain of Sidi Bel Abbès. The methods aim at producing a reliable estimation of the potential zones based on the groundwater static level. The model is constructed as a compromise between simplicity as well as efficiency. As model features, we employ seven thematic maps which are: digital elevation model, slope, drainage, lineaments, land use, rainfall and lithology. Results show that the proposed method achieves better performance compared to several machine learning methods (RF, SVM and Neural Networks) in terms of accuracy (84.94%) and κ (0.80) coefficient with excellent AUC (0.80). The predicted maps can be used to improve the management of groundwater resources in the region.