Explainable Artificial Intelligence to visualize the unseen

Groundwater is unseen yet provides half of the world’s freshwater supply. The world's population is currently facing freshwater supply shortages, necessitating precise measurement of groundwater resources. However, because groundwater is found beneath our feet in aquifers, it is expensive if not impossible to accurately estimate. We have developed models to estimate groundwater potential using two explainable artificial intelligence techniques: Explainable Boosting Machine (EBM) and Generalized Additive Model (GAMI-net). We predict groundwater potential using a comprehensive spring location database and environmental conditions (such as elevation, slope, aspect, curvature, precipitation, etc.). We tested the models in western Nepal, where natural springs are particularly important and the majority of people rely on gravity-based natural spring systems. Both EMB and GAMI-net predicted groundwater potential well, with validation scores of 0.89 and 0.9, respectively. We find that elevation, slope, total precipitation, and their combinations have a significant influence on groundwater potential. We highlight a few cases and recommend management strategies to improve water availability in western Nepal.