Abstract
This study utilized the Random Forest (RF) algorithm to assess groundwater potential (GWP) in the mid-mountain region of the Coquimbo region, north-central Chile. A comprehensive evaluation of twenty-one factors, primarily derived from Digital Elevation Models (DEM) and satellite data, was conducted against a database of 3822 groundwater discharge points. The majority of them consisted of shallow wells with relatively low yields. The main objective was to develop a groundwater potential (GWP) map for the study area. Among the factors considered, six variables, including two anthropogenic factors (distance to roads and presence of agricultural communities) and four natural factors (slope, elevation, concavity, and ruggedness index), were identified as the most influential indicators of GWP. The RF approach demonstrated excellent performance, achieving an Area Under the Curve (AUC) value of 0.95, sensitivity of 0.88, specificity of 0.86, and kappa coefficient of 0.74 in the test set. The majority of the study area exhibited low GWP, while only 14% of the area demonstrated high or very high GWP. In addition to providing valuable guidance for future hydrogeological investigations in the region, the GWP map serves as a valuable tool for identifying the areas that are most vulnerable to water shortages. This is particularly significant, as the region has been severely affected by extended drought, making water supply a critical concern.