Assessing groundwater potential in the guder watershed, Ethiopia, using geospatial technologies, Google Earth Engine, and machine learning

2026-01-11

Guta Tolossa Werati, Abera Gonfa Abdissa,
Assessing groundwater potential in the guder watershed, Ethiopia, using geospatial technologies, Google Earth Engine, and machine learning,
Environmental Challenges,
Volume 21,
2025,
101325,
ISSN 2667-0100,
https://doi.org/10.1016/j.envc.2025.101325.
(https://www.sciencedirect.com/science/article/pii/S2667010025002446)
Abstract: Groundwater is vital for river flows, wetlands, and ecosystem health, particularly during dry seasons and droughts, yet increasing population pressures threaten its availability. This study aimed to map groundwater potential zones in the Guder watershed, Ethiopia, using an integrated approach combining geospatial analysis, Google Earth Engine (GEE), and machine learning. Key influencing factors, including land use/land cover (LULC), slope, drainage density, lineament density, topographic wetness index (TWI), rainfall, soil texture, geology, and geomorphology, were collected from multiple sources. LULC was classified using GEE, while DEM-derived variables were generated in ArcGIS Pro. Rainfall data were interpolated using Inverse Distance Weighting, and all datasets were normalized and integrated into two classification approaches: Analytical Hierarchy Process (AHP) with GIS overlay and Random Forest Classifier (The AHP-GIS model classified zones into very high (17.8 %), high (59.3 %), moderate (22 %), low (0.8 %), and very low (0.1 %) categories. The RFC model produced very high (20.8 %), high (57.1 %), moderate (21 %), low (0.9 %), and very low (0.2 %) zones. The RFC achieved high sensitivity and low false positive rates (0.90, 0.89, and 0.84). Validation included AHP consistency checks, spatial comparison with observed wells, and ROC-based performance metrics, confirming the reliability of the models. These findings demonstrate that integrating machine learning with geospatial tools enhances the accuracy and resolution of groundwater mapping. The approach identifies additional high-potential zones compared to traditional methods, providing a robust, scalable framework for sustainable water resource planning in data-scarce regions.
Keywords: Guder watershed; Groundwater potential zones; Geospatial technologies; Google Earth Engine, machine learning