Modeling Groundwater Potential Under Climate Change Scenarios in Eastern Rwanda Using the MaxEnt and RF Machine Learning Algorithms

Groundwater in semi-arid East Africa is under increasing pressure from population growth and climate change, yet its spatial distribution and future trajectory remain difficult to predict in regions with limited data. This study utilizes two distinct modeling approaches to map current groundwater potential (GWP) and project how it will change across Eastern Rwanda under three climate scenarios SSP1-2.6, SSP2-4.5, and SSP5-8.5 — for mid-century (2040–2059) and late-century (2080–2099) across Eastern Rwanda. Model training used 153 georeferenced boreholes from the Rwanda Water Resources Board (70/30 split) and twelve conditioning factors covering topography, hydrology, land cover, and geology. Both models performed strongly: RF achieved a macro-average AUC of 0.973 with an F₁-score of 0.85; MaxEnt attained a test AUC of 0.896. Land use/land cover and topographic wetness index were the dominant predictors in both models, with geology, rainfall, and soil texture as secondary controls. Under SSP5-8.5, MaxEnt projects Very Low GWP zones to expand by 29–48% late-century across most districts, while RF indicates Low-zone contraction exceeding -15% and Very High-zone losses of -7 to -9% in Kayonza. Nyagatare emerges as the most resilient district; Bugesera and Kirehe are most vulnerable across all scenarios. Model convergence under SSP5-8.5 provides cross-validated evidence particularly valuable given regional data constraints. The findings deliver spatially explicit, scenario-specific GWP projections directly applicable to Rwanda's National Adaptation Plan and offer a transferable methodology for groundwater risk assessment across semi-arid Africa.