Integrating field measurements, Green-Ampt modelling, and machine learning to map infiltration heterogeneity and water resource planning in the Ethiopian highlands

Soil infiltration governs groundwater recharge, runoff, and erosion, yet it is often oversimplified in hydrological planning. This study integrates field measurements, Green-Ampt modelling, and Support Vector Machine (SVM) classification to map infiltration heterogeneity across five landscapes in Debre Markos, Ethiopia. Ninety-seven double-ring infiltrometer tests revealed strong variability: grasslands sustained rapid infiltration (> 30 cm/hr), grazing lands showed moderate but suppressed rates (10–36 cm/hr), and urban/bare lands exhibited severe reductions (< 20 cm/hr). Green-Ampt parameters (saturated hydraulic conductivity, suction head, and moisture deficit) were calibrated against field data, achieving robust agreement (R² = 0.81, RMSE = 0.42 cm/hr). Coupled with SVM, the hybrid framework improved predictive accuracy (87%) and produced spatially explicit infiltration maps that identified recharge-prone zones and vulnerable urban areas. These findings demonstrate that infiltration is not a uniform soil property but a driver of water security and climate resilience. Conservation of grasslands, restoration of degraded grazing lands, and adoption of permeable urban infrastructure are recommended to sustain recharge and reduce erosion. The findings have implications for SDG 6 (Clean Water and Sanitation) by identifying recharge‑prone zones that could support groundwater protection and for SDG 13 (Climate Action) by providing evidence to inform climate‑resilient watershed management. This study demonstrates the value of integrating field measurements, physically based modelling, and machine learning to provide site‑specific evidence for sustainable water resource planning in Ethiopia’s highlands.