Study region The Upper Tekeze Basin (UTB), northern Ethiopia, is a semi-arid basin characterized by high rainfall variability, recurrent droughts, rugged topography, and limited surface water availability. Study focus This study integrates the Soil and Water Assessment Tool (SWAT), Random Forest (RF) modeling, and geospatial analysis to identify rainwater harvesting (RWH) potential zones. A total of 140 inventory sites (equal suitable/non-suitable) were derived from field surveys, with 70% used for training and 30% for testing. Multicollinearity was low (VIF < 5). SWAT simulated runoff with good performance (R & sup2; = 0.77, NSE = 0.76 for calibration; R & sup2; = 0.75, NSE = 0.75 for validation). RF identified slope (17.5%), drainage density (11.5%), and topographic wetness index (10.1%) as key controlling factors. New hydrological insights for the region A fuzzy gamma overlay (gamma = 0.9) with constraints indicated that 16.9% of the basin is suitable for RWH (very high: 2.3%; high: 14.5%). Model performance was strong (AUC = 0.91; accuracy = 81.4%), and field validation confirmed agreement with existing RWH structures. Results highlight the dominant role of topography and runoff processes and demonstrate the effectiveness of integrating hydrological modeling with machine learning for spatial RWH planning.