Bayesian model averaging approach to predict future rainfall erosivity in the Gumara-Maksegnit watershed, Ethiopia

The increasing impact of climate change on rainfall patterns and intensities necessitates robust methodologies to predict rainfall erosivity, a critical factor in soil erosion and hydrology. This study focuses on the Gumara-Maksegnit watershed in the Ethiopian highlands, a region highly vulnerable to soil erosion due to intense rainfall, rugged topography, and limited weather stations. Employing Bayesian Model Averaging (BMA), we predicted rainfall and rainfall erosivity under the RCP 4.5 and RCP 8.5 scenarios, addressing data limitations and improving prediction accuracy. Using historical rainfall data from 1971 to 2005 for calibration and validation, the study ensembled outputs from five Regional Climate Models (RCMs). Monthly rainfall predictions revealed that the BMA technique outperformed individual RCMs in capturing observed rainfall trends. BMA achieved an R 2 of 0.69 to 0.85 for monthly predictions, compared to 0.35–0.75 for individual RCMs, with reduced biases. The analysis indicates that projected rainfall is expected to surpass observed values by 12.5% and 20% under the RCP 4.5 and RCP 8.5 scenarios, respectively, with the peak months (July and August) experiencing the most significant increases. Rainfall erosivity is projected to rise by 12% under RCP 4.5 and 22% under RCP 8.5, with maximum value exceeding 900 MJ mm ha -1 h -1 yr -1 . The study demonstrates the effectiveness of the BMA technique in improving prediction accuracy, particularly in data-scarce regions. Incorporating geographic factors such as topography and altitude and integrating machine learning with BMA may improve its ability to capture extreme rainfall events, which were not explicitly evaluated in this study. • BMA outperformed individual RCMs for monthly rainfall predictions. • Rainfall projected to increase by 12.5% under RCP 4.5 and 20% under RCP 8.5 scenarios. • Peak rainfall increments are expected in July and August under future climate scenarios. • RCP 8.5 predicts a 22% rise in rainfall erosivity, reaching 900 MJ mm ha -1 h -1 yr -1 . • BMA enhances rainfall prediction reliability, vital for erosion risk in data-scarce areas.