Bayesian Model Averaging for Satellite Precipitation Data Fusion: From Accuracy Estimation to Runoff Simulation

Precipitation plays a vital role in the hydrological cycle, directly impacting water resource management and the prediction of flood and drought risks. This study explores an approach by applying the Bayesian Model Averaging (BMA) algorithm to merge multiple precipitation datasets, aiming to improve the accuracy of precipitation estimates for hydrological simulations. The BMA framework combines four widely used precipitation datasets—CHIRPS, ERA5, GSMaP, and IMERG—across the Ganjiang River Basin in China from 2008 to 2020. To evaluate the performance of the merged dataset, researchers analyzed it alongside its individual components and the MSWEP dataset at daily, monthly, and seasonal scales. Several key metrics, including CC, RMSE, and KGE, were used for assessment. Additionally, the Variable Infiltration Capacity (VIC) hydrological model was employed to examine how these datasets influence runoff simulation. The results indicate that the BMA-merged dataset significantly enhances precipitation estimation accuracy compared to individual datasets. It achieved the highest CC (0.72) and KGE (0.70) at the daily scale and demonstrated superior seasonal performance, particularly in minimizing biases during autumn and winter. When applied to hydrological simulations, the BMA-driven VIC model closely mirrored observed runoff patterns, proving its effectiveness for long-term runoff predictions in the region. Overall, this study underscores the potential of the BMA approach to refine precipitation inputs for hydrological models, offering valuable insights for sustainable water resource management and risk mitigation in complex hydrological environments.