Evaluating multi-source precipitation data for streamflow simulation using the SWAT model in the Alpine Manas River...

Due to the influence of topography, climate, and other factors, the applicability of precipitation products varies significantly across different regions. Investigating the differences among multi-source precipitation products and their suitability for streamflow simulation in cold and alpine basins not only helps to reveal the advantages and limitations of each product but also plays an important role in improving the accuracy and application of hydrological models. Combining comprehensive indexes for precipitation magnitude and precipitation event detection capability, and the SWAT model, this study compares and analyzes the accuracy of four precipitation products (CMFD, MSWEP, ERA5-LAND, IMERG) in the arid regions of Xinjiang and explores their applicability in streamflow simulation for the alpine Manas River Basin. All four products demonstrate consistent intra-annual precipitation variation patterns. Significant discrepancies emerge specifically during summer. IMERG yields the lowest precipitation estimates, underestimating monthly totals with underestimations reaching approximately 39% to 45.7% in spring and autumn. Conversely, IMERG overestimates precipitation by 6.43% to 35.2% in July and August. Both CMFD and MSWEP underestimate precipitation during the rainy season while overestimating it in the dry season. The annual mean precipitation is overestimated by 4.05% to 6.59%. ERA5-Land exhibits a substantial overall overestimation of approximately 45.6%. CMFD demonstrates optimal overall precipitation performance with higher station correlation and relatively stable behavior across precipitation intensity levels. MSWEP and ERA5-Land exhibit comparable performance, though MSWEP shows smaller daily-scale biases than CMFD. MSWEP-based daily streamflow simulations yielded optimal results, achieving Nash-Sutcliffe Efficiency (NSE) > 0.65 during both calibration and validation periods with PBIAS < 15%. CMFD and ERA5-Land generated similar streamflow simulations, consistently overestimating streamflow by 12.79 ~ 27.07% in both periods. This study offers valuable references for streamflow simulation and water resource management in data-scarce alpine basins.