Machine learning-based precipitation dataset for the Yarlung Zangbo River Basin: Generation, evaluation, and environmental factor analysis

Yarlung Zangbo River Basin, in the Southern Tibetan Plateau. Mountainous precipitation data often exhibit spatiotemporal inconsistencies and substantial biases, limiting hydrological applications. We develop a machine learning framework to merge CN05.1, ERA5-Land, CHIRPS, and TPMFD with environmental variables, producing a high-precision multiple-product merged precipitation dataset (MMPD). Performance is evaluated at daily, monthly, and annual scales, along with nine extreme indices. Shapley Additive Explanations (SHAP) quantifies environmental factors' influence on precipitation intensity. The results show that the MMPD significantly enhanced precipitation accuracy, achieving relative improvements in Modified Kling-Gupta Efficiency (KGE) of 60%, 23%, and 11% at daily, monthly, and annual scales, respectively, compared to the best-performing individual precipitation product (CN05.1 at daily scale and TPMFD at monthly/annual scales), and with notable enhancements in capturing extreme precipitation. The model interpretation results indicated that longitude, relative humidity, cloud cover, latitude, soil moisture, slope, elevation, and wind speed were the explaining variables with the highest importance for precipitation intensity. Different variables are characterized by specific statistical thresholds that promote precipitation intensity, indicating the impacts of Indian monsoon moisture transport, orographic lifting, and others. This framework provides an interpretable paradigm for machine learning-driven meteorological attribution and offers practical and theoretical guidance for developing high-precision precipitation products, especially in mountainous river basins with complex terrain. • Machine learning fusion improved precipitation accuracy by 11–60% at all temporal scales. • SHAP provides robust interpretability for the machine learning models. • Longitude and relative humidity are key influencing factors of precipitation intensity. • SHAP reveals statistically inferred thresholds for environmental variables affecting precipitation intensity.