Forecast of heavy precipitation and characteristic of key factors in key areas of the Sichuan-Tibet Railway based on FY-4 satellite

Focusing on the forecast and early warning of heavy precipitation in key areas of the Sichuan Tibet Railway,based on the high spatiotemporal resolution Fengyun-4 satellite data and ERA5 reanalysis data from the summer of 2020-2024, combined with the LightGBM machine learning algorithm, a heavy precipitation classification forecast model for key areas of the Sichuan-Tibet Railway was constructed by region. The model's interpretability was analyzed using SHAP values, and its forecasting ability was systematically evaluated. The physical characteristics of key forecasting factors were deeply analyzed. The results showed that the model achieved a critical success index of 0.41 for heavy precipitation forecasts in key regions, with a probability of detection reaching 0.76, demonstrating strong forecasting capabilities. Regional model analysis indicates that, the probability of detection for heavy precipitation is 0.78 and the critical success index is 0.49; The probability of detection in the southeastern region of the plateau is 0.69, and the critical success index is 0.33, indicating significant regional differences in forecasting effectiveness. SHAP and statistical analysis shows that heavy precipitation in the southeastern plateau is mainly dominated by satellite brightness temperature difference factors (such as TB_6.25-7.1 and TB_13.5-10.7) that reflect cloud thickness and deep convection development, while thermal instability parameters (such as CAPE, K index) and low-level vertical motion are the main indicators for heavy precipitation in western Sichuan. 60 minutes before the occurrence of heavy precipitation, key satellite parameters and physical parameters already exhibit statistically significant differences. Satellite parameters begin to show notable evolutionary characteristics as early as 150 minutes prior to the precipitation., providing quantitative reference for early warning. This study, through interpretable machine learning methods, not only improved the objective forecasting ability of heavy precipitation, but also deepened the understanding of regional heavy precipitation mechanisms, providing scientific support for disaster prevention and reduction in key areas of the Sichuan Tibet Railway.