Research on Extreme Rainstorm Prediction in Chengdu Region Based on a Multi-Weight Scheme Machine Learning Approach

This study focuses on extreme rainstorm events in the Chengdu region during 2025 and constructs an extreme rainstorm prediction model based on a multi-weight scheme machine learning framework. By integrating eight distinct sample weighting strategies, the research systematically analyzes 11 rainstorm episodes in the Chengdu area-eight used for model training and three reserved for independent validation. The study combines ERA5 reanalysis data with station observation data to develop a feature engineering system encompassing multi-level meteorological variables and derived features. Results indicate that different sample weighting strategies significantly influence model performance, with Scheme 6 (Custom 1) demonstrating optimal results across multiple evaluation metrics. The model effectively captures the spatial distribution characteristics of extreme precipitation and shows good predictive capability for stations experiencing rainfall ≥ 50 mm. This research provides a novel technical approach and methodological support for refined extreme rainstorm forecasting in the Chengdu region.