Abstract:This study presents the development and implementation of the Rapid-Refresh Integrated Seamless Ensemble (RISE) forecasting system, an innovative multi-source data fusion system designed to provide “100-meter-scale, minute-level-update” weather forecasts. It was originally created to support meteorological services during the Beijing 2022 Olympic and Paralympic Winter Games. Here, this article demonstrates the significant improvements over several years after the games in short-time forecasting and nowcasting capabilities, which have been achieved through the integration of key technologies and the application of machine learning and deep learning methods within the high-resolution forecasting framework of the RISE system. The system’s novel key features are reflected in the refined short-time forecasting and nowcasting for heavy precipitation and thunderstorm gale， as well as in the nowcasting for the initiation of severe convection. The technologies integrated into the RISE system include: a bias-corrected, high-resolution gridded precipitation analysis scheme, the machine learning-based gridded precipitation short-time forecasting and nowcasting algorithms, a novel dynamic-statistical ensemble method for gust prediction using multi-source data fusion, an interpretable deep learning model for nowcasting convectively high winds or thunderstorm gale, a nowcasting method for convective initiation that integrates satellite observations and storm tracking, and an integration scheme for multiple numerical models. Then, comprehensive verification analyses confirm that these methodologies have significantly enhanced the forecast accuracy for precipitation and thunderstorm gale, particularly for the 0-6 h short-time forecasting and nowcasting during flood seasons. Finally, this study concludes with a critical discussion of existing challenges and potential future directions for “100-meter-scale, minute-level-update” weather forecasting.