The GRAPES_RAFS (Rapid Analysis and Forecast System) is based on GRAPES (Global and Regional Assimilation and Prediction System) model and GRAPES_3DVAR system, running on a high intermittent assimilation cycle to provide high resolution mesoscale analyses and short time numerical weather predication guidance for weather forecasting. The GRAPES_RAFS assimilates radiosonde observation and a lot of observations with high temporal and spatial resolution, such as aircraft, VAD wind profiles, surface station observation data, et al. Herein, the framework and flowchart of GRAPES_RAFS are technically described, and compared with the forecasting products of GRAPES_MESO, the short time nowcasting capability of this system and the critical techniques influencing its forecasting performance are also discussed. The research results show that the GRAPES_RAFS system is effective in providing more accurate short time nowcasting forecasts initialized with recent data than GRAPES_MESO system forecasts. The results also show that, 〖JP2〗high resolution observations, the background error covariance of GRAPES_3DVAR system,〖JP〗 the dynamical framework and physical processes of GRAPES model are keys to GRAPES_RAFS. The short time nowcasting performance of GRAPES_RAFS is a challenging task in the case that the assimilated observation data for GRAPES_RAFS are sparse.