A high-resolution (9-km) diabatic data assimilation system-the Local Analysis and Prediction System (LAPS), has been developed and used to initialize the real-time fifth-generation Pennsylvania State University-National Center for Atmospheric Research Mesoscale Model (MM5) at the Central Weather Bureau in Taiwan. During 2003, the more extensive network of four high quality Doppler radars and the access to satellite data from the Geostationary Operational Environmental Satellite (GOES-9) provided an excellent opportunity for advancing the short-range precipitation forecasts over the Taiwan area. The parallel forecasts of four tropical cyclones (Tropical Storm Morakot and Vamco, Typhoon Dujuan, and Tropical Storm Melor) that affected Taiwan in 2003 are performed, both with and without the inclusion of the LAPS cloud analysis scheme. Except for the inclusion of the LAPS cloud field, the model integrations are identical in all other respects. Forecast results demonstrate that using LAPS to diabatically initialize MM5 leads to an improved prediction of tropical cyclones in terms of the storm's track, intensity, cloud pattern, and movement of rainbands in the early portion of model prediction. During the first 6-h of the forecast, the heavy rainfall prediction associated with the cases studied was improved when the LAPS cloud analysis scheme was included. The assimilation of data from Doppler radars, and GOES-9 satellite, played an important role in the improvement of storm hydrometeorological features in the model initial condition and thus had a beneficial impact on reducing the model spin-up time. However, further studies are needed to clarify the reasons for the poor performance in simulating the typhoon eyewall. This paper represents a major step toward building a short-range mesoscale modeling system that predicts more realistic storm structures and rainfall distribution over the Taiwan area in real time. The overall results suggest that the impact of the LAPS/MM5 system can be significant for short-range, high spatial-re solution, rainfall prediction associated with a tropical cyclone, especially for the heavy rainfall occurring during the early hours of the model integration.