Abstract: A forecast model for short-time heavy precipitation (greater than 20 mm h⁻¹) for 1–12 h at 1-h intervals in Fujian Province and its neighboring provinces is established based on neighborhood analysis with an optimal TS (threat score). This model utilizes real-time precipitation observational data from automatic weather stations, 0–120 min quantitative precipitation forecast from SWAN-QPF (Severe Weather Automatic Nowcasting) of the China Meteorological Administration, and forecast precipitation data from global and regional models during April–September 2021–2023. A neighborhood test with a radius of 40 km is adopted. The results indicate that the accuracy of the nowcast in the initial hours is greatly improved by incorporating real-time precipitation observations to forecast the short-time heavy precipitation (persistence forecast) compared with forecasts based on multimodel optimal weight integration. After the optimal elimination threshold correction, the TS can reach 37.5% for 2021 and 32.2% in 2022 for 1-h forecast lead time when the forecast precipitation is calculated using 10 min real-time precipitation prior to the forecast production, with a neighborhood radius (R_i) of 0.5° and the average of the top 5 (N_top=5) heavy precipitation stations within R_i. The TS can reach 22.2% for 2021 and 19.5% in 2022 in 2-h forecast lead time. The TS of the consensus forecast combining global and regional models with optimized weights can reach 16.2% for 2021 and 16.6% in 2022 for 1-h forecast lead time (18.0% and 14.2% in 2-h forecast lead time, respectively) when the forecast precipitation is calculated using R_i=0.6° for each model and N_top=15. The SWAN-QPF short-time heavy precipitation predictions revised by the optimal elimination threshold (with R_i=0.3° and N_top=15) are also better than the multimodel optimal weight integration, but they are less effective than predictions using real-time observational precipitation in the first few hours. For the 3–12-h forecast lead time, multimodel optimal weight integration is better than the other two methods. The above mentioned multisource data, revised by the optimal elimination threshold, are further integrated with optimized weights for 1–4-h forecast lead time, whereas multimodel optimal weight integration is adopted in the 5–12-h forecast lead time to establish the short-time heavy precipitation model for 1–12-h forecast lead time at 1-h intervals. The parameters trained with 2021 and 2022 data are applied to forecast short-time heavy precipitation in Fujian in 2023, yielding TS of 42.7%, 28.8%, 23.1%, and 20.2% for the 1–4-h forecast lead time, respectively, with all values exceeding 17% for the 5–12-h forecast lead time.