The WRF model was used to simulate five rainfall events in the upper Minjiang River from 2018 to 2022, and FNL and ERA5 data were selected as the initial field to analyse the spatial distribution, hourly processes and simulation errors of the rainfall, and to assess the influence of the underlying surface on simulation of the rainfall process. The results indicate that the WRF model can effectively simulate the sequential process of rainfall at multiple temporal resolutions when compared to the original field data and measured information. This capability enhances rainfall magnitude and distribution predictions. However, the model still struggles to accurately simulate peak rainfall and underestimates the measured cumulative and surface rainfall. Furthermore, initial field data with a spatial and temporal resolution of 0.25°×0.25° and 3-h are more effective for numerical simulation of rainfall in the upper Minjiang River, with the 10 to 14-hour forecast period yielding the best results. The elevation and land use condition updates have improved the cumulative rainfall results, the overall rainfall peaks are still lower than the measured values.Integration of updated initial field data can improve numerical model simulations of precipitation peaks and processes.