Abstract: The distribution of meteorological stations in China is sparse and uneven, which can bring some difficulties for flood simulation of many small and medium-sized watersheds, where station-based precipitation information is usually insufficient. Merging satellite and ground-measured datasets is an effective method to obtain high spatiotemporal precipitation datasets, but the adaptability of the merged precipitation in flood simulation needs further studies. In this paper, Tunxi Basin is taken as an example to evaluate the capacity of these merged precipitations in flood simulation. Two important near-real-time satellite precipitation products in GPM era, IMERG_Early and GSMaP_NRT, are applied in this study. BP neural network model is adopted to merge them with ground measurements separately and then the two merged precipitation datasets are employed into Xin’anjiang model for flood simulation. Furthermore, in order to represent potential application of merged precipitation in the poor-gauged catchment, the number of meteorological stations in this area is reduced gradually to explore whether merged precipitation is still appliable when fusing only a little measured information. The results show that the merged precipitation is reliable no matter how much the in-situ information is available. Specifically, the averaged DC was above 0.8 and the qualification rate of Q_m and ΔH was larger than 70% and 90%, respectively. When there are limited meteorological stations within the watershed, the merged precipitation can provide more reliable simulation results compared with gauged observations. Moreover, the simulation results in Qianyang basin show that merged precipitation based on 4 stations could present as reliable performance as 12 meteorological stations did in forecasting flood events. Therefore, merging satellite precipitation and station-based information can provide reliable precipitation datasets for flood forecasting, especially for small and medium-sized watersheds with limited in-situ information.