Abstract: High-quality precipitation data is crucial for accurate hydrologic simulations.Remote sensing precipitation products offers an alternative to the gauge observations, however the accuracy of remote sensing precipitation products still needs to be further improved.To achieve a more precise precipitation dataset, we propose a spatio-temporal dynamic data fusion method that combines a LSTM-based spatial interpolation method with a Dynamic Bayesian Model Averaging(DBMA)data fusion method.Based on this method, precipitation data derived from IMERG-E and gauge observations are fused.Taking the Xiangjiang River Basin as an example, the reliability of the method as well as accuracy of the merged precipitation products are verified by evaluating the accuracy of the merged precipitation data and hydrological simulation.Results indicate that the merged precipitation data can better reflect the spatial distribution characteristics of precipitation in the Xiangjiang River Basin and have high correlation with the gauge observations, with the Correlation Coefficient(CC)reaching more than 0.65 at several stations, and most of the stations reaching the same or even higher level than CMA.Furthermore, compared with IMERG-E and CMA,the merged precipitation dataset enhances the performance of hydrologic simulations.With the use of this dataset, the Nash-Sutcliffe Efficiency coefficient(NSE)in flood event simulations increases from 0.41 and 0.80 to 0.84,respectively.