Background  Soil moisture plays an important role in the exchange of substance and energy in the ground-air interface, which is also a key variable in drought monitoring and conservation of water and soil. Therefore, it is of great importance to obtain reliable soil moisture data.

Methods  In this study, the performances of three remote sensing soil moisture datasets, namely, SMAP (soil moisture active passive), ASCAT(advanced scatterometer) and AMSR2(advanced microwave scanning radiometer 2), over the Xiang River basin from April 2017 to October 2019 were evaluated using four evaluation indices which are R (correlation coefficient), BIAS (relative bias), RMSE (root mean squared error) and ubRMSE(unbiased root mean squared error). The CLSMDAS (China land soil moisture data assimilation system) dataset was used as the reference data. Based on the evaluation results, products with relatively higher accuracy among these products were selected for the data fusion process with EnKF (Ensemble Kalman Filter) method. Then the accuracy of the fusion data set was evaluated using the reference data and compared with the original remote sensing datasets to verify the effectiveness of the EnKF algorithm in data fusion.

Results  1) SMAP had the highest correlation with the reference data set, but there was a large deviation in a small part of the grid point scale, and there were individual outliers. ASCAT had a smaller deviation from the reference data set, but the correlation with the reference data set was lower than that of SMAP. However, AMSR2 cannot capture the change characteristics of soil moisture in the Xiangjiang River Basin, and seriously underestimates soil moisture. SMAP and ASCAT, which performed better among these three products, were selected for the fusion process. 2) The accuracy of merged soil moisture based on EnKF was high both in grid and basin scales. And the performance of the merged soil moisture on the evaluation indices were significantly improved, compared with the SMAP and ASCAT.At grid scale, compared with SMAP, the BIAS values of the merged soil moisture were lower at 42% grids. Compared with ASCAT, RMSE values and ubRMSE values of the merged soil moisture are improved at 80% grids, while R values were higher at 90% grids. At basin scale, compared with SMAP, the BIAS, RMSE and ubRMSE value of the merged soil moisture were improved 50%, 3%, and 3%, respectively. Compared to ASCAT, the R, BIAS, RMSE and ubRMSE values of the merged soil moisture were improved 56%, 65%, 27%, and 26%, respectively.

Conclusions  It is proved that a high-accuracy soil moisture dataset can be obtained through the fusion of multi-source remote sensing datasets, which provides important reference data for soil moisture monitoring and is of great significance for water and soil conservation and disaster reduction in the Xiang River basin.