Abstract: Here the multi-source data fusion of soil water content was proposed to improve the regional applicability and accuracy of grid data using decomposition and reconstruction of the discrete wavelet transform. The products of soil water content were estimated to test the feasibility and applicability of the improved fusion from the ESA-CCI, SMAP, and ERA5-Land. The data was also collected from the ground station in the Yellow River Basin, including the main agroclimatic region from June to September 2016-2018. The results showed that the improved fusion effectively captured the multi-scale feature information from the original data source. These multi-source and multi-scale feature information was then weighted to merge and reconstruct at each decomposition level. Finally, the regional applicability was effectively improved in the different agroclimatic regions, together with the accuracy of spatiotemporal structure and fluctuation. The merged data of soil water content was fully assessed, with the root mean square error (RMSE) of 0.053 m3/m3, the bias of 0.001 m3/m3, and the correlation coefficient (r) of 0.721. Furthermore, the merged data of soil water content was also much better than any single original one, in terms of the overall performance of spatiotemporal decomposition. The higher time-series RMSEs of ERA5-Land soil water content were found in the transition region from the humid to the semi-arid agricultural climate zone, as well as the SMAP and ESA-CCI soil water content in the rainy region. By contrast, the spatial-series RMSEs steadily fluctuated with a narrow range interval (0.07-0.08 m3/m3), especially the amplitude of the fluctuation was controlled within 0.1 m3/m3 during increasing precipitation. The merged data of soil water content shared the lower positive bias (0.005 m3/m3) in the time series and lower negative bias (-0.002 m3/m3) in the spatial series, indicating the higher positive bias of ERA5-Land and ESA-CCI soil water content, together with the overall negative bias of SMAP soil water content. There was a similar spatiotemporal range in the correlation coefficient of the merged and ERA5-Land soil water content, indicating a slightly lower in the time series (0.578), and much higher in the spatial series (0.497). The higher matching of the merged data was achieved in the structural characteristics of the multi-scale spatiotemporal fluctuation spectrum in the observation. The higher total power of cross wavelet and stronger ability for the merged data were observed to characterize the spatiotemporal fluctuations of real soil water content. There was consistent spatiotemporal fluctuation within the time scale of 25d. The fusion scheme was considered in the multi-dimensional trends and structure characteristics from the three data sources at different spatial scales in the fusion processing of multi-level wavelet decomposition and weighted merging level by level, compared with the Triple-Collection Analysis, and Cumulative Distribution Function Matching. Hence, the maximum layer was determined for the wavelet decomposition. There was a slightly lower linear slope of the merged soil water content, compared with the observations, due to the common feature of "overestimation of lower value, and underestimation of high value" of ERA5-Land, ESA-CCI, and SMAP soil water content. Feasible approaches in the future can be expected to improve the fusion data sources and algorithm, such as the spatiotemporal continuity and accuracy of ESA-CCI and SMAP soil water content. The observations were introduced into the fusion processing, and then the advanced fusion was developed to merge the more useful information relating to spatiotemporal characteristics of soil water content. In short, a better expectation was obtained to improve the regional soil water content. The finding can provide feasible and effective applications in the regional ecological environment, sustainable agricultural development, soil and water conservation, as well as disaster prevention and mitigation.