Satellite Soil Moisture Data Reconstruction in the Temporal and Spatial Domains: Latent Error Assessments and Performances for Tracing Rainstorms and Droughts

Intermittent records of satellite soil moisture data are major obstacles that constrain their hydrometeorological applications. Based on the European Space Agency Climate Change Initiative (ESA CCI) soil moisture combined product, two machine learning models were employed to reconstruct soil moisture in China during 1979–2019 in both temporal and spatial domains, and latent errors for reconstructed series, as well as their performances for tracing climate extremes, were analyzed. The results showed that with the homogeneity of available data over space, the spatial approach performed well in reproducing the spatial heterogeneity of soil moisture (with medians of the correlation coefficient (CC) above 0.8 and root mean square errors (RMSEs) ranging from 0.02 to 0.03 m3∙m−3). The temporal approach (CC values of 0.7 and RMSEs ranging between 0.02 and 0.03 m3∙m−3) was superior in capturing the seasonality features and the timely and accurate mapping of short-term soil moisture dynamics impacted by rainstorms. However, both approaches failed to identify the location and severity of droughts accurately. The findings highlight the benefits of combining the strengths of both temporal and spatial gap-filling approaches for improving the estimation of missing values and hydrometeorological applications.