GAP Filling of SMAP Soil Moisture Products Using Different Approaches

Satellite soil moisture products have great potential for many applications, such as drought monitoring and landslide warning. However, these applications often require accurate and continuous soil moisture records, and the missing values in satellite soil moisture datasets often hamper the usefulness of these products for such applications. This study firstly proposed and compared three approaches, i.e., linear regression, linear rescaling, and random forest to fill the missing values in the SMAP soil moisture products in both temporal and spatial dimensions, based on the seamless ERA5 data from 2016 to 2019. Then, a total of twelve auxiliary data were incorporated into the training datasets of random forest to improve the accuracy of gap-filled SMAP data. Finally, the gap-filled SMAP data were compared with the original SMAP data and validated by in situ measurements from 1071 sites worldwide. The results indicate: 1) when using only the ERA5 datasets, the random forest performs better than linear regression and linear rescaling methods in the training phase, but its skill degrades noticeably in the validation phase; 2) by adding the auxiliary data, the performance of random forest improves significantly in the validation phase; 3) the gap-filled SMAP data maintain or even exceed the accuracy of the original SMAP soil moisture, demonstrating the effectiveness of the proposed gap-filling method.