| Summary: | Spatio-temporally continuous and high-quality soil moisture (SM) is very important for assessing changes in the water cycle and climate, especially over the Tibetan plateau (TP). Data fusion is an important method to improve the quality of SM product. However, limited observation overlaps between different satellite SM products, caused by inherent gaps, make it difficult to fuse them to create a continuous and high-quality product. In this study, an SM spatio-temporal continuity and quality simultaneously improving algorithm is proposed. The first step of the approach is obtaining spatio-temporally continuous reference data, including land surface temperature (LST), normalized difference vegetation index (NDVI), Albedo, and digital elevation model (DEM). The second step is training the general regression neural network (GRNN) model with all available essential climate variables (ECV) and Fengyun (FY) SM. The last step is predicting the spatio-temporally continuous and high-quality SM using the trained GRNN derived by the spatio-temporal continuity reference data. An implementation of the algorithm on the TP showed that, compared with the original ECV and FY SM, both the continuity and quality of the fused SM product were largely improved in terms of coverage (72.5%), correlation (R = 0.809), root mean square error (0.081 cm<sup>3</sup> cm<sup>-3</sup>) and bias (0.050 cm<sup>3</sup> cm<sup>-3</sup>). The algorithm showed a good performance in obtaining spatio-temporal variation fusion weights over the TP. This spatio-temporally continuous and high-quality SM of the TP will help advance our understanding of global and regional changes in water cycle and climate.
|
|---|
