Evaluating Machine Learning and Geostatistical Methods for Spatial Gap-Filling of Monthly ESA CCI Soil Moisture in China

Obtaining large-scale, long-term, and spatial continuous soil moisture (SM) data is crucial for climate change, hydrology, water resource management, etc. ESA CCI SM such a large-scale long-term (longer than 40 years until now). However, there exist gaps, especially the area of China, due to limitations in remote sensing as complex topography, human-induced radio frequency interference (RFI), vegetation disturbances, The gaps make cannot achieve continuity, which entails study gap-filling methods. In order develop suitable methods fill whole we compared typical Machine Learning (ML) methods, including Random Forest method (RF), Feedforward Neural Network (FNN), Generalized Linear Model (GLM) with geostatistical method, i.e., Ordinary Kriging (OK) this study. More 30 passive–active combined from 1982 2018 other biophysical variables Normalized Difference Vegetation Index (NDVI), precipitation, air temperature, Digital Elevation (DEM), type, situ International Soil Moisture (ISMN) were utilized Results indicated that: (1) gap frequent found not only cold seasons areas but also warm areas. ratio pixel numbers can be greater 80%, its average around 40%. (2) ML all up. Among RF had best performance fitting relationship between variables. (3) Over simulated areas, comparable OK, they outperformed FNN GLM greatly. (4) networks, achieved better OK method. (5) We explored various strategies SM. demonstrated that strategy constructing monthly model one simulating another disturbance performance. Such combining suggested filling China.