In Situ Observation-Constrained Global Surface Soil Moisture Using Random Forest Model

The inherent biases of different long-term gridded surface soil moisture (SSM) products, unconstrained by the in situ observations, implies spatio-temporal patterns. In this study, Random Forest (RF) model was trained to predict SSM from relevant land feature variables (i.e., temperature, vegetation indices, texture, and geographical information) precipitation, based on data International Soil Moisture Network (ISMN.). results RF show an RMSE 0.05 m3 m?3 a correlation coefficient 0.9. calculated impurity-based importance indicates that Antecedent Precipitation Index affects most predicted moisture. coordinates also significantly influence prediction reduced 0.03 after considering coordinates), followed texture. pattern compared with European Space Agency Climate Change Initiative (ESA-CCI) product, using both time-longitude latitude diagrams. indicate captures spatial distribution daily, seasonal, annual variabilities globally.