Soil moisture initialization for climate prediction: Assimilation of scanning multifrequency microwave radiometer soil moisture data into a land surface model

Abstract

Climate model prediction skill is currently limited in response to poor land surface soil moisture state initialization. However, initial soil moisture state prediction skill can potentially be enhanced by the assimilation of remotely sensed near‐surface soil moisture data in off‐line simulation. This study is one of the first to evaluate such potential using actual remote sensing data together with field observations. Here the ensemble Kalman filter (Kalman, 1960) is used to assimilate scanning multifrequency microwave radiometer derived near‐surface soil moisture data from 1979 to 1987 into the catchment‐based land surface model (CLSM). CLSM is used by the NASA Goddard Modeling and Assimilation Office global climate model. Enhancement to land surface soil moisture initialization skill is evaluated for Eurasia using the ground soil moisture measurements collected in Russia, Mongolia, and China. As initial model and observation error predictions were poor, the assimilation improved both the surface and root zone soil moisture estimates only when the observation error was less than the model error. This emphasizes the need for good quality remotely sensed soil moisture data sets, together with reliable observation and model error assessments, in order to ensure improved soil moisture estimates through data assimilation. When the relative magnitude of predicted observation and model error was matched to the error determined from field observation comparison, improvements in root zone and surface soil moisture estimates were guaranteed given unbiased model and satellite observations.