EnOI Optimization for SMOS Soil Moisture Over West Africa

In land surface or numerical weather prediction (NWP) models, a soil moisture initialization scheme is important not to drift the prognostic variables to errors. We propose a novel approach for a stationary data assimilation scheme of ensemble optimal interpolation (EnOI) effective for soil moisture and ocean salinity (SMOS) soil moisture initialization. For the optimization of EnOI, the satellite retrieval error specification was conducted rather than ensemble evolution. As combining two ensembles generated from a satellite retrieval and a land surface model, this approach is termed as "two-step EnOI" in this study: (first step) the SMOS soil moisture retrieval ensembles (i.e., errors in brightness temperature, landscape, and geophysical parameters) were merged with SMOS L3 data; (second step) the data assimilation result from the first step was further used for the observations of the EnOI. This two-step EnOI was compared with a sequential ensemble Kalman filter (EnKF) evolving model state ensembles over time but assuming global constant a priori random errors for the SMOS observations. The point-scale comparison results showed that two-step EnOI was better matched with the field measurements than the SMOS L3 data and a sequential ensemble KF scheme. On meso-scale, a spatial average of two-step EnOI reached that of a sequential ensemble KF with the significantly reduced ensemble size. These results suggest that the performance of two-step EnOI is comparable to a sequential ensemble KF but computationally more effective. From this, it is illustrated that appropriate error specification of satellite retrieval is more important than a sequential evolution of model state ensembles, and brightness temperature ensemble mean can reduce the SMOS retrieval biases without sequential evolution.