The potential of 2D Kalman filtering for soil moisture data assimilation

We examine the potential for parameterizing a two-dimensional (2D) land data assimilation system using spatial error auto-correlation statistics gleaned from a triple collocation analysis and the triplet of: (1) active microwave-, (2) passive microwave- and (3) land surface model-based surface soil moisture products. Results demonstrate that, while considerable spatial error auto-correlation exists in the errors for all three products, the inclusion of this information into a 2D assimilation system does not significantly improve the performance of the system relative to a one-dimensional (1D) baseline. This result is explained via an analytical evaluation of the impact of spatial error auto-correlation on the steady-state Kalman gain, which reveals that 2D filtering requires the existence of large auto-correlation differences (between the assimilation model and the assimilated observations) in order to enhance the analysis relative to a 1D filtering baseline. As a result, large error auto-correlations alone (in both the model or observations) are not sufficient to motivate the application of a 2D land assimilation system. These results have important consequences for the development of land data assimilation systems designed to ingest satellite derived surface soil moisture products for water resource and climate applications.