Abstract:
We present an ensemble Kalman filter (EnKF) for assimilation of land surface observations into the Catchment Model of the NASA Seasonal-to-Interannual Prediction Project (NSIPP). By using one- and three-dimensional versions of the EnKF in a twin experiment, we assess the importance of horizontal error correlations for large-scale soil moisture estimation.

After calibration of model error parameters, the average estimation error in the root zone soil moisture is reduced by 58 % (relative to no assimilation) when horizontal error correlations are taken into account and by 44% when horizontal error correlations are neglected. The higher estimation accuracy of the three-dimensional filter must be traded off against its higher computational burden.