Soil moisture and surface flux estimation over the Hydros OSSE site: Assimilation of multi-scale active and passive L-band microwave observations using the ensemble Kalman smoother.

A land data assimilation framework, centred on the Ensemble Kalman Smoother (EnKS) is used to estimate surface and root zone soil moisture as well as surface fluxes over the Hydros OSSE site through the assimilation of synthetic radiometer and radar observations.

In previous work it was argued that soil moisture estimation is a reanalysis-type problem, and that smoothing (a.k.a. batch estimation) techniques are more appropriate than filtering techniques. Early experiments focussed on a simple extension of the ensemble Kalman filter (EnKF) in which the state vector was distributed in time. This yielded improved estimates of surface and root zone soil moisture in a synthetic experiment. However, the augmentation of the state vector increased the computational burden of calculating the covariances and the Kalman gain term and also required a larger number of ensemble members for convergence. This rendered the approach unsuitable for spatially distributed problems.

The ensemble Kalman smoother (EnKS) is a more computationally efficient alternative in which observations are used as they become available to update the ensemble at prior estimation times in addition to the current forecast ensemble. It is considered superior to other smoother algorithms such as the Ensemble Smoother as it uses the EnKF as its first guess, guaranteeing an estimate at least as good as the EnKF. Each update with a subsequent observation results in a slight change in mean and a reduction in ensemble variance. As observations further into the future are used, the improvements become negligible, indicating that they are beyond the decorrelation time. Previous work using real ESTAR observations from SGP97 demonstrated the advantages of using the EnKS in land data assimilation. It yielded improved soil moisture and surface flux estimates compared to the EnKF while incurring only a modest increase in computational burden. Furthermore, it was shown that the EnKS could be implemented as a fixed lag smoother, further reducing the computational expense. It was shown that the lag required to estimate soil moisture at depth is greater than that at the surface.

Here, the EnKS is applied at the Hydros Observing System Simulation Experiment (OSSE) site in Arkansas to combine higher resolution but noisy backscatter data (3km) with more accurate but coarse resolution radiobrightness temperatures (36km) in order to estimate soil moisture and surface fluxes at 6km resolution. ``True'' soil moisture from April 1 to July 30 1994 at 1km resolution was obtained using the TOPLATS model, and the Hydros forward microwave emission and backscatter model were used to generate synthetic observations with a revisit time of 3 days. Surface and root zone soil moisture as well as surface fluxes are estimated four times daily to capture the diurnal cycle. The added value of including radar observations is compared to assimilation of radiometer observations only. The impact of constructing a spatially distributed state vector, rather than assimilation on a grid of independent soil columns/pixels is evaluated. It will be demonstrated that this data assimilation framework can be used to make mission design trade-offs for future soil moisture missions.