L-band Radiative Transfer Parameter Estimation with a Particle Smoother
Montzka, Carsten1; Grant, Jennifer2; Drusch, Matthias3; Hendricks Franssen, Harrie-Jan1; Vereecken, Harry1
1Research Centre Jülich, GERMANY; 2Lund University, SWEDEN; 3European Space Agency, ESA-ESTEC, NETHERLANDS

The parameter estimation performance in a data assimilation (DA) framework is directly linked to the accuracy of the forcings, the observations, and the process description. Forcings and observations are typically prone to errors which are propagated through data assimilation to the parameter estimates and state updates. To reduce the impact of errors on simulation results, smoother techniques are used to reduce the effect of single erroneous forcings or observations on the parameter estimation by including observations at earlier times. Moreover, a present state variable can be updated under consideration of future observations. In this paper we present a Particle Smoother with sequential importance resampling (SIR-PS) for sequentially backward smoothing of particle weights in state and parameter space. It is applied to estimate the temporal variability of vegetation opacity and soil surface roughness from remotely observed L-band brightness temperatures. These parameters are key in the radiative transfer processing used for the generation of a soil moisture product. Except for extreme events such as storms, vegetation opacity and soil surface roughness should not vary too fast, so that a smoothing technology is still able to capture their seasonal variability. In situ observed soil moisture and soil temperature is used to drive the microwave forward model L-MEB (L-band Microwave Emission of the Biosphere). Observed Soil Moisture and Ocean Salinity (SMOS) brightness temperature data are assimilated into the Particle Smoother framework. The study aims are i) to improve the soil moisture estimation by the retrieval of radiative transfer parameters, and ii) to reduce the impact of single erroneous model forcings/observations by a smoother technique. The results show that the developed data assimilation system based on L-MEB and a sequential importance resampling particle smoother is able to simultaneously estimate the parameters vegetation opacity and soil surface roughness, even if the latter changes over time. In general, the proposed approach performs well for the selected experimental conditions.