BAYESIAN TIME SERIES APPROACH AND ITS APPLICATION TO RETRIEVE GROUND AND VEGETATION VARIABLES FROM SMOS

M. Holmberg¹, J. Lemmetyinen¹, P. Richaume², A. Colliander¹, A. Kontu¹, J. Tamminen¹

¹Finnish Meteorological Institute, ²Centre d’Etudes Spatiales de la Biosphère CESBIO

The retrieval of geophysical variables from passive microwave observations is based on the inversion of empirical and physics-based forward models. Retrieval problems are often underdetermined due to the limited information in satellite observations and the highly complex nature of the observed scenes. This lack of information is resolved by introducing necessary prior information about the retrieved geophysical variables, for example, through Bayesian theory. While the contemporary approach is to consider each satellite overpass largely independently, this limits the amount of information that can be incorporated into the retrieval algorithm. Here, we present an extension of the contemporary retrieval into a joint retrieval of geophysical variables over a longer time series. By considering time series retrieval, information on the temporal behavior of geophysical variables can be incorporated into the retrieval process. This allows, for example, retrieving unknown time-constant parameters and disentangling geophysical signals with different temporal autocorrelation lengths.

We demonstrate the method by applying it to European Space Agency’s Soil Moisture Ocean Salinity (SMOS) satellite data over the Finnish Meteorological Institute’s Sodankylä research site in Northern Finland. In our demonstration, we retrieved two typical land surface parameters: soil permittivity and vegetation optical depth, with prior information on their respective temporal autocorrelation lengths. In addition, we retrieved two parameters that are usually not retrieved: soil surface roughness correction parameter and vegetation single scattering albedo, by assuming them as time-constant parameters. We compared the retrieved soil permittivity and vegetation optical depth with in situ data from the Sodankylä research station and found good agreement with the reference measurements. We will also discuss the role of the two time-constant variables in the retrieval problem. The presented method is general and can be applied to other passive microwave data sets, particularly for land surface and ocean applications.