A Comparison of Polarimetric Decomposition Techniques for Soil Moisture Estimation

The estimation of soil moisture represents a fundamental step in environmental monitoring activities such as land and water monitoring since it plays a crucial role for the forecasting of natural disasters, e.g. flooding events. The peculiarity of Synthetic Aperture Radar (SAR), a microwave imaging sensor, to ensure both a wide coverage and a fine spatial resolution, makes it a fundamental tool to retrieve soil geophysical parameters, such as roughness and moisture, at the watershed scale. However, in the case of dense vegetated areas, the use of single polarization SAR data does not allow the discrimination between surface and vegetation scattering signatures. This leads to inaccurate soil moisture estimates. In order to overcome this drawback promising results have been obtained by using SAR polarimetry. Accordingly, new polarimetric decomposition techniques have been proposed to separate vegetation and ground scattering components. In this paper a comparison between different polarimetric decomposition techniques for the estimation of soil moisture under vegetated covers is accomplished. Once the coherence matrix T is decomposed by the eigenvalue decomposition, the H-? plane is considered to classify the dominant scattering mechanism within ach resolution cell [1]. When the bare surface scattering turns out to be dominant, the extended Bragg (X-Bragg) formulation is used to model the scattering and soil moisture is achieved via the X-Bragg inversion approach. Otherwise, the scattering mechanism is considered as the result of ground and vegetation (or volume) components. In this case, the coherence matrix of the volume term has to be subtracted from the matrix T in order to retrieve the characteristic of the underlying soil. The purpose of this paper is the modeling of the volume coherence matrix by using different well-known approaches in order to compare the corresponding soil moisture estimates. After the removal of the volume term, the remaining ground component is properly modeled [1] and hence inverted for soil moisture estimation. Finally the soil moisture estimates from both bare and vegetated soil give the total soil moisture estimation. The proposed techniques are applied to the fully polarimetric data acquired by the NASA/Airborne Synthetic Aperture Radar (AIRSAR) during the Soil Moisture Experiments in 2003 (SMEX03). In particular, L-band data concerning the southern Oklahoma areas have been considered [2] since ground measurements of soil characteristics, both roughness and moisture, are available in those zones but the implemented processing chain can be operated over Italian regional sites.