Model-based Tensor Decompositions for Soil Moisture Estimation

Abstract

Current and future SAR missions offer the potential to obtain high-resolution soil moisture estimates with large coverage and frequent revisit times. This work extends model-based polarimetric inversion methods from coherency matrices to tensors by integrating an additional spatial data dimension. The resulting data tensor is decomposed into model-based components by solving an optimization problem. Tensors offer a larger observation space in comparison to coherency matrices, allowing the use of more complex physical models that cover a more extensive range of scenarios. The proposed three-component decomposition characterizes image areas in terms of model-based scattering mechanisms. Physical parameters, including the soil dielectric constant, are directly obtained through optimization. The proposed method is extensible and is well suited for geophysical parameter estimation from increasingly available multidimensional SAR data.