A Physically Constrained Inversion for Super-resolved Passive Microwave Retrieval of Soil Moisture and Vegetation Water Content in L-band

Abstract: Remote sensing of soil moisture and vegetation water content from space often requires underdetermined inversion of a zeroth-order approximation of the forward radiative transfer equation in L-band---known as the $\tau$-$\omega$ model. This paper shows that the least-squares (LS) inversion of the model is not strictly convex due to its saddle point structure. It is demonstrated that the widely used unconstrained damped least-squares (DLS) inversion of the model could lead to biased and physically unrealistic retrievals---chiefly because of the existing preferential solution spaces that are characterized by the eigenspace of the model Hessian. In particular, the numerical experiments show that for sparse (dense) vegetation with a shallow (deep) optical depth, the DLS tends to overestimate (underestimate) the soil moisture and vegetation water content for a dry (wet) soil. This paper proposes a new Constrained Multi-Channel Algorithm (CMCA) that confines the retrievals by an a priori information of the soil type and vegetation density. Unlike the existing algorithms, the presented approach can account for slowly varying dynamics of the vegetation water content over croplands through a temporal smoothing-norm regularization. We also demonstrate that depending on the resolution of the constraints, the algorithm leads to super-resolved soil moisture retrievals with much higher resolution than the spatial resolution of the radiometric observations.