On the Use of Anisotropic Covariance Models in Estimating Atmospheric DInSAR Contributions

When studying geophysical processes through Differential SAR Interferometry (DInSAR), it is often necessary to estimate and subtract signals due to atmospheric inhomogeneities. To this end, stochastic models are often used to describe atmospheric phase delays in DInSAR data. As a first approximation, these can be modelled as isotropic, though this is a simplification, because SAR interferograms often exhibit anisotropic atmospheric signals. In view of this, it is increasingly advocated the use of anisotropic models for atmospheric phase estimation. However, anisotropic models lead to increased computational complexity in estimating the correlation function parameters with respect to the isotropic case. Moreover, performances can degrade when use is made of interferograms with only a few sparse points usable for computations, such as in the case of persistent scatterers interferometry (PSI) applications, especially when this estimation has to be done in an automated way on several tens of interferograms. In the present work we critically analyse the main aspects connected with the use of anisotropic models for DInSAR atmospheric delays, and we evaluate the advantage given by anisotropic modeling of atmospheric phase in the case of sparse grids of points. Through analysis of APS simulated data, we observe that a slight increase in the performances of reconstruction approaches can be obtained when sufficient sampling densities are available; based on these results, some recommendations for operational atmospheric phase estimation procedures are proposed.