Modeling InSAR tropospheric delay based on their spatiotemporal characteristics: Application to postseismic displacements of the 2021 Maduo earthquake, China

Interferometric Synthetic Aperture Radar (InSAR) measurements suffer from undesirable errors caused by tropospheric delays. Generally, two classes of methods are used to reduce InSAR tropospheric errors: Methods based on independent external information and methods using directly the InSAR data themselves (i.e., data-driven methods). External information methods use GNSS data, meteorological data, atmospheric model outputs, etc., and can be reliable but the external information is usually of significantly lower spatial resolution than needed to correct InSAR data. Data-driven methods, on the other hand, are based on the InSAR data directly and thus do not require any external data. Given that tropospheric delays are usually divided into two components, i.e., the stratified and turbulent components, and that these two components have different spatiotemporal characteristics, they are usually treated separately in data-driven correction methods. However, during such separated error reduction, the existence of one component affects the mitigation performance of the other component, which results in somewhat biased reduction of the tropospheric delays.

Therefore, in this study we propose a new method to simultaneously model and mitigate the InSAR turbulent and stratified delays by taking their spatiotemporal characteristics as a priori information. In this method, which we call DetrendInSAR, the turbulent delay is regarded as a spatially slow-changing process and can therefore be fitted by position-related polynomials within a small area (e.g., 1 km x 1 km); the stratified delay can be linearly fitted with the local terrain height; and these a priori information is combined to establish a solvable mathematical model with respect to the tropospheric delay based on a novel pixel-by-pixel window-based modeling strategy. Besides, the displacement signals in the InSAR observations are assumed to be a temporally smooth process and therefore providing additional constraints for distinguishing between the displacements and turbulent delays in the DetrendInSAR modeling process. We validate the DetrendInSAR method using simulated datasets and a 16-month-long Sentinel-1 SAR data sequence of the postseismic deformation after the 22 May 2021 Maduo earthquake, China. We compare our results with the traditional data-driven strategy that fits a ramp and a terrain-related linear function over the whole image based on far-field signals and suppresses the turbulent delay by temporally averaging adjacent SAR-image acquisitions. The results obtained from ascending and descending orbits illuminate the logarithmic decay of the postseismic deformation after this earthquake. We also calculated the one-year postseismic east-west and vertical displacements of this earthquake, indicating that poroelastic rebound contributed to the postseismic deformation, rather than only the afterslip considered in previous studies.