Tropospheric corrections for InSAR: Statistical assessments and applications to the Central United States and Mexico

The rapid expansion of SAR data availability and advancements in InSAR processing methods has enabled the formation of ground displacement time series for many parts of the world where such research was previously hindered by decorrelation due to sparse temporal sampling and SAR operating frequency. In particular, free and open data access from the European Sentinel-1 constellation and the future NASA-ISRO SAR (NISAR) mission is encouraging the global community to move towards automated, cloud-based processing that can accommodate these rapidly growing data volumes and facilitate the use of a suite of corrections to the data. A key challenge is related to path delays introduced when the radar signal propagates through the troposphere. Tropospheric corrections estimated from empirical, phase-based methods and those using independent data from weather models, GPS, and radiometers have been included in open-source packages such as TRAIN, PyAPS and GACOS. Users within the InSAR community have reported varying degrees of success using these methods in a range of areas around the world. However, the various statistical metrics used to evaluate the reliability of tropospheric corrections are not consistently applied and often depend on the area and the spatial scale over which they are evaluated. Examination of a simple metric such as the overall reduction in phase variability within an interferogram does not allow the user to determine whether the improvement was at large or short length scales. We present a review of existing tropospheric correction methods and statistical performance metrics, providing guidelines for global assessment and verification of the efficacy of tropospheric correction methods. We summarize the assumptions and limitations for each correction method as well as each statistical performance metric. We examine two regions with different atmospheric characteristics - one Sentinel-1 swath covering the central United States and one swath covering south central Mexico, including part of the Pacific coast. As the SAR community moves towards reliance on global and automated InSAR processing platforms that incorporate tropospheric corrections, approaches such as those examined here can aid researchers in their efforts to evaluate such corrections and include their uncertainties in derived products such as surface displacement time series, coseismic offsets, processes that correlate with topography, and signals with smaller magnitude or larger spa