Improvements and evaluation of methods for reducing tropospheric effects in InSAR

Radar refraction in the troposphere is the primary source of noise in coherent Interferometric Synthetic Aperture Radar (InSAR) measurements. As InSAR expands to a more diverse base of applications and users seeking to push the limitations of the method, properly characterizing tropospheric delays remains an outstanding prohibitive challenge. We present new methods for both empirical and numerical model-based corrections that leverage machine learning algorithms. When comparing our correction results with standard approaches, we find our new approach to significantly improve upon existing methods with an ~80% reduction of total tropospheric noise in individual interferograms. We test these methods in the central United States and south-central Mexico -- respectively sampling both mountainous and low-relief terrain -- and show improvements are effective down to spatial scales of ~20 km. These advances may allow resolution of nonlinear deformation signals previously hidden by low signal-to-noise ratios, and allow for more confidence when interpreting interferograms and InSAR time series.