InSAR and GPS Time Series Analysis in Areas with Large Scale Hydrological Deformation: Separating Signal From Noise at Varying Length Scales in the San Joaquin Valley

Areas of large-scale subsidence are observed over much of the San Joaquin Valley of California due to the extraction of groundwater and hydrocarbons from the subsurface.These signals span regions with spatial extents of up to 100 km and have rates of up to 45 cm/yr or more. InSAR and GPS are complementary methods commonly used to measure such ground displacements and can provide important constraints on crustal deformation models, support groundwater studies, and inform water resource management efforts. However, current standard methods for processing these data sets and creating displacement time series are suboptimal for the deformation observed in areas like the San Joaquin Valley because (1) the ground surface properties are constantly changing due largely to agricultural activity, resulting in low coherence in half or more of a SAR frame, and (2) the deformation signals are distributed throughout the SAR frames, and are comparable to the size of the frames themselves. Therefore, referencing areas of deformation to non-deforming areas and correcting for long wavelength signals (e.g. atmospheric delays, orbital errors) is particularly difficult. We address these challenges by exploiting pixels that are stable in space and time, and use them for weighted spatial averaging and selective filtering before unwrapping. We then compare a range of methods for both long wavelength corrections and referencing via automatic partitioning of non-deforming areas, then benchmark results against continuous GPS measurements. Our final time series consist of nearly 15 years of displacement measurements from continuous GPS data, and Envisat, ALOS-1, Sentinel SAR data, and show significant temporal and spatial variations. We find that the choice of reference and long wavelength corrections can significantly bias long-term rate and seasonal amplitude estimates, causing variations of as much as 100% of the mean estimate. As we enter an era with free and open data access and regular observations plans from missions such as NISAR and the Sentinel constellation, our approach will help users evaluate the significance of observed deformation at a range of spatial scales and in areas with challenging surface properties.