Comparison of InSAR Processing Techniques Applied to Cultivated Areas within California’s Central Valley

Abstract:

With the upcoming launch of the joint NASA-ISRO synthetic aperture radar (NISAR) mission, it is expected that upwards of 85 terabytes of data will be generated each day. To better manage and exploit data volumes of this size for geophysical investigation, the automation of interferometric synthetic aperture radar (InSAR) processing is increasingly critical. The Jet Propulsion Laboratory-Advanced Rapid Imaging and Analysis (JPL-ARIA) project aims to fully automate processing of InSAR data and produce standard surface displacement products for the general science community. Here, we assess the quality of ARIA InSAR products against independently processed InSAR datasets over California’s Central Valley. The Central Valley features well documented subsidence signals, complex spatial and temporal variations in land use, and independent observations of surface displacements from Global Positioning System (GPS) stations.

We examine a rich SAR dataset with 174 image acquisitions spanning 2015-2020 from the Sentinel-1 mission’s Track 144. We compare results independently processed using ISCE and GMTSAR against the ARIA products (also constructed with ISCE). Techniques mainly differ in their approach for downsampling, filtering, and unwrapping to produce the final interferograms. Spatially, the differences between the final interferograms are largest within the agricultural regions, particularly for fields that are cultivated for part of the time series and that lay fallow for other parts. These intermittently-cultivated fields are associated with temporally complex InSAR coherence, so the different approaches for masking and filtering result in significant differences in the final unwrapped phase . We quantify the spatially heterogeneous character of the differences between the individual interferograms processed by the three approaches using spatial structure functions. Independent land use maps offer an opportunity to evaluate the impact of these differences across the entire study region by allowing us to examine spatial structure functions binned by land use type. We find that all three processing approaches are similar across most of our study region. Unwrapping errors are minimal but do exist, and we assess the contribution of these to expected errors in the final time series.