Making InSAR processing hardisks small again: Estimating the impact of Quantized Compressed SAR images on the precision of PS InSAR time series

Abstract

Modern satellites produce massive datasets. So, handling and processing bulk Synthetic Aperture Radar (SAR) imagery often represents a high entry level for researchers to tackle scientific challenges at regional or global scale. SAR imaging has been used as a remote sensing tool for studying Earth’s surface, such as measuring topography, terrain discrimination, forestry, and differential interferometry (InSAR) for monitoring the Earth’s surface deformation at millimetre scale. Due to the growing application of SAR imaging and the advancement of InSAR technique, more SAR satellites have been launched over the years. Moreover, those satellites have increased its temporal and spatial sampling rate, which contributes to the current rapid and massive data volume available for InSAR processing. Storage (on-line and off-line hardisks) requirements for InSAR processing is therefore constantly growing - over the past 7 years, Sentinel-1 SAR data downloaded by users has been increased by 620% (Serco, 2022). In the future, more SAR satellites with higher resolutions will be launched, not only increasing the carbon footprint by storing massive data in energy-intensive data centre, but also putting higher pressure on the computing resources of both the platforms and individual users for its scientific exploitation. In this study, we explored compression algorithms to downsize Sentinel-1 single look complex (SLC) images by 2 to 4 times. 162 ascending SLC images covering an area of around 19,335 km2 over the Pearl River Delta Region in the southern China were used in the test. In order to evaluate the performances of these compressed images for ground deformation monitoring, we compressed SLCs  generated by ISCE, calculate interferograms from compressed SLCs, and then compute time series of surface displacements using StaMPS InSAR processing software. Bulk SLC images can be compressed using the Julia package developed in this study and only decompress during calculation of interferograms, therefore images will not be saved in their expanded format. Our error analysis for signal reconstruction and the processed time-series results suggests that original 32-bit complex images can be can be optimally compressed using different quantization methods, reducing the storage required to handle large processing InSAR tasks. We confirmed that complex radar images retrieved from SAR satellites to be compressed up to a factor of 4 times, and achieving data reduction without sacrificing significant ground displacement precision.