New Advances of the P-SBAS Approach for an Efficient Parallel Processing of Large Volumes of Full-Resolution Multitemporal DInSAR Interferograms

This article presents an innovative, parallel implementation of the Small BAseline Subset (SBAS) approach to automatically and efficiently process large volumes of multitemporal differential synthetic aperture radar (SAR) interferometry (DInSAR) interferograms generated at the native full spatial resolution of the SAR images. The starting point of the developed full-resolution parallel-SBAS (FR P-SBAS) technique involves some algorithmic extensions for improving the quality of the DInSAR time series to effectively analyze extended deformations and localized displacement phenomena, such as those affecting single buildings and infrastructures. The main focus of the work is on the efficient and scalable FR P-SBAS processing chain implementation, extensively exploiting graphical processing unit (GPU) architectures. Moreover, the presented scalability analysis demonstrates the GPU capability of efficiently generating full-resolution displacement time series starting from large DInSAR datasets. Furthermore, it is also shown that the implemented processing solution easily allows us to deal with SAR data acquired through the Stripmap and TOPS modes. To assess the quality of the generated DInSAR products, an extensive experimental analysis is also shown, based on long sequences of X-Band COSMO-SkyMed Stripmap and C-Band Sentinel-1 TOPS acquisitions relevant to the Campi Flegrei Caldera (Southern Italy), which is monitored through a dense GNSS network. The presented results demonstrate the effectiveness of the FR P-SBAS processing chain in retrieving multifrequency and multiplatform displacement time series at the full spatial resolution with subcentimetric accuracy and in very short time frames, from a few hours for the COSMO-SkyMed datasets up to some tens of hours for the Sentinel-1 case.