A Cloud Computing Solution for the Efficient Implementation of the P-SBAS DInSAR Approach

We present an efficient Cloud Computing (CC) implementation of the Parallel Small BAseline Subset (P-SBAS) algorithm, which is an advanced Differential Interferometric Synthetic Aperture Radar (DInSAR) technique for the generation of Earth surface displacement time series through distributed computing infrastructures. The rationale of our approach consists in properly distributing the large data volumes and the processing tasks involved in the P-SBAS chain among the available (virtual and/or physical) computing nodes of the CC infrastructure, so that each one of these elements can concurrently work on data that are physically stored on its own local volume. To do this, both an ad hoc management of the data flow and an appropriate scheduling of the parallel jobs have been also implemented to properly handle the high complexity of the P-SBAS workflow. The proposed solution allows minimizing the overall data transfer and network load, thus improving the P-SBAS efficiency and scalability within the exploited CC environments. The presented P-SBAS implementation has been extensively validated through two experimental analyses, which have been carried out by exploiting the Amazon Web Services (AWS) Elastic Cloud Compute (EC2) resources. The former analysis involves the processing of a large (128 SAR images) COSMO-SkyMed dataset, which has been performed by exploiting up to 64 computing nodes, and is aimed at demonstrating the P-SBAS scalable performances. The latter allows us to show the P-SBAS capability to generate DInSAR results at a regional scale (150 000 km 2 in Southern California) in a very short time (about 9 h), by simultaneously processing 18 ENVISAT frames that correspond to a total of 741 SAR images, exploiting in parallel 144 AWS computing nodes. The presented results confirm the effectiveness of the proposed P-SBAS CC solution, which may contribute to further extend the frontiers of the DInSAR investigation at a very large scale.