A Novel Parallel Computational Framework for Processing Large InSAR Data Sets
Zinno, Ivana1; Imperatore, Pasquale1; Elefante, Stefano1; Casu, Francesco1; Manunta, Michele1; Mathot, Emmanuel2; Brito, Fabrice2; Farres, Jordi3; Lengert, Wolfgang3; Lanari, Riccardo1


The use of satellite data has been a key point in both scientific and commercial applications since the beginning of the Earth Observation (EO) satellite era. Current and past EO satellites have been providing to the user community not only an extraordinary amount of data but also a large variety of information. The number of possible EO applications that can lead to technological breakthroughs and economic revenue is more and more increasing. Such a growth demands and requests the creation of an EO ecosystem that is meant to be a loosely related cloud computing community, in which EO data, results and processing tools can be efficiently managed and shared by users.
A well-assessed EO methodology is the Differential SAR Interferometry (DInSAR) that has been demonstrated to be a remarkable technique to detect and monitor ground displacements within a centimetre accuracy [1]. In particular, one of the most extensively used methodologies for ground displacement time series analysis is the Small BAseline Subset (SBAS) algorithm [2]. Due to its characteristics, the DInSAR technique has been widely used not only in Geoscience but also in hazard monitoring and risk mitigation, mainly thanks to the availability of large SAR data archives, that have been acquired during the past decade by ERS, ENVISAT and RADARSAT satellites. Nowadays, there are still several fully functioning SAR constellations (e.g. COSMO-SkyMed, TerraSAR-X, RADARSAT-1/2) that will be soon assisted by the forthcoming Sentinel-1 satellite. In particular, the advent of Sentinel-1 sensor, which will be able to provide a large amount of SAR data (exceeding 1 Terabyte per day) [3] with improved revisit time and geographical coverage (with respect to the ERS/ENVISAT class), will have a great impact from application perspective.
Nonetheless, classical computational approaches are not suitable to cope with the processing of this huge data stream, and, in particular, the use of SBAS sequential implementation can be extremely time-consuming. Thus, the time-critical SBAS results exploitation demands new computational approaches, that should also include emerging parallel computing platforms.
The development of this novel paradigm will play a crucial role for the creation of an EO ecosystem. Here, processors, data and final products are shared and easily accessible for the scientific community, aiming at disseminating value added information and knowledge. In particular, such an ecosystem is envisaged to be an effective path to address the issue of processing and analyzing the present (ERS, ENVISAT, etc.) and future (Sentinel-1) huge SAR data stream.


The aim of this work is to present a novel parallel computational framework for DInSAR, focusing on the parallel SBAS algorithm development [2]. The final goal of the investigated computing environment is to process large SAR data sets within a time-frame that can span in a few hours for emergency scenarios and in a few days for standard elaboration aimed at hazard monitoring, with an enormous impact on applications exploitation. This challenging task have required to design a parallel SBAS-DInSAR version in which the algorithm makes efficiently use of the multiple processing units, by properly considering both shared memory and distributed memory architectures. It is worth emphasizing that programming a parallel system for the execution of a single complex application is still a challenging problem [4]. To do that, firstly, the processing chain has been analysed with reference to the inherent granularity, dependence structure, and regularity. Secondly, a proper task partitioning has been performed to identify those parts of the code that can be executed concurrently. Finally, the cloud-oriented parallel SBAS-DInSAR algorithm has been developed and validated within OpenNebula environment that is an open-source software for deploying private and hybrid clouds [5], based on Cloudera 3, an open-source Apache Hadoop distribution available through the ESA-CIOP (Cloud Operational Pilot) platform [6].
The proposed parallel SBAS-DInSAR algorithm has been implemented within the cloud computing framework provided by the ESA-CIOP environment (Fig. 1) which is used under the ESA-SSEP (SuperSites Exploitation Platform) flagship of the Helix Nebula project.

Figure 1. Web page for SAR data selection within ESA-CIOP project dedicated portal

A preliminary performance evaluation has been carried out comparing the results of the full SBAS processing chain - from the raw data download up to the generation of deformation time series - implemented both in the sequential and parallel version. The experimental analysis has been conducted by considering the whole ASAR archive (64 acquisitions) acquired from ascending orbits over the Napoli Bay, a volcanic and densely urbanized area in Southern Italy. The parallel SBAS-DInSAR processing chain has been tested on the ESA - CIOP platform exploiting 4 nodes, each one with two cores (Intel® Xeon®Processor X5650 - 2.66 GHz) and 16 GB of RAM, showing encouraging performance results leading to a speed-up of 375 %, with respect to the SBAS sequential version.


[1] R. Goldstein, and H.A. Zebker, "Mappings small elevation changes over large areas: Differential radar interferometry," J. Geophys. Res., vol.94, no.B7, pp.9183-9191, 1989.
[2] P. Berardino, G. Fornaro, R. Lanari, and E. Sansosti,"A new algorithm for surface deformation monitoring based on small baseline differential SAR interferograms," IEEE Trans. Geosci. Remote Sens., vol.40, no.11, pp.2375-2383, Nov.2002.
[3] P. Snoeij, E. Attema, M. Davidson, B. Duesmann, N. Floury, G. Levrini, B. Rommen, B. Rosich, "Sentinel-1 radar mission: Status and performance," IEEE Aerospace and Electronic Systems Magazine, vol.25, no.8, pp.32-39, Aug. 2010.
[4] H. El-Rewini and M. Abd-El-Barr, Advanced Computer Architecture and Parallel Processing, John Wiley & Sons, Inc, 2005.
[5] A. Goscinski, R. Buyya, J. Broberg (Eds.), Cloud Computing: Principles and Paradigms, JohnWiley & Sons, Inc., New Jersey 2011.
[6] http://ciop.eo.esa.int