Leveraging GPUs for handling large SAR data volumes for the NISAR mission

Abstract

The NASA-ISRO Synthetic Aperture Radar (NISAR) mission [4] will redefine the future of earth science in terms of both the quality as well as the quantity of data that will be downlinked daily. The InSAR Scientific Computing Environment (ISCE) [1], a powerful, modular, and extensible framework, is expected to be used as the processing engine for generating tens of terabytes of SAR data products from the NISAR mission on a daily basis. Until recently, the ISCE framework was designed to exploit traditional CPU-based computational architectures. Simulations with SAR data from ESA's Sentinel-1 constellation have shown that these traditional implementations maybe insufficient to handle the daily influx of data from the NISAR mission. This paper describes the initial efforts by the NISAR project team to explore the use of Graphics Processing Units (GPUs) to accelerate SAR data processing. Preliminary results from “GPU-izing” interferogram generation workflows indicate that processing times can be reduced by an order of magnitude without loss in precision, potentially setting a new standard for radar processing in the world of “Big SAR Data”.