Sentinel-1 Normalized Radar Backscatter processing on the High-Performance Data Platform terrabyte

Abstract

The rapid growth of satellite-based Earth Observation (EO) data is accompanied by growing challenges in continuous processing of global data sets into high-quality geoinformation. To support the analysis of EO data in the context of environmental and global change, it is essential not only to provide EO data, but also to process them into global decadal data sets and Analysis Ready Data (ARD). Therefore, the German Aerospace Center (DLR) and the Leibniz Supercomputing Centre (LRZ) of the Bavarian Academy of Sciences and Humanities are building up an innovative high-performance data analytics platform (HPDA), called terrabyte. In addition to the satellite data of the Copernicus program, United States Geological Survey (USGS) and National Aeronautics and Space Administration (NASA), derived ARD products are made available as well. The Sentinel-1 ARD product is provided as Normalized Radar Backscatter (NRB), compliant with the Committee on Earth Observation Satellites (CEOS) ARD (CARD) product family specifications (PFS). The processing of the Sentinel-1 NRB is performed by the Multi-SAR System, a pre-processing system for Synthetic Aperture Radar (SAR) data from various space missions (ERS-1/2, Envisat ASAR, RADARSAT-1/2, ALOS PALSAR-1/2, TerraSAR-X, and Sentinel-1) [2, 3]. The Multi-SAR System comprises ortho-rectification, Kennaugh decomposition, multi-looking, radiometric calibration, image enhancement using the multi-scale multi-looking (MSML) approach, and normalization based on a hyperbolic tangent function for the final lossless data compression. The system can be used for the processing of all Level-1 SAR data (focused, slant range, single-look, complex) independent of their geometric, radiometric, and polarimetric resolution. Its main benefit is the usage of the same workflow and similar parameter configuration file, independent of different SAR input data formats and interfaces. The processor is implemented at DLR’s German Remote Sensing Data Center (DFD). For the adaption of the NRB product to the CARD specification and the processing on the HPDA, the Multi-SAR System has been developed further. This concerns the radiometric terrain correction [4], the creation of mandatory output layers (e.g., the Gamma Naught Radiometric Terrain Corrected image, local contributing area mask) and the containerization of the software. For a first application, the adopted Multi-SAR Sentinel-1 NRB processor was integrated in the Sentinel-1-Based Water and Flood Monitoring processor (WFMP) [5] which is implemented and used operationally at the Center for Satellite Based Crisis Information (ZKI) at DLR. The WFMP is a near real time emergency mapping application, based on a convolutional neural network using Sentinel-1 amplitude data. The main advantage of the integration of Multi-SAR into the WFMP is the automatic and optimized data conversion from Sentinel-1 Ground Range Detected (GRD) images to a water and flood map in one processing chain. With the integration of this first application we want to gain experience in the operational processing on the platform and in performance optimization. In this presentation we describe the Sentinel-1 NRB processing workflow on the HPDA platform terrabyte, the challenges that have to be overcome in big data processing, as well as the processing steps and the CARD-compliant output layers of Multi-SAR. Furthermore, the adaptation of Multi-SAR to the WFMP will be presented. In order to investigate whether different NRB products lead to different results, the WFMP will be applied on NRBs generated by Multi-SAR and SNAP. To cover different topographical characteristics, a comparative study with globally distributed, quality checked Sentinel-1 datasets [6] will be carried out.