Geocoded bursts: Single ARD product to support backscatter and InSAR analysis with Sentinel-1 SAR data
Descartes Labs has developed a single Analysis Ready Product to support both SAR backscatter as well InSAR analysis using Sentinel-1 data. We have implemented an efficient geocoding-based approach (Zebker, 2017) that lets us generate ARD products at the granularity of a single Sentinel-1 TOPS burst. These ARD products are on grids compatible with widely used optical imagery products; thus allowing users to use these ARD products within frameworks/data engines developed for optical data analysis. To enable scalable generation of this product, we also developed two key tools - a global TOPS burst index and a mechanism to access parts of the ESA Sentinel-1 SLCs quickly, efficiently and cost-effectively. We built a global database of individual TOPS bursts by parsing the annotation files of all the Sentinel-1 IW mode SLCs; and also devised a mechanism to update the burst index with new footprints from new SLC products if the instrument was turned on over previously un-imaged regions. This burst index also allowed us to overcome well known issues caused by inconsistent spatial framing of SLC data over the lifetime of the Sentinel-1 mission. To access individual bursts from ESA’s zipped SAFE files, we borrowed ideas from the neuro-imaging and genomics community that let us index the zip files, thereby allowing us to reconstruct contiguous streams of data from partial streams of compressed data. Note that we need to make a complete pass on the SLC files once to index the compressed data stream and in the process. We label the bursts contained in each product and extract the SAR metadata for faster access. SAR metadata by itself can assist with data staging decisions for various analytics pipelines and can be easily transformed into formats recognized by popular open-source tools like GMTSAR, ISCE, sarpy (SICD) and SNAP (BEAM-DIMAP). Our preliminary tests show that we are able to stage SAR metadata and SLC imagery in a numpy array for a single burst in about 3 seconds from SLCs (original zip format) in our Google Cloud buckets and about 7 seconds when pulling in imagery from NASA’s ASF archive using Earthdata authentication. Our burst indices and fast data access mechanism, can be leveraged by themselves for any large scale deployment of spatio-temporal analytics pipelines with Sentinel-1 data. With access to the SAR metadata and the imagery, we are able to calibrate, apply thermal noise corrections and resample the data onto a UTM grid (10m Northing x 2.5m Easting) while accounting for the slant range delay in phase. The resulting products exhibit coregistration properties that are better than those of ESA GRD products, and support detailed backscatter and change detection analysis. In addition, because phase is preserved in the geocoding process, these ARD products can be used for insar analysis by simple cross-multiplication. Our Sentinel-1 processing chain is highly scalable since processing of an individual burst is not very resource intensive and allows us to rapidly build InSAR-ready stacks of ARD products for time-series analysis of targeted sites. We will show some examples of how the use of geocoded bursts has enabled completely interactive deformation analysis within our jupyter-notebook environment for targeted sites. We are also exploring the feasibility of rolling out a single Radiometric Terrain Correction factor layer for backscatter normalization of stacks of geocoded bursts.