Global scale InSAR analytics with Sentinel-1
Descartes Labs (DL) has built a global index of TOPS mode SLC bursts, allowing us to access any individual burst from Sentinel-1’s mission history in a few seconds without having to download complete SLC zip archives. Exploiting the scalability afforded by burst-based data access, DL has built two large scale pipelines for generating analysis ready InSAR products - one that generates global scale moderate resolution wrapped interferograms and another that generates a targeted full resolution stack of coregistered geocoded SLCs. We specifically note that DL has already processed the entire Sentinel-1 SLC archive (V-transmit) to populate the moderate resolution global InSAR product. DL’s moderate resolution global InSAR product consists of all interferometric pairs with a temporal baseline of less than 25 days. Burst-based wrapped interferograms - with coherence and phase bands, are generated using geocoded bursts [2] as an intermediate product and spatially averaged using a Gaussian filter to a posting of 20 m (resolution of ~80 m). The global InSAR product is primarily designed for - Global scale change detection applications for agriculture and sustainability-related applications with coherence, Deformation feasibility assessment of AOIs using coherence and Quicklook moderate resolution deformation time-series estimation using both phase and coherence. DL’s coregistered, geocoded SLC stacks are generated on demand over AOIs/ infrastructure of interest and are automatically updated as new images are released by ESA. Geocoded SLCs consist of two bands - real and imaginary, at a posting of 10m Northing and 2.5m Easting, and the phase of each pixel has already been adjusted for the image’s slant range to facilitate generation of flattened interferograms with simple band math operations. Pixel/point-cloud based InSAR algorithms [3] are used to derive deformation history from these full resolution stacks. In addition to deformation history, DL also uses scalable computing to estimate deformation uncertainties on a pixel-by-pixel basis using an ensemble-based approach [3]. Geocoded SLC stacks over new AOIs / facilities can be generated on demand in a few minutes using scalable computing. DL has also extended this pipeline to use data from other SAR sensors (e.g., TerraSAR-X and Cosmo Skymed). The most important feature of DL’s InSAR products are that they are accessible to the end user using the same API calls as other remote sensing datasets like Landsat and Sentinel-2, thus enabling deployment of machine learning and analytics tools on these interferometric datasets, just like other widely used datasets. This allows us to quickly iterate over interferometric datasets using a suite of analytics tools, without having to worry about customizing data access.