The STAC and Dask based solution for flood mapping allows cloud computing in close proximity to the data thereby freeing the user of the burden of downloading sentinel-1 radar images while using the scalable resources of the host for processing. In combination with the intuitive API, this solution is an independent and reusable module that can be easily integrated in existing workflows. Furthermore, the open-source Python package allows expert users to adapt the flood mapping workflow to their own insights and demands. This sets it apart from the current Global Flood Monitoring implementation, which provides only statically produced flood maps.
Map floods with Sentinel-1 radar images. We replicate in this open source Python package the work of Bauer-Marschallinger et al. (2022) on the TU Wien Bayesian-based flood mapping algorithm. This implementation is entirely based on dask and data access via STAC with odc-stac. The algorithm requires three pre-processed input datasets stored and accessible via STAC at the Earth Observation Data Centre For Water Resources Monitoring (EODC).
- Sentinel-1 Sigma Nought (SIG0) Backscatter in 20 meter sampling distance: The SIG0 product is radiometrically calibrated and georeferenced backscatter data generated from Sentinel-1 level-1 Interferometric Wide (IW) Swath Ground Range Detected (GRD) High resolution products using TUWien Sentinel-1 preprocesssing workflow. The preprocessing workflow includes the following steps: applying precise orbit file, radiometric calibration, thermal noise removal, and range doppler terrain correction.
- Sentinel-1 harmonic parameters (HPAR): HPAR describes typical seasonal Sentinel-1 backscatter variations on a 20 m sampling distance. The harmonic oscillators , sine and cosine functions, are used to describe periodicities in the time series of backscatter data. This helps discern noise (e.g. speckle) or transient events (e.g. floods) from long-term trends of seasonal changes brought about by varying water content, e.g., inundation or soil moisture, and progression of vegetation structure.
- Sentinel-1 Mean Projected Local Incidence Angle (PLIA): PLIA is the angle between the surface normal and the looking direction of the satellite (local incidence angle – LIA) which is further projected into the range plane. It can be computed for each individual Sentinel-1 acquisition during the terrain correction using SNAP Sentinel-1 Toolbox.
It is foreseen that future implementations can also use data from other STAC catalogues. The latter entails development of dynamic harmonic parameter estimation in favor of the static HPAR dataset. Furthermore, the Dask implementation allows remote processing of the data at the EODC with the aid of a Dask Gateway. This has the added benefit that one can process in close proximity of the data without rate-limiting file transfers over the internet. The output data consists of a Bayesian decision of flood (1) or no flood (0) per pixel, or the Bayesian posterior probability of flood per pixel.