Pre-processing and ancillary data – Sentinel-1 IW GRDH Vertical-Vertical polarization data is ingested and pre-processed using the Sentinel Application Platform (SNAP). The pre-processed analysis-ready data is 20 m
sigma nought calibrated backscatter. The use of the Copernicus DEM for geocoding requires a transformation from orthometric height to ellipsoidal height. This incorporates geoid undulations from Earth Gravitational Models.
The Sentinel-1 Global Backscatter Model (S1-GBM) is used to determine areas where flood detection is improbable because of surface structures on radar return signal. A near-global Height Above Nearest Drainage (HAND) index
is generated to allow for the exclusion of areas where flooding is unlikely due to the hydrologic-topographic conditions. The affected population of a flood is derived from the GHSL-POP layer of the GHSL dataset. The GHSL and
the WSF 2015 datasets are used to determine where flooding should be ignored due to the likelihood of artefacts introduced by interactions between SAR signals and built-up areas. Th3 Global forest change dataset is used for
gap-filling in areas where S1-GBM is limited by temporal coverage.
Flood detection – Three flood detection algorithms are applied to the Sentinel-1 data to derive the observed flood extent layer of this product. The first is a change detection algorithm which requires as input the
current Sentinel-1 scene, a previous overlapping scene from the same orbit, and the previous flood extent map. The second is a fuzzy logic-based approach using region growing. Inputs are the Sentinel-1 scene, a DEM, a dataset
of areas where flooding is not prone and a reference water layer. The third is a probabilistic approach taking in the Sentinel-1 scene, a projected local-incidence layer and harmonic model parameters for seasonal backscatter
variation simulation. Pixels are assessed for flooding probability. An ensemble flood mapping algorithm uses the output of all three previous algorithms to decide if each pixel is classed as flooded or not. If 2 out of 3 algorithms
determine flooding, then a pixel is classed as flooded.
The output layers of this product are listed here. It should be noted that many further processing steps are required to generate some of these layers and details of these can be found on the product site (
Welcome to GFM | EODC Public Wiki).
Output layers:
- Observed flood extent (GeoTIFF, shapefile)
- Observed water extent (GeoTIFF, shapefile)
- Reference water mask (GeoTIFF, shapefile)
- Exclusion mask (GeoTIFF)
- Uncertainty values (GeoTIFF)
- Advisory flags (GeoTIFF)
- S1 metadata (KML)
- S1 footprint (KML)
- S1 schedule (KML)
- Affected population (GeoTIFF)
- Affected land cover (GeoTIFF)
