Use of Sentinel-1 SAR data in assessing the accuracy of LISFLOOD-FP in modelling (compound) flooding in estuaries

Estuaries are among most vulnerable parts of our planet in terms of flood risk because they are constantly exposed to flood sources from at least two directions. While different flood modelling tools are helping us to be better prepared for flood events, state-of-the-art space technologies are providing useful high resolution (temporal and spatial) data to quantify and monitor real flood impacts and consequences, regardless of night and cloudiness.

In this study we are: a) applying LISFLOOD-FP hydrodynamic model for modelling compound flood event in the Dyfi estuary, Wales (UK) and b) using Sentinel-1 SAR data to map flood event from the same period and to validate flood inundation model. The selected flood event (> 300 m³/s and around 70 cm surge) caused large flooding along the estuary, particularly in the upper parts. Modelled results are shown as water surface elevation and water depth classified raster maps, which are used later for comparison with the SAR image.

Raw Sentinel-1 SAR (GRD) image downloaded from Copernicus Browser had to be pre-processed in ArcGIS PRO (The Synthetic Aperture Radar toolset) to remove unwanted noise and correct distortions. Further, RGB color composite was produced from the calibrated SAR image and used for extracting water bodies/flooded areas. It was achieved by applying deep learning tools integrated in ArcGIS PRO which classify pixels (wet/dry) based on a previously trained sample. Resulting raster was then compared with the modelled flood extent, quantifying the differences. Matching was very good in the upper parts where major flooding was recorded (> 80 % agreement), while the model was slightly less accurate in the lower estuary and along the salt marsh zone due to larger DEM uncertainty in those areas. However, when selected shallow areas (e.g. 0-1 cm or 0-2 cm class) were removed from the model, matching between modelled and observed flood extent was higher.