Flood mapping using Sentinel-1 data

Flooding in the lower Shannon river basin, December 2015. Analysed Sentinel-1 image.

The European Space Agency’s Sentinels Scientific Data Hub is an invaluable repository of earth observation data. Many GIS users are already familiar with multispectral satellite data, such as that provided by Landsat, and the opportunities for analysis which it provides. The ESA are now also contributing to the ever growing catalogue of freely available, multispectral data with their Sentinel-2 products.

However Ireland’s climate and northerly latitude highlights some major drawback of multispectral imagery and anyone hoping to evaluate the extent of flooding on the west coast during the recent storms will have realised the extent of the problem. Storm events tend to occur in the winter months, when there are long hours of darkness and generally involve almost complete cloud coverage, limiting the usefulness of multispectral data. Data acquired from satellite based Synthetic Aperture Radar systems, such as Sentinel-1, do not have the same drawbacks. As SAR does not depend on visible light, can penetrate cloud coverage, and has very low return from water bodies it makes it the ideal tool for mapping flood extents. In a radar image the delineation between water and non-water can be clearly delineated.

An example of this can be seen in the image below, which was produced from Sentinel-1 data, acquired on December 13th 2015. It shows an area north of Limerick city and the River Shannon and Head-race Canal are both clearly visible. In order to produce flood extents, which are compatible with existing GIS data, the raw Sentinel-1 dataset must first be pre-processed. The processing techniques include calibrating the pixel values and speckle filtering the data. The dataset must also undergo terrain correction to correct for effects such as foreshortening, layover, and shadow. Following this it can be transformed from ground range geometry to a specific coordinate reference system such as WGS84 or Irish Transverse Mercator.

After the pre-processing stage a histogram of backscatter coefficients can be generated and is used to help determine a value which most accurately reflects the threshold between water and non-water. Finally, the resulting binary raster image is converted into a vector dataset for analysis with existing datasets.

Mapping of this nature has obvious applications in flood risk management and planning. It can assist engineers and planners make more informed decisions and so help reduce the risk from future floods. In addition the fast turnaround possible between the data being acquired to the flood extents being produced means these maps could assist emergency services during a flood event.