A methodology for mapping annual flood extent using multi-temporal Sentinel-1 imagery

A novel automated approach for mapping nonurban flood extents using satellite-based Sentinel-1C-band SAR data is presented and applied in the Republic of Ireland. The methodology mapped the maximum flood extent of over 25,000 detected water bodies across Ireland over a five-year period (2016–2021). Flood extents were classified using a semi-automatic tile-based histogram thresholding approach and refined using a series of post processing filters. These included multi-temporal filters which exploited the timing and orbit position of the Sentinel-1 satellite to identify likely false positives, and contextual information based filters which used external information such as topography and land use to estimate the likelihood of flooding and constrain its extent. A topographic correction algorithm was also applied to hydrologically-correct flood bodies. The methodology produced a user accuracy of above 0.95 for each annual map when compared to observed flood level data, overall accuracy values above 0.97 and kp values above 0.7 when compared to mapping process based on Sentinel-2 validation imagery, and the capacity to detect over 90% of recorded permanent water bodies larger than 2 ha. As an additional step, confidence information was provided for each flood polygon based on the quantity of consistently flooded pixels used in its definition. The mapping process represents a stable and systematic nonurban flood mapping methodology that enables inter-annual comparison of flood extent at gauged and ungauged sites.