Locating flood embankments using SAR time series: A proof of concept

Abstract

River embankments limit flood risk, alter the flood peak, and change sediment deposition rates, but in many areas of the world embankment locations are not available in geodatabases. This lack of information reduces the accuracy of predicted flood extent within inundation models. The objective is therefore to determine the positional accuracy of embankments using time series of Sentinel-1 Synthetic Aperture Radar (SAR) data with local gauge data. A new method is proposed to: (1) establish the relationship between stage and extent for river cross sections; with stage obtained from gauges and flood extent from SAR-derived flood images, and (2) infer embankment location from plateaus in these plots. The predicted embankment location was compared with airborne LiDAR data for two case studies: the One Hundred Foot Wash (UK) and the Yolo Bypass (Sacramento, USA). The median absolute error for cross sections located in low lying floodplains with low vegetation was 59 m and 110 m, respectively. This increased to 78 m and 197 m when vegetated and cross sections with high ground were included. Application of the method could provide key flood defence information to update geodatabases for locations where data are missing or not widely/freely available, and to improve inundation models up to global scales.