Flood detection from multi-temporal SAR data using harmonic analysis and change detection

Abstract

Flood mapping from Synthetic Aperture Radar (SAR) data has attracted considerable attention in recent years. Most available algorithms typically focus on single-image techniques which do not take into account the backscatter signature of a land surface under non-flooded conditions. In this study, harmonic analysis of a multi-temporal time series of >500 ENVISAT Advanced SAR (ASAR) scenes with a spatial resolution of 150m was used to characterise the seasonality in backscatter under non-flooded conditions. Pixels which were inundated during a large-scale flood event during the summer 2007 floods of the River Severn (United Kingdom) showed strong deviations from normal seasonal behaviour as inferred from the harmonic model. The residuals were classified by means of an automatic threshold optimisation algorithm after masking out areas which are unlikely to be flooded using a topography-derived index. The results were validated against a reference dataset derived from high-resolution airborne imagery. For the water class, accuracies > 80% were found for non-urban land uses. A slight underestimation of the reference flood extent can be seen, mostly due to the lower spatial resolution of the ASAR imagery. Finally, an outlook for the proposed algorithm is given in the light of the Sentinel-1 mission.