Evaluation of a change detection approach to map global flood extents using Sentinel-1

Abstract: Floods are the most frequent disaster in the world and flood exposure is increasing globally. Flood mapping of past events can be a useful aid in not only disaster risk management, but also in evaluating and validating global flood models (GFMs) which are being used to assess and predict these floods. There are however numerous ways of mapping floods, and it is uncertain how well these perform as validation data. In this paper, a change detection approach based on a combination of synthetic aperture radar (SAR) and cloud-computing to map past flood events is evaluated (hereinafter CD- SAR). Eight flood events were chosen over a wide range of hydroclimatic conditions, regions, and flood types. These eight events were mapped with CD-SAR and compared to four GFM outputs and two flood maps of past events from two commonly used databases. The spatial agreement between CD-SAR and the GFMs showed a considerable variation between regions and models. The agreement was however shown to share a similar interval to previous validation studies, albeit in the lower portion. CD-SAR also showed similar performance to the comparison between the GFMs and the outputs from the databases of mapped past flood events. The results were also analysed for how the flood extent and flood-edge distribution of CD-SAR compare to both the GFMs and the database outputs. The results showed a similar variation in distribution as the spatial agreement but did not follow the same trend for all regions. The flexibility and high resolution of CD-SAR allow it to cover events over a wider range of regions and of any size and it can be a viable tool to map past flood events and could be used to evaluate GFMs. However, CD-SAR needs further evaluation as uncertainties still exist due to the inherent characteristics of SAR and the revisit times of the satellites using SAR.