Reference Water Mapping based on Sentinel-1 and -2 data

Abstract

Flood events are one of the most dangerous natural hazards on Earth. By using instruments of Earth Observation, information about the actual water extent and the event’s process can be provided. Subsequently, a targeted rescue of victims can be supported and the potential damages can be reduced. An important task is to determine the event’s true surface water extent and to distinguish between areas which are affected by an event and those which are generally inundated. Therefore observations obtained during a flood event are compared to reference information which represents the open surface water bodies under normal conditions. Uncertainties within the reference information could directly distort the result of flood detection. In this thesis multiple methods for defining valuable reference information are observed and compared. One focus lies on the diverging properties between surface water detection based on optical (Sentinel-2) or radar observations (Sentinel-1). Therefore Sentinel-1 and Sentinel-2 water maps were processed by fully automatic processing chains developed by the German Aerospace Centre’s (DLR) research group ’Natural Hazards’. A flexible work-flow for defining reference information from time-series of water maps is presented. As a result diverging properties of reference information calculated from different sensors, time periods and other adaptions of the work-flow are presented. By the use of these findings the thesis should support potential use of satellite based reference water mapping. The aim is to improve the understanding of characteristics of certain parameters like the used sensors or the considered period of time. Finally, an example for the use of sensor specific differences in flood detection is given. All results are tested in Bihar (India) where temporal changes of the surface water areas appear quite often. The second test region is the region surrounding Zaragoza (Spain) where the water areas are hardly affected by temporal changes.