Flood monitoring based on multi-temporal Sentinel-1 data - a synergistic approach of amplitude data with interferometric coherence

Abstract

In the present study, multi-temporal Sentinel-1 data from different regions where severe flood events were indicated are examined. Based on an existing operational algorithm for flood classification and extraction, an improvement using interferometric coherence information is investigated. Using interferometric pairs of complex SAR data complementary to the time series data of the crisis event, information about the flooded vegetation and the water areas could be retrieved. The approach implemented in the existing flood service is limited to the use of amplitude information along with auxiliary data sources, such as Digital Elevation Models (DEM) and permanent water area masks. In the proposed approach, coherence maps with a temporal resolution of 12 days are considered as an ancillary source of information that could significantly improve misclassifications over water lookalike areas or croplands. Due to the randomness of the interferometric phase over the water surfaces the presented interferometric coherence values are significantly low. Furthermore, due to the consistency of the combined Synthetic Aperture Radar complex image pair information, coherence values variations seem to be less affected from wind variations over water areas or vegetation, where the amplitude values could be significantly altered. Based on this approach, a fuzzy logic classification method is implemented for region with different characteristics, in order to examine the coherence contribution to the classification results. Two different datasets from Evros River in Greece, in the years 2014-2015 and Shannon River in Ireland, in the years 2015-2016, are examined and presented in this Master thesis. A time series analysis and statistical evaluation is provided for the flood events and flood delineation maps are produced based on the near real time Sentinel-1 flood service (S1-FS), implemented on Remote Sensing Data Center (DFD). In order to obtain coherence maps with satisfactory spatial resolution a refined methodology is adopted and a processing chain for integration of interferometric coherence information in the existing flood service is produced. Subsequently, a comparison between the different approaches is conducted, in order to determine the level of contribution of the interferometric coherence information on the final flood classification results. Main products are classification maps, where different scenarios are examined. Overall, with the integration of interferometric coherence data, significant improvements are obtained, especially regarding the water lookalike areas over agricultural and urban areas.