Satellite radar time series for flood detection using SAR Sentinel-1 data

Abstract

All countries are affected by natural hazards in different ways. The recently launched Synthetic Aperture Radar (SAR) Sentinel-1 A and B twin satellites provides large amounts of high-resolution radar data due to their, at most frequent, 6th day revisit time globally. The higher revisit times and data accuracies of the Senitnel-1 provides more information and knowledge which have been used by researchers for assessing the damage and extent of natural hazards. In this thesis, satellite time series have been created to investigate a possible improved flood extent detection using SAR Sentinel-1 data compared to using single radar products. Flooded areas were detected from low backscatter values and were found to range between -16 to -21 dB for the Norwegian study areas (Gudbrandsdalen and Birkeland) and between -20 and -25 dB for the study area located in India (Kerala). The study was conducted by processing VV polarized Sentinel-1 data for three different study areas before creating image stacks. False color composites were created to detect areas of change between the processed single and/or stacked products. The selected reference products to be composited were assigned the red band and the flood products the green and blue bands for all created composites. Areas of change were compared to derived Normalized Difference Water Index (NDWI), Height Above Nearest Drainage (HAND) and Change Detection and Thresholding (CDAT) calculations in addition to visual findings using optical Sentinel-2 and PlanetScope data. For the Gudbrandsdalen study area, the results showed an increased flood extent detected in the false color composites created using a single flood and a reference stacked products. The findings were supported by optical data, derived NDWI and HAND index calculations. By the work of this thesis, calculations derived from NDWI and CDAT proved to be useful tools for flood extent estimations and visual comparison against the created false color composites and optical findings. The HAND index was successfully used to detect features such as radar shadow in addition to investigating areas of likely inundation. The study successfully created an improved flood mapping method using satellite radar time series for assessing flood extents.