Development and Validation of a Flood Depth Mapping System using SAR-derived Water Mask

Abstract:

Floods are a common disaster that can be triggered by hydro-meteorological hazards such as hurricanes, tropical cyclones, and atmospheric rivers. With the recent proliferation of synthetic aperture radar (SAR) imagery for flood mapping due to its all-weather imaging capability, the opportunities to detect flood extents are growing compared to using only optical imagery. Flood extent mapping algorithms can be considered mature, but flood depth mapping is still an active area of research. Depth estimation is essential to assess the damage caused by a flood and its impact to infrastructure. In this study, we introduce prototype algorithms for flood depth mapping based on two different approaches. The first approach estimates the water height by subtracting the DEM from the highest inundated elevation. Once a water mask is produced, water surface elevation is generated by interpolating the DEM values located along the floodplain boundaries. After testing various interpolation methods, two dimensional Gaussian convolution kernels implemented were found to have the least amount of artifacts. The second approach estimates the flood depth based on Height Above the Nearest Drainage (HAND) algorithm. In this case the water height is adaptively set to two standard deviations above the mean elevation of each flood plain. A case study over Bangladesh related to flood season in 2017 showed promising results using HAND method to obtain the flood depth map. Quantitative accuracy assessment was conducted using daily observations at the water level monitoring stations operated by the Flood Forecasting and Warning Centre. Water levels derived using our approach resulted in 1.4m of RMS error when compared to the reference data. The algorithm is further developed to make it operational for which several different case studies are planned to assess its accuracy and viability. Ultimately, our goal is to develop a software which will improve our end-user partners decision-making workflows.