[A-OS11] Continental Oceanic Mutual Interaction - Planetary Scale Material Circulation

[AOS11-09] Deep learning Method of Flood Area Segmentation with SAR Satellite Data and Possibilities of Its Applications to Planetary Science

*Naruo Kanemoto¹, Kentaro Tanaka¹, Shota Uchida¹, Shusuke Yasui¹ (1.Space Shift, Inc.)

Keywords:Synthetic Aperture Radar (SAR), Flood Detection, Deep Learning, Remote Sensing

Floods are known as one of the greatest natural disasters because of the enormous damage they cause to people's lives and property. Modern deep learning techniques have the potential to accurately predict the extent of flooding and depth of the water over a wide area in near real-time. In this paper, we introduce the method to detect flood areas from a pair of SAR images acquired in the same area at two different times. We add interferometric coherence and SAR intensity to the training set of our deep learning network to give information on surface displacement between the two time periods of a pair of SAR image. The proposed method is evaluated by deep learning models trained by the datasets with and without coherence images. The results showed that the use of coherence image improved the accuracy of flood mapping. For inundation depth estimation, a combination of DEM and Ground Truth inundation depths was used for learning, enabling highly accurate estimation. This technique of water area estimation using SAR satellite data is also effective for hydrological understanding of other planets, and we will discuss its effectiveness for maintaining civilization when human civilization expands to outer space in the future.