2:20 PM - 2:40 PM
[4K3-J-13-02] Estimation of Flood Inundation Area with Conditional Generative Adversarial Networks
Keywords:pix2pix, disaster prevention, GAN
In the event of a flood disaster, grasping the inundated area is important for appropriate evacuation behavior and disaster prevention activities. A flood hazard map has been released nationwide, and the range of assumed inundation is shown based on preliminary simulation. However, no technique has been established for grasping the inundation range in real time in the event of a disaster. In this research, in order to instantaneously estimate the inundation range from the observation information, we constructed the inundation area estimation model with Conditional Generative Adversarial Networks and verified the accuracy. Specific procedures of modeling are as follows.
1: Implemented physically based flood simulation under various flooding scenarios (levee breakdown location and inundation scale).
2: Pseudo observation data of inundation was prepared by randomly extracting the inundation depth from the calculation mesh of the inundation simulation.
3: Developed the predictor that estimates the inundation depth distribution from imaged inundation observation information with pix2pix.
With the constructed model, the reproducibility of the inundation simulation results was tested for the Arakawa Delta area. From several tens to 100 of the flood observation information, the inundation range could be reasonably estimated.
1: Implemented physically based flood simulation under various flooding scenarios (levee breakdown location and inundation scale).
2: Pseudo observation data of inundation was prepared by randomly extracting the inundation depth from the calculation mesh of the inundation simulation.
3: Developed the predictor that estimates the inundation depth distribution from imaged inundation observation information with pix2pix.
With the constructed model, the reproducibility of the inundation simulation results was tested for the Arakawa Delta area. From several tens to 100 of the flood observation information, the inundation range could be reasonably estimated.