Responding to the flood emergency in Germany with the use of high tech from the space

Abstract

Nowadays, it is very often to see in the news around the world how wild rivers outside the control destroyed houses and flooded cities. This July in Germany, heavy rains caused floods and 182 casualties in North Rhine-Westphalia and Rhineland-Palatine. In this context, one of the first questions was, which villages were affected? Optical images only showed clouds. SAR (Synthetic Aperture Radar) images yielded poor results of flood detection. In this context, DLR took aerial photography of < 0.15 m resolution to perform flood damage analysis. The images were analysed by one operator by means of the ESRI deep learning framework. The flood damage training samples were obtained manually from 1.5 km² of image and the Mask-RCNN (Mask Region-Based - Convolutional Neural Network) model/algorithm was used to train them. The post-processed Mask R-CNN output were delivered in time to the federal institutions for their humanitarian actions of rescue and reconstruction. The results of the accuracy assessment methods will be published soon in a scientific paper.