A Physics-Guided Neural Network for Flooding Area Detection Using SAR Imagery and Local River Gauge Observations

The flooding extent area in a river valley is related and proportional to river gauge observations. The higher the water elevation, the larger the flooding area. That relationship is used for the water elevation prediction based on satellite images. Whereas in this study, we apply it conversely. We aim to provide a method that maps water extent on the image without supervision and to evaluate its efficiency depending on the water elevation in the river, which is not common in other studies. To achieve this, we propose a physics-guided neural network for flooding area detection. Our approach takes the Sentinel-1 time-series images and the corresponding water elevations in the river as input data. We apply the Pearson correlation coefficient between the predicted sum of water extent areas and the local water level observations of river water elevations as the loss function. The effectiveness of our method is evaluated in five different study areas by comparing the predicted water maps with reference water maps obtained from digital terrain models and optical satellite images. The highest Intersection over Union score achieved by our models was 0.89 for the water class and 0.96 for the non-water class. Additionally, we compared the results with other unsupervised methods, such as Otsu thresholding, Chan-Vese segmentation, gaussian mixture model, and spectral clustering. The proposed approach provided a higher IoU than the benchmarking methods, especially for SAR images registered during low water elevation in the river.