Integration of remote sensing and crowdsourced data for fine-grained urban flood detection

In the context of frequent global flood disasters, flood detection is of great significance for emergency management and human sustainable development, especially in urban areas with increasing population and socio-economic activities. However, there are similar reflection/scattering characteristics between flooded and non-flooded land use and land cover (LULC) classes in complex urban environments, which limits the accurate detection of floods. In this study, we develop a new method for fine-grained and accurate flood detection by integrating multi-temporal Sentinel-1 synthetic aperture radar (SAR) images, OpenStreetMap (OSM) data, and convolutional neural networks (CNNs). We take the 2017 Houston flood event as a test case, where the study areas are divided into 6 fine-grained LULC classes, i.e., residential areas, service areas, main roads, forest, grassland, and waterways. Based on the information of finegrained LULC classification, the proposed method performs more prominently than the baseline methods for urban flood detection. Specifically, compared with such baseline methods, F1 score, overall accuracy (OA) and Kappa increase by more than 3.96%, 4.53%, and 9.26%, respectively. The integration of remote sensing and crowdsourced data provides a new perspective for flood detection in complex urban environments, thus supporting emergency management.