Knowledge-Driven Flood Intelligent Monitoring (KDFIM) Method: Analyzing the Kakhovka Dam Destruction Incident

Active microwave remote sensing data, such as Sentinel-1 synthetic aperture radar (SAR), are indispensable for flood monitoring and emergency response due to their all-weather imaging capabilities of the Earth’s surface and global coverage. Nevertheless, flood monitoring based on SAR still faces significant challenges stemming from the inherent noise and interference present in the background, which further complicates the dynamic and urgent nature of flood events. To address these challenges, this study utilizes Sentinel-1 SAR data, complemented by Sentinel-2 multispectral data and digital elevation model (DEM), to construct a knowledge-driven flood intelligent monitoring method (KDFIM). First, KDFIM integrates both satellite imaging knowledge and ground object scattering knowledge to construct a spatiotemporal SAR feature fusion module, which is designed to mitigate the noise and interference present in the background and to extract multispectral SAR features for flood detection. Subsequently, leveraging ground object spectral feature knowledge, a flood knowledge implementation module is constructed to facilitate the adaptive extraction of flooding extent. Finally, based on the physical understanding of water bodies, a knowledge-driven multiple parameters calculation module is developed to enhance the three-dimensional dynamic flood analysis. The KDFIM was validated using the flood event triggered by the destruction of the Kakhovka Dam, achieving an inundation extraction accuracy of 98.46 ± 0.39% and a Kappa coefficient of 0.9691 ± 0.08. The knowledge embedded in the KDFIM is replicable and transferable, helping to reduce risks associated with water-related disasters and ultimately achieve the goal of building resilience among impoverished and vulnerable populations, thereby reducing their exposure and susceptibility to climate-related extreme events.