Improved Knowledge-Driven Flood Intelligent Monitoring (KDFIMv2): A Case Study of the 2024 Bangladesh Flood

Monitoring large-scale floods and tracking their evolution are essential for effective disaster response, particularly in regions where floods have widespread and dynamic impacts. Satellite-based flood detection using Synthetic Aperture Radar (SAR) and optical data faces challenges such as low spatial and temporal resolution, incomplete coverage, and cloud interference, which complicates the reliability of optical data. These issues hinder timely flood monitoring, which is critical for disaster management. This study introduces the Improved Knowledge-Driven Flood Intelligent Monitoring (KDFIMv2) method, which integrates SAR and optical data to improve flood monitoring by enhancing both spatial and temporal resolution.

The main challenge in large-scale flood monitoring is low spatiotemporal resolution, caused by limited SAR sensor coverage, low temporal observation frequency, and cloud interference affecting optical data. KDFIMv2 addresses these challenges through three key modules: 1) Surface Scattering Knowledge-Driven Flood Inundation Extraction, 2) Physical Knowledge-Driven Feature Fusion, and 3) Mathematical Knowledge-Driven Flood Information Extraction. The Surface Scattering Knowledge-Driven Flood Inundation Extraction module integrates SAR and optical data to extract flood information from satellite images. It tackles cloud cover and cloud shadows, which hinder water surface extraction in optical data, especially during floods. By combining SAR’s surface scattering capability with optical image spectral data, this module ensures accurate flood detection even under cloudy conditions. The Physical Knowledge-Driven Feature Fusion module improves adaptability by extending potential flood areas based on existing data. Using knowledge of flood dynamics, it infers the evolution of flood levels across a basin, filling gaps caused by cloud interference or incomplete satellite coverage, offering a more comprehensive flood monitoring solution. The Mathematical Knowledge-Driven Flood Information Extraction module uses mathematical models to calculate flood parameters such as depth, duration, and spread, providing a holistic assessment of the flood’s impact. This allows authorities to quantify flood disasters and track their evolution over time.

KDFIMv2 was applied to monitor floods in Bangladesh from June to December 2024. Results showed that KDFIMv2’s flood depth estimates had a mean error of only 0.1 meters, with 75% of the area within 0.2 meters and 95% within 0.5 meters. The method mitigated cloud cover and observational limitations, enabling flood tracking with a 30-meter resolution every two days. KDFIMv2 overcomes the limitations of current flood monitoring systems, offering high-accuracy flood evolution tracking. This study advances flood monitoring techniques and contributes to a better understanding of the impacts of floods on climate change adaptation and disaster resilience. By enhancing flood monitoring accuracy, KDFIMv2 plays a crucial role in reducing risks for vulnerable populations and contributes to achieving the United Nations Sustainable Development Goals (SDGs), particularly SDG 1 (No Poverty) and SDG 11 (Sustainable Cities and Communities).