Enabling seamless spatiotemporal flood monitoring via bridging observational gaps with a case study on the 2024 Bangladesh flood

Accurate and timely flood monitoring is critical for effective disaster management. Remote sensing observations offer large scale monitoring; however, they frequently suffer from observational gaps caused by cloud cover, limited swaths, and extended revisit cycles, severely hindering continuous tracking. To overcome these physical barriers, this study proposes an enhanced Knowledge-Driven Flood Intelligent Monitoring framework (KDFIMv2). Rather than relying on isolated imagery, KDFIMv2 proposes a synergistic architecture that integrates optical and SAR observations to maximize single day inundation extraction, employs topographic routing constraints for spatiotemporal gap filling, and incorporates dynamic water surface elevation calculations for three-dimensional parameter retrieval. The framework was evaluated using the 2024 Bangladesh flood. Quantitatively, KDFIMv2 achieves high depth estimation precision with prediction errors predominantly within 0.1 m. Crucially, benchmarking experiments assessing the spatial distributions of flood inundation demonstrate that KDFIMv2 maintains high classification performance across rising, peak, and recession stages, achieving a mean Overall Accuracy and F1 metric exceeding 0.9. While baseline methods exhibit severe stability fluctuations under observational gaps, spatiotemporal analysis reveals that KDFIMv2 achieves superior temporal consistency, with the proportion of stable pixels exceeding 70 %. The results highlight the potential of KDFIMv2 for near-real-time flood monitoring and disaster response, offering a scalable and interpretable solution for flood risk assessment and management.