Multitemporal Flood Damage and Recovery Assessment of the Amnok River Using SAR and Optical Satellite Imagery with a Deep Learning Model

Floods pose persistent risks across the Korean Peninsula, with impacts particularly severe in North Korea due to limited disaster-response capacity.Rapid and systematic inundation mapping using satellite imagery is therefore essential for assessing flood impacts in such information-restricted and inaccessible environments.This study presents a multitemporal analysis of the 2024 Amnok River flood by integrating Sentinel-1 synthetic aperture radar (SAR) and Sentinel-2 optical imagery with a deep learning (DL)-based flood mapping framework.First, a multimodal semantic segmentation model was trained on an open-source flood mapping dataset and subsequently applied to multitemporal Amnok River satellite imagery to derive a time series of water-extent maps.By aggregating pre-flood water extents, flood-induced changes were detected and tracked across the event timeline, enabling a detailed reconstruction of spatial and temporal inundation dynamics.The earliest post-event observations revealed substantial river expansion and widespread flooding along low-lying areas, followed by rapid hydrological recession within the first month.However, pronounced cross-border differences were observed.North Korea experienced a maximum inundated area nearly 1.7 times larger than China and retained flooded zones for a longer duration.Integration with pre-flood land-cover classifications further revealed classspecific impacts, with agricultural land constituting the majority of inundation in North Korea, whereas built-up areas were most affected in China.Temporal patterns indicated slower recession in agricultural and grass-covered areas, reflecting prolonged water retention and elevated soil moisture conditions.Overall, this study demonstrates the value of combining multitemporal satellite observations with DL-based mapping to characterize flood progression in data-scarce regions.The proposed framework provides a reproducible basis for large-scale and temporally consistent flood monitoring, offering practical utility for future disaster preparedness and response.