Abstract The intensification of extreme rainfall under climate change amplifies fluvial geohazards, creating a critical management paradox for riparian vegetation, which is essential for bank stability yet can increase flood risks. Conventional management, reliant on static hydraulic models, fails to capture the dynamic nature of these systems. Here we show a Geospatial Digital Twin (GDT) framework that integrates high-resolution unmanned aerial vehicle imagery with a deep learning model to create a living, high-fidelity representation of a river corridor. Applied to the Yi-dong stream, South Korea, our GDT accurately segmented seasonal vegetation changes and translated them into dynamic hydraulic roughness maps. By simulating distinct management scenarios, we quantified the trade-offs between flood mitigation and geomorphic stability, revealing that a balanced strategy involving willow planting enhances stability without dramatically increasing flood risk. By providing a virtual laboratory to test management outcomes before implementation, the GDT framework offers a powerful, scalable tool for developing climate-adaptive strategies.
Park et al. (Thu,) studied this question.