AI-driven digital twinning quantifies the trade-off between flood conveyance and riparian stability under climate extremes

Anthropogenic climate change amplifies hydrological extremes, challenging flood models reliant on static hydraulic assumptions. Riparian vegetation is essential for ecological resilience and bank stability but is often removed to maximize flood conveyance. To quantify this trade-off, we introduce a Scenario-Based Cognitive Digital Twin framework integrating UAV sensing, AI-driven segmentation (Residual Attention U-Net), and dynamic roughness parameterisation for a single ~ 10 m reach. Applied to a temperate headwater stream, we show static models underestimate growing-season flow resistance by up to 64% in the reach-averaged Manning coefficient, corresponding to an approximately 39% reduction in roughness-controlled conveyance. Through physics-informed simulation validated against in-situ observations (n = 179; RMSE = 0.0087 m s −1 against the wavelet-denoised trend; NSE = 0.783), we evaluated geohazard trade-offs under a 3× extreme-discharge stress-test. Results indicate full vegetation clearance carries an elevated erosion-risk profile, reducing the modelled bank-stability proxy by approximately 50%. Conversely, a nature-based selective intervention increases the stability proxy by 20% with only ~ 1% conveyance loss. By coupling AI-derived structure with dynamic hydraulics, this cognitive twin implements geomorphologically constrained defence strategies to support climate-adaptive river management at the reach scale.