Storm-driven erosion and morphological change on the nearshore-beach-dune continuum poses a significant threat to coastal infrastructure. Therefore, accurate and timely storm-impact prediction is essential for coastal management and infrastructure protection. 

This study presents a hybrid modelling framework combining the physics-based model XBeach with a deep neural network emulator (CNN-ConvLSTM) to predict spatio-temporal bed-level change. The emulator was trained on XBeach simulations and validated against four historical storm events of varying intensity. 

The emulator achieves strong cross-storm generalisation with R² = 0.70–0.88 and RMSE ≈ 0.03–0.05 m across storms spanning minor to severe conditions in the case study area, namely the Norderney island as part of the southern North Sea coast. Computational effort is reduced by several orders of magnitude relative to direct XBeach simulations, enabling rapid ensemble-based storm scenario assessment. The proposed framework offers a practical tool for real-time coastal risk management and infrastructure protection planning along the German Wadden Sea coast.