Coastal morphodynamic emulator for early warning short-term forecasts

The use of numerical models to anticipate the effects of floods and storms in coastal regions is essential to mitigate the damages of these natural disasters. However, local studies require high spatial and temporal resolution numerical models, limiting their use due to the involved high computational costs. This constraint becomes even more critical when these models are used for real-time monitoring and warning systems. Therefore, the objective of this paper was to reduce the computational time of coastal morphodynamic models simulations by implementing a deep learning emulator. The emulator performance was evaluated using different scenarios run with the XBeach software, which considered different grid resolutions and the effects of a storm event in the morphodynamic patterns around a breakwater and a groin. The morphodynamic simulation time was reduced by 23%, and it was identified that the major restriction to reducing the computational cost was the hydrodynamic numerical model simulation. [Display omitted] •The implemented emulator had a high performance in simulating different complex coastal environments.•The transposed convolution layers improved the emulator performance.•A morphodynamic model emulator was developed, reducing the simulation time by 23%.•This emulator, using different input and output numerical model grid resolutions, reduced the simulation time by 86%.•Hydro-morphodynamic results from other numerical models can be used under the proposed methodology.