Surrogate modelling for uncertainty analysis of dike breach floods

Dike breaches are sudden and severe events that need to be predicted and managed for public safety. Important flood characteristics such as flood arrival times are required for effective decision making. Current tools, such as hydrodynamic models, offer this information at long simulation times to be used during an emergency situation. Surrogate models leave out complex physical descriptions of the system in favour of simplified hydraulic principles. In this study, we couple a conceptual breach outflow hydrograph model with a fast flood arrival time model. The low computation times of the coupled framework allow for 10.000 simulations to be conducted in around 5 minutes. We investigate the effect of two sources of uncertainty on the breach outflow and on the flood arrival times: the uncertain moment of dike failure due to wave overtopping, and the uncertain river water level forecast with a number of days lead-time. The results illustrate that the models effectively compute the breach outflow and flood arrival times, and that the river water level causes greater uncertainty than the wave overtopping failure mechanism. We further conclude that the method shows the potential of surrogate modelling frameworks for flood emergency decision making thanks to the low data demand and computational cost.