Design discharges corresponding to large return periods are generally highly uncertain because of the relatively short data set of measured discharges. The uncertainty in these discharge predictions can be decreased by extending the data set of annual maximum discharges with historic flood events. However, efficient model approaches, in terms of computational times and model accuracy, are required to reconstruct historic flood events. In this study, the suitability of three data-driven surrogate models are evaluated, namely: a linear regression model, an artificial neural network and a support vector machine. The Rhine river flood event of 1809 is used as a case study. Although all types of surrogate models are capable of reproducing the maximum discharge during the 1809 flood event, the use of an artificial neural network resulted in the smallest 95% uncertainty interval.@en

Structures integrated in a grass-covered dike may increase erosion development. Currently, safety assessment methods for flood defences are only applicable for a conventional grass-covered dike and the effects of structures on dike cover erosion are poorly understood. Since many dikes have a road on top, it is important to study the effect of such a road structure on erosion onset during wave overtopping. To investigate this effect, a coupled hydrodynamic–eroison model was developed. The erosion onset caused by overtopping waves was predicted by combining the time-varying bed shear stresses from the hydrodynamic model with a depth-dependent erosion model. The results show that roads on top of a dike increase the erosion of the neighbouring grass cover. This increase in erosion may have a negative impact on dike stability. Therefore, we recommend considering effects of constructions on top of dike profiles during safety assessments. Explicitly, consideration of the roughness transitions in the safety assessments of dikes is recommended.

@en

The present study aimed to include the turbulence effects derived from a road located over the crest of a dike, in its probabilistic safety assessment. This was done by building two different computation fluid dynamics models (RANS K-); one of a dike with a road on top and one without it. Both models were validated with experimental data collected from the Wave overtopping simulator experiments performed in the Netherlands. These models were used to produce training data sets which were later used for constructing emulators (computationally cheaper models) which allowed to reduce the computational burden from the required stochastic modelling. These new emulators allowed to calculate bottom shear stress time series in different locations along the dike profile. With these time series, it is possible to estimate the potential scouring depth per wave volume routed. These emulators allowed to model different probabilistic overtopping scenarios without running the CFD models again. The results showed that when assessed under extreme climate scenarios, the presence of a road may reduce the dike safety by almost 50% with respect to the case where no road is present. In addition, it is also concluded that the spatial grass quality distribution is a more important factor for determining dike safety than the spatial grass cover thickness. @en