Low lying coastal areas have always been attractive for people to live, but are also prone to flooding. In The Netherlands, half of the population lives in the coastal area below mean sea level where two-thirds of the economic value is located. Coastal dunes protect the hinterland from flooding as a primary sea defence along the major part of the Dutch coastline. The envisaged protection level of the densely populated and economically valuable areas is one of the world's highest, with a normative failure probability of O(1×10⁻⁵ year⁻¹). The extreme storm events of this order of magnitude are not observed in (known) recent history. Hence, design, evaluation and maintenance of flood defence systems that can resist these extreme normative conditions rely on models and data extrapolations.

The safety assessment method for the Dutch dune coast includes an empirical dune erosion model (DUROS+) and boundary conditions that represent the normative loading (semi-probabilistic). Every six years, a safety assessment is performed for a series of cross-shore dune profiles along the coast. The prediction skill of the DUROS+ model is limited due to its empirical nature and the omission of the alongshore dimension. The model only takes the upper part of the cross-shore profile explicitly into account and does not account for effects of wave obliquity and coastal curvature. Several dune erosion models have been developed over the last two decades that provide the means to model the dune erosion process in a more comprehensive way. In addition, probabilistic methods have improved and computational power has increased, potentially allowinga probabilistic safety assessment of complex dune areas to be performed.

The aim of this thesis is the reduction of uncertainty in dune erosion prediction, in particular at complex dune coasts, in order to improve dune safety assessment methods. To that end, state-of-the-art process-based models are employed to study the influence of the cross-shore bathymetry, wave obliquity and coastline curvature on dune erosion. In addition, a Bayesian Network is introduced as a probabilistic method to account for their uncertainties.

Reducing uncertainty in dune erosion prediction is envisaged by deploying a more comprehensive dune erosion modelling approach and by using a tailored probabilistic approach. The development of the dune erosion modelling capabilities concerns the improvement of the 1D cross-shore modelling skills and the expansion of the coverage by providing a 2DH modelling approach for complex areas. The introduction of a more advanced and computationally expensive dune erosion model leads to other requirements for the probabilistic approach. The probabilistic methods presently applied require either relatively large numbers of simulations (Monte Carlo) or continuous limit state functions (FORM). These requirements inhibit the comprehensive process-based modelling of complex dune areas as considered in this thesis. The use of a Bayesian Network model provides a flexible way to overcome the limitations of the probabilistic methods and allows model, field and laboratory data to be combined as well as experts' opinions to estimate the dune erosion rate.

The process-based XBEACH model is deployed in 1D mode to investigate the influence of distinct parts of the cross-shore profile. Intercomparison of model simulations of different cross-shore profiles led to the conclusion that the upper profile part is of major importance for the dune erosion volume under extreme conditions, while the lower profile part only has marginal influence. The DUROS+ model, utilized in the current Dutch dune safety assessment method, takes the upper profile part explicitly into account, but the lower part implicitly. Hence, it is not strictly needed to modify the safety assessment method regarding the cross-shore bathymetry influence, provided that the offshore profile part does not influence the forcing significantly.

The influence of wave obliquity on dune erosion is studied by XBEACH simulations in 2DH for a simplified alongshore uniform coastal stretch with time invariant hydraulic loading. For an incident wave angle of 40°, 30% more erosion is found with respect to the reference case with shore normal wave direction. The 2DH model shows additional stirring of the sediment, which is related to the wave driven alongshore current enhancing the cross-shore transport.

A 2DH curvilinear XBEACH model is used to investigate the governing phenomena that play a role in dune erosion on a curved coastline. The series of model simulations with an alongshore uniform bathymetry and time invariant hydraulic loading shows a strong relation between the dune erosion volume and the coastal orientation with respect to the (incident) wave angle. Only a marginal dependency on the coastal radius is found for approximately shore normal waves. At the area along the curved coast where the incident wave angle is 45° with respect to the local coastal orientation, 30 to 50%more erosion is found relative to the reference case with a straight coast and shore normal wave direction. The alongshore variation in alongshore current, relative wave obliquity, wave height and wave set-up as well as their interactions make the erosion rate primarily spatially varying rather than coastal radius dependent.

A fully probabilistic evaluation (Monte Carlo) of the failure probability of the first dune row is carried out for the major part of the Dutch dune coast where a 1D dune erosion model (DUROS+) is applicable. In nearly the full study area, the first dune row meets the normative safety level. At the limited number of locations where first dune row's failure probability exceeds the normative safety level, either these areas are not part of the primary sea defence or secondary dune bodies are present landward.

As an alternative to the probabilistic methods Monte Carlo and First Order Reliability Method (FORM), a Bayesian Network is introduced that estimates the relative dune erosion due to the effects of wave obliquity and coastal curvature, with respect to a reference case with a straight coast and shore normal wave direction. The Bayesian Network uses the XBEACH model results concerning wave obliquity and coastal curvature. The Bayesian Network has been utilised to evaluate the failure probability of the first dune row, of the major part of the Dutch dune coast, with the effects of wave obliquity and coastal curvature included. The failure probabilities are a factor O(10) higher for the major part of the study area, with respect to the reference without wave obliquity and coastal curvature effects.

This thesis stresses the importance of wave obliquity and coastal curvature for the assessment of dune safety against flooding. The effects of these phenomena are investigated based on 2DH XBEACH model simulations in a schematised configuration and show that wave obliquity and coastal curvature can lead to significantly larger dune erosion volumes and thus larger dune failure probabilities. Hence, there is a need to include the effects of wave obliquity and coastal curvature into future versions of the dune safety assessment method, combined with proper validation and a tailored probabilistic method.