Operational wave forecasting plays an important role in ensuring safe navigation and in the prediction of tidal windows for harbour approach channels. The underlying nearshore process-based wave models need to be accurate for a wide range of different conditions, from more common mild wave conditions to the occasional high energy (storm) conditions. In this work, an innovative hybrid modelling approach is proposed to improve the accuracy of operational wave forecasts. An operational wave model is combined with a machine learning model which is trained using wave measurements within the wave model domain. This hybrid modelling approach is applied to the Dutch North Sea, covering four major harbour approach channels.

The final hybrid operational wave model results in a significant average error decrease compared to just the process-based model, amounting to 21.7% for the wave energy density and 25.3% for the wave direction. The error reduction for the spectral wave parameters is even larger, with a 33.3% smaller error in spectral wave height and a 38.8% smaller error in spectral wave period. As this approach is generically applicable to spectral wave models, it contains the potential for significant improvements in operational modelling.@en

When performing physical or numerical experiments in a wave flume, it is important to distinguish between the incident and reflected wave components. Recently wave separation methods including nonlinear effects were presented which can be applied for a large range of conditions and reduces the error in the wave separation. However, these nonlinear methods also result in a complex system of equations and require more wave gauges. In this work, physical model experiments with a high spatial resolution of wave gauges were carried out to validate different wave separation methods. All tests were analysed with a Linear method, the nonlinear method described in Eldrup and Lykke Andersen (2019) (ELA method) and a Modified ELA method. In this Modified ELA method, the model complexity depends only on the condition number of the phase difference matrix making the accuracy less dependent on wave gauge positions and noise. The results show that both Nonlinear methods are always preferred over the Linear method. A detailed analysis of the number of wave gauges shows that for the same distance between first and last gauge, 6 to 8 wave gauges are required for a converged solution independently of the wave gauge position. In an ideal situation – with an ideal number of wave gauges and ideal spacing between them – the ELA and Modified ELA methods should give nearly identical results. However, for the non-ideal conditions encountered in practice, the Modified ELA is shown to be more robust.@en

Coastal structures are often designed to a maximum allowable wave overtopping discharge, hence accurate prediction of the amount of wave overtopping is an important issue. Both empirical formulae and neural networks are among the commonly used prediction tools. In this work, a new model for the prediction of mean wave overtopping discharge is presented using the innovative machine learning technique XGBoost. The selection of features to train the model on is carefully substantiated, including the redefinition of existing features to obtain a better model performance. Confidence intervals are derived by tuning hyperparameters and applying bootstrap resampling. The quality of the model is tested against four new physical model data sets, and a thorough quantitative comparison with existing machine learning methods and empirical overtopping formulae is presented. The XGBoost model generally outperforms other methods for the test data sets with normally incident waves. All data-driven methods show less accuracy on oblique wave data, presumably because these conditions are underrepresented in the training data. The performance of the XGBoost model is significantly improved by adding a randomly selected part of the new oblique wave cases to the training data. In the end, this new model is shown to reduce errors on all data used in this work with a factor of up to 5 compared to existing overtopping prediction methods.

@en

In coastal engineering context, the use of video imagery is widespread. Especially in field settings along sandy coasts, many types of data have been derived from video imagery, often using non-learning analysis techniques. Recent advances in the field of computer vision and deep learning allow for the automation of image segmentation. In this paper, these techniques are combined with video imagery of physical model tests, resulting in an innovative non-intrusive measurement technique. This technique is validated for three different applications: the measurement of surface elevation, wave run-up and bed level development. In addition to demonstrating its potential as an alternative for existing measurement instruments, it is shown that the added detail in the spatial or temporal domain provided by the technique can lead to new insights. Examples of this are the detailed analysis on the variability of the run-up height over the width of the flume and the spatial distribution of run-up velocities over the slope.@en

In order to improve the current aeolian modeling, a process-based model is developed by implementing biological and physical processes adapted from different existing models. The current version of the model is capable of simulating barchan and parabolic dunes and the initial stage of coastal dune formation: Embryo dunes. Potentially the model can be used to determine the influence of tidal ranges, storm frequencies, armoring, salinity and precipitation on dune building processes. This will result in a greater insight in the general behavior of coastal systems and on the other hand the model can also be used for engineering purposes.@en