EVALUATION OF OVERTOPPING MODEL PERFORMANCE USING NOVEL EXPERIMENTAL DATA FROM INDUSTRIAL DESIGN PROJECTS

Abstract

A critical requirement in successfully planning to mitigate wave overtopping is the ability to predict the frequency at which coastal defences will be overtopped. Many empirical formulae have been developed to predict wave overtopping rates for specific structural typologies and hydrodynamic conditions. More recently, Machine Learning methods have been deployed in an effort to make models that generalize across a wide range of structures and environments. A critical enabling factor has been the compilation of systematically parameterized physical model data, culminating in the EurOtop extended database. Practitioners now have the luxury of choosing between multiple, high-quality models. In addition, given the rapid advancement in usability of modern Machine Learning frameworks, training their own bespoke models is an increasingly realistic option. What is now needed is more information, showing how these models perform on unseen data, in a practical design context, in order to continue to refine guidance around their use.