Implication of General Regression Neural Network (GRNN) for Wave Overtopping Prediction of Vertical Sea Defences

Commentary

Can artificial intelligence help predict when waves will overtop a sea wall?

Wave overtopping is a critical consideration in the design of coastal defences. Predicting it, however, is difficult because it depends on a complex combination of wave conditions, water level and the geometry of the structure.

In this early application of machine learning to coastal engineering, a General Regression Neural Network (GRNN) was trained using the European CLASH wave overtopping database, which contains around 18,000 experimental records collected from research facilities around the world.

Rather than asking one model to represent every type of coastal structure, the study investigated whether an AI model trained specifically on vertical sea defences could perform better. The results showed that the specialised model produced considerably stronger predictions than a model trained across many different structure types and also outperformed the conventional EurOtop prediction approach considered in the study.

The work highlights an idea that remains highly relevant in engineering AI today: more data are not always better if they describe fundamentally different physical systems. Carefully selected, problem-specific data can produce a more useful predictive model.