Prediction of Incipient Breaking Wave-Heights Using Artificial Neural Networks and Empirical Relationships

The accurate prediction of shallow water breaking heights is paramount to better understanding complex nonlinear near shore coastal processes. Over the past 150 years, numerous empirical relationships have been proposed based on scaled laboratory datasets. This study utilizes a newly available field collected full-scale dataset of breaking wave conditions to investigate the accuracy of published empirical models and a novel artificial neural networks (ANN) model in predicting the final breaking wave-height for laboratory-scaled and full-scaled ocean waves. Performance is measured by comparison against both the field datasets and 465 separate datasets from 11 independent laboratory studies. The relationship of Rattanapitikon and Shibayama [2000 “Verification and modification of breaker height formulas,” Coastal Eng. J. 42 (4), 389–406.] outperformed all available empirical models when tested against only laboratory datasets, but was superseded by the relationship of Robertson et al. [2015 “Remote sensing of irregular breaking wave parameters in field conditions,” J. Coastal Res. 31 (2), 348–363.] when tested against only field datasets. However, this study noted that models developed based on scaled laboratory tests tend to underestimate the ocean full-scale breaking wave-heights. The training and testing of the ANN model were accomplished using 75% and 25% of the combined field and laborartory datasets. The ANN models consistently outperformed predictive accuracy of empirical models. Sensitivity analysis of the trained ANN models quantified the relative impact of individual wave parameters on the final breaking wave-height.