An investigation of the performance of a data-driven model on sand and shingle beaches

In this paper Canonical Correlation Analysis (CCA) is used as the basis for predicting beach profile changes over a timescale of years. In particular, datasets of wave measurements and beach profiles at two locations with different sediment types and wave exposure are used. The study sites are located in Christchurch Bay (South Coast of England) in which the beaches are classified as mixed shingle and sand beaches and in North Carolina (East Coast of the USA) which has sandy beaches. The datasets comprise detailed bathymetric surveys of beach profiles covering a period of more than 15 years for the New Forest dataset and over 27 years for the Duck dataset. The structure of the datasets and the data handling methods are described. The application of the CCA method is discussed as well as the ability of the CCA to provide useful forecasts of the beach profile at each site. Some results and interpretation of the analysis are presented. Predicted beach profiles elevations agreed well with the measured for up to eight years of forecast at both of the sites. This implies that there is no clear degradation in prediction accuracy over the forecast period. Furthermore the level of error obtained using this approach is commensurate with that found with dynamical modelling. This kind of prediction could be of practical use for engineers and planners.