Calibration data requirements for modelling subaerial beach storm erosion

Numerical coastal erosion models are used by engineers to predict the magnitude of storm erosion at the coastline. In particular, the 'dry' region of the beach, extending landwards from the waterline to include the beach face, berm and dunes, is a particular focus of interest to managers, planners and other coastal practitioners. However, the choice of the most appropriate numerical model to predict subaerial beach erosion requires careful consideration of the strengths and limitations of each model, and the quality and quantity of field data available for both model calibration and prediction. While no model can perfectly replicate observed upper beach erosion, this study specifically assesses the quantity of field calibration data required to achieve optimum model performance for coastal storm erosion modelling applications. Two of the most commonly used, but differently formulated, coastal erosion profile models are compared: the process-based and more complex model XBeach, and the semi-empirical and significantly simpler model SBEACH. A rigorous calibration technique (the Generalised Likelihood Uncertainty Estimation) is applied to both models, using a comprehensive dataset of pre- and post-storm topographic measurements collected over four differing storm events at Narrabeen-Collaroy Beach in southeast Australia, as well as at two adjacent embayed beaches. When applying the two numerical models using their default parameters (i.e., with no model calibration), SBEACH was found to be the more skilful model and XBeach default parameters were found to have no predictive skill along this stretch of coastline. Once calibrated with detailed field observations obtained before and after a single storm event, XBeach validation skill rose considerably and provided predictions of subaerial beach profile change with greater skill (87% better Brier Skill Score on average for the same calibration data) than SBEACH. Overall XBeach model performance was found to marginally improve when field observations obtained from additional storm events were included in the calibration process, whereas SBEACH showed negligible improvement. A comparison was made of the options available for transferring previously-calibrated parameters to adjacent locations, finding spatial proximity to be the most sensitive indicator of model performance for a transferred parameter set. Due to the alongshore variability in pre-storm topography as well as wave exposure at the sites tested, the spatial coverage of data was more important for the calibration process than the magnitude of the individual storm(s) used for a specific calibration. The results of this study underscore the need for careful consideration of the available calibration field data when choosing and optimising coastal erosion models.