Splinter, K.D., Davidson, M.A., and Turner, I.L. Monitoring data requirements for shoreline prediction: How much, how long, and how often?
Predicting future shoreline position and the subsequent vulnerability of coastal communities is of great interest to coastal managers and engineers. A number of empirically-based equilibrium shoreline models have been proposed, but all have been formulated on unique and relatively short-term datasets. As such, the transferability and calibration sensitivity of these models is unknown. Here we use a newly developed shoreline model, ShoreFor, driven by changes in offshore wave steepness and wave power, to determine model calibration coefficient sensitivity based on two key constraints in any monitoring program: 1) the duration of data collection of necessary shoreline and wave data (N); and, 2) the shoreline sampling interval (dt). The analysis was undertaken at two distinct sites: a 5-year data set from Gold Coast, QLD, Australia, a long, straight, open coast comprising medium grain sand with a strong annual cycle in the shoreline; and a 6-year data set from Narrabeen-Collaroy, NSW, Australia, which is a coarser, swash-aligned embayed beach whose shoreline variability is driven predominately by storms. In both cases, shoreline data were extracted weekly from video imagery and hourly offshore wave data were obtained from buoy measurements. Shoreline calibration datasets were then sub-sampled in both time (N) and frequency (dt). Results presented in this paper suggest calibration dataset requirements are beach-type dependent and linked to the rates of sediment exchange between the nearshore and shoreline. The storm-dominated site, Narrabeen-Collaroy, required more frequently sampled shoreline data to converge on model free parameters and maintain model hindcast normalized mean square error (NMSE) below 0.5. As a general ‘rule of thumb' a minimum of two years, sampled at dt ≤30 days was required to consistently resolve model coefficients at both sites, while calibration datasets longer than three years (roughly 50% of the data set) resulted in minimal improvement in model hindcast skill at the two sites examined.