Shoreline-position forecasting: Impact of storms, rate-calculation methodologies, and temporal scales

Despite the considerable research that has sought to describe past and pred ict future shoreline change, little consensus has emerged on the best metho dology for forecasting future shoreline positions. While a certain degree o f heterogeneity in approach is warranted given the variability in coastal g eomorphology and sediment-transport processes, the prediction error associa ted with each method has not been evaluated in great detail. In this study, measured shoreline positions from Delaware and New York were used to calculate long-term erosion rates and make predictions to subseque nt, known positions. Rates were calculated using end-point and linear-regre ssion methods, including and excluding storm-specific shorelines. Those rat e computations that included storm-specific shorelines yielded consistently poor predictions (average factor-of-three increase in error) compared with non-storm erosion rates, regardless of rate-calculation method. Linear-reg ression predictions, on average, performed better than end-point rate predi ctions, reducing error by over 70% in New York and 34% in Delaware for rate s including storm shorelines, and between 4 and 31% for non-storm data (DE and NY, respectively). Predictions (hindcasts) were also made to 19(th) cen tury shoreline positions using rates computed with modern, non-storm data. The positions predicted along relatively undeveloped stretches of the coast were within the 95% confidence interval associated with the prediction. Hi ndcasts made in areas characterized by heavy development and/or beach nouri shment projects were poor, as would be expected given the recent alteration of the natural sediment-supply system. For all locations, inclusion of 19( th) century data reduced uncertainty in forecasts of 21(st) century shoreli ne positions by roughly 44%. These results show that forecasts derived from linear-regression rates using non-storm, 19(th) and 20(th) century data pr oduce the lowest prediction error and uncertainty in the long-term trend.