On Forecasting Future U.S. Shoreline Positions: A Test of Algorithms

Abstract

Coastal development is proceeding rapidly in the United States while at the same time coastal erosion is almost ubiquitous. The result is that dollar losses due to coastal storms and flooding are reaching economically intolerable levels. Avoiding even larger losses in the future requires effective land use management practices. These typically take the form of building setbacks to serve as protection for a time comparable to the expected lifetime of new coastal structures, usually 30 or 60 years. However, determining adequate setbacks requires estimating long-term shoreline change trends from temporally poorly sampled historical shoreline position data bases. As such, it is important to recognize these data limitations so that faulty analyses and potentially costly errors of prediction can be minimized or avoided. Since there is a well-known relationship between long-term shore retreat and sea level rise, we have used temporally complete sea level records as surrogate data sets to evaluate various shoreline prediction algorithms (e.g., end point method, linear regression, and a technique based on the minimum description length criterion). Predictions from subsets of sea level data sampled temporally to mimic shoreline data sets are compared to actual sea level values. It has been found that in a clear majority of cases, linear regression gives superior results. Predictions shaped or influenced by higher-order polynomial schemes can sometimes be superior to those obtained from linear regressions, but they can also be extremely inaccurate.