The performance of shoreline detection models applied to video imagery

Abstract

Digital images of the intertidal region were used to map shorelines and the intertidal bathymetry along four geomorphically and hydrodynamically distinct coastlines in the United States, United Kingdom, The Netherlands, and Australia: Mapping methods, each of which was originally designed to perform well at only one of the four sites, were applied to all four sites, and the results were compared to direct topographic surveys. It was determined that the rms errors of image-derived versus directly surveyed elevations depended on the prevailing hydrodynamic conditions as well as differences in each of the four different mapping methods. Before these differences were accounted for, rms errors ranged from 0.3 to 0.7 m. An empirical correction model that computed local estimates of setup, swash, and surf beat amplitudes reduced errors by about 50%, with residual rms errors ranging between 0.1 and 0.4 m. The model required tuning only one parameter that determined how each method was affected by swash at infragravity and incident wave frequencies. In environments where all methods successfully identify shorelines, the methods can be used somewhat interchangeably. The diversity of methods is advantageous in situations where one or more methods are likely to fail (e.g., lack of color imagery, high degree of alongshore variability). This remote sensing methodology can be applied to the shoreline and inter-tidal mapping problem across diverse nearshore environments.