Accounting for positional uncertainty in historical shoreline change analysis without ground reference information

Systematic shifts in shoreline position are important indicators of environmental change. Shoreline position interpreted from historical aerial imagery is frequently used to assess shoreline change. Although most published studies do not formally consider the effects of source error and interpretation error when analysing shoreline change, the effects of these errors may be significant. This article introduces and evaluates a new uncertainty-aware approach to assessing shoreline change in the presence of positional uncertainty and without ground-reference data, which is typical for historical coastline analysis. The overlapping double buffer (ODB) approach extends the epsilon band model to account for the effects of source and interpretation errors on shoreline position, and assesses the degree of overlap in order to distinguish between significant change and noise. This approach improves upon standard shoreline change analytical techniques that use regularly spaced shore-normal transects to measure the direction and magnitude of shoreline change without accounting for error. Shoreline position interpreted from historical aerial images for four sites along the Michigan (USA) coast between 1938 and 2010 were used to demonstrate the feasibility of the ODB approach. An epsilon band was constructed around each shoreline with radius equal to the combined source and interpretation error for each image. These bands were merged and intersected to test whether the observed change was real or an artefact caused by uncertainty in the data sources. The most significant advantage of the ODB approach is its ability to account for uncertainty in shoreline position where ground-reference information does not exist, as is often the case with historical aerial imagery. Results indicate that the overlapping epsilon band method is viable for analysing change in linear features in the absence of a higher-accuracy reference dataset, and that source error contributed more to the positional uncertainty than did interpretation error.