Extent, accuracy and repeatability of bare sand and vegetation cover in dunes mapped from aerial imagery is highly variable

Abstract

Vegetation cover on coastal sand dunes has been increasing worldwide since at least the 1940s. Analysis of aerial and satellite imagery has been the principal source used to measure this change, however no studies have systematically evaluated the accuracy of remotely sensed estimates. Using established land cover classification methods and in-situ field measurements, we show that both the extent and accuracy of remotely sensed areas of bare sand and vegetation in dunes varies with image resolution and classification method. We found that supervised methods of classification (semi-automatic), whilst mapping a greater extent of bare sand and being more accurate than manual digitisation, had poor repeatability, exhibiting a relatively large range of bare sand and vegetation extent between classifications replicated under the same conditions. In contrast, areas of bare sand and vegetation classified by manual digitisation had high repeatability but a relatively low percentage of observed agreement with data collected in the field. For all classification methods, observed agreement with field data generally increased with image resolution. Our results demonstrate that users of land classification data in dunes should be cautious when interpreting trends of bare sand and vegetation cover due to substantial repeatability error in supervised classification methods, and relatively poor observed agreement with field data of manual classification. We recommend that analysis of bare sand and vegetation cover in dunes should be based on multiple replicates using supervised classification, employing the highest resolution imagery available and that all results presented should also include the range measured by multiple replicates.