Detecting Subpixel Changes in the Coastal Vegetation Line With Sentinel-2 Imagery

Abstract

Monitoring coastal erosion often requires detecting changes in the vegetation line (VL). Traditional approaches rely on aerial photography and manual annotation, restricting analysis to a limited number of sites. In this article, we present a deep learning framework that leverages Sentinel-2 (S2) imagery to detect VLs and estimate erosion rates at scale. We mitigate resolution and geolocation limitations using an interpolation process that averages detected VLs across multiple scenes over time. This is aided by the guidance band (GB)—an additional input channel that indicates the approximate location of a VL. Despite the 10-m resolution of S2, this approach achieves subpixel precision. Specifically, we detect VLs with a mean absolute error of 2.2 m when validated against manual annotation on aerial photography. As part of the work, we produce the Sentinel-2 Irish Vegetation Edge (SIVE) dataset. To support reproducible research, the SIVE dataset and all accompanying code are released openly. This work demonstrates the potential of S2 imagery for scalable, automated coastal erosion monitoring and provides a foundation for future methodology extensions.