Automated Lake Shoreline Mapping at Subpixel Accuracy

Abstract

Accurate assessment of lake extent and variation is essential for understanding the global hydrological cycle and necessitates the use of satellite data. The intrinsic scale of lake dynamics is finer than the spatial resolution of available images. Lake change detection requires mapping lake shorelines with high positional accuracy as they are employed for calculating changes in area and volume. There exist many techniques for mapping lake shorelines from multispectral images, and their accuracy decreases as the image spatial resolution degrades. In addition, lake shorelines obtained from images have a zagged appearance, which not only provides an unrealistic representation but also overestimates the lake perimeter. This letter presents a method for subpixel lake shoreline mapping. The approach is used to map subpixel shorelines from Landsat Enhanced Thematic Mapper Plus images acquired in the lake-rich Arctic Coastal Plain of northern Alaska, where lakes dominate the landscape and change significantly over time. High performance of the proposed approach is demonstrated by the agreement of subpixel shorelines with high-resolution QuickBird panchromatic images. The subpixel shorelines estimated by the proposed approach are smooth and lead to precise area estimation and displacement error reduction when compared with the shorelines obtained using image segmentation. The proposed approach has great potential in regional-scale shoreline mapping.