Seagrass meadows cover about 0.05-0.15% of the world’s ocean and are some of the most productive systems on Earth. Direct and indirect human-derived impacts have led to significant seagrass declines worldwide and the alteration of...
moreSeagrass meadows cover about 0.05-0.15% of the world’s ocean and are some of the most productive systems on Earth. Direct and indirect human-derived impacts have led to significant seagrass declines worldwide and the alteration of services linked to their biodiversity. Effective conservation and the provision of sustainable recovery goals for ecologically significant species, such as green turtles (Chelonia mydas), are limited by the absence of reliable information on seagrass extent. This is especially true for the Wider Caribbean Region (WCR) where many conservation initiatives are under way, but impaired by the lack of accurate baseline habitat maps. To assist with such a fundamental conservation need using high resolution remote sensing data, both environmental and methodological challenges need to be tackled. First, the diversity of environments, the heterogeneity of habitats, and the vast extent of the targeted region mean that local expertise and field data are seldom available. Second, large-scale high-resolution mapping requires several hundred Landsat 5 and 7 images, which poses substantial processing problems. The main goal of this study was to test the feasibility of achieving Landsatbased large-scale seagrass mapping with limited ground-truth data and acceptable accuracies. We used the following combination of methods to map seagrasses throughout the WCR: geomorphological segmentation, contextual editing, and supervised classifications. A total of 40 Landsat scenes (path-row) were processed. Three major classes were derived (“dense seagrass”, “medium-sparse seagrass”, and a generic “other” class). Products’ accuracies were assessed against (i) selected in situ data; (ii) patterns detectable with very high-resolution IKONOS images; and (iii) published habitat maps with documented accuracies. Despite variable overall classification accuracies (45-85%), the resulting thematic maps were deemed acceptable to (i) regionally provide, following their critical evaluation, an adequate baseline for further large-scale conservation programs and research actions; and (ii) regionally re-assess carrying capacity estimates for green turtles. They certainly should represent a drastic improvement relative to current regional databases.