A Multi-scale Segmentation Approach to Mapping Seagrass Habitats Using Airborne Digital Camera Imagery

Abstract

The purpose of this study was to map the areal extent and density of submerged aquatic vegetation, principally the seagrasses, Zostera marina and Ruppia maritima, as part of ongoing monitoring for the Barnegat Bay, New Jersey National Estuary Program. We examine the utility of multiscale image segmentation/object-oriented image classification using the eCognition software to map seagrass across our 36,000 ha study area. The multi-scale image segmentation/ object oriented classification approach closely mirrored our conceptual model of the spatial structure of the seagrass habitats and successfully extracted the features of ecological interest. The agreement between the mapped results and the original field reference was 68 percent (Kappa � 56.5 percent) for the four category map and 83 percent (Kappa � 63.1 percent) for the presence/absence map; the agreement between the mapped results and the independent reference data was 71 percent (Kappa � 43.0 percent) for a simple presence/absence map. While the aerial digital camera imagery employed in this study had the advantage of flexible acquisition, suitable image scale, fast processing return time, and comparatively low cost, it had inconsistent radiometric response from image to image. This inconsistency made it difficult to develop a rule-based classification that was universally applicable across the 14 individual image mosaics. However, within the individual scene mosaics, using the eCognition software in a “manual classification” mode provided a flexible and time effective approach to mapping seagrass habitats.