Abstract
Knowing instantaneous locations and numbers of individuals in animal populations is a major requirement for wildlife and conservation ecology. Recent advances in very high spatial resolution digital-imaging systems and in object-based image-analysis methods offer great potential for developing remote sensing in new application arenas, including direct mapping and counting of birds, mammals and other larger organisms. We present the successful application of an automated object-based image-mapping strategy that has been applied to total mapping, using aerial image data, of Phoeniconaias minor (Lesser Flamingo) individuals at Kamfers Dam, a large perennial lake in central South Africa. The object-based method used quadtree image segmentation and sequential object-brightness thresholding to identify individual birds with high accuracy (>99% compared to human visual interpretation). Accuracy-assessment results are presented, with discussion of the error factors related to the object-based method and the reference data. An under-estimation by the object-based method of less than 0.5% is indicated. The automated procedure mapped 81 664 Lesser Flamingos at this one site, which is 30% above the most recent estimated size of the entire southern Africa Lesser Flamingo population, indicating a need for further work to reassess this population's size.
Acknowledgements
The authors wish to thank Peter Hohne and Jahn Hohne for their support of our work at Kamfers Dam, Ekapa Mining for paying for the images, Herbert and Brenda Booth for access to their property (Kamfers Dam), Enrico Oosthuysen (Department of Environment Affairs & Nature Conservation, Northern Cape Province, South Africa) for assistance with topographic maps for georegistration of the Kamfers Dam image data, Graham McCulloch for the Sua Pan image data, Kate N. Farrell for assistance with Kappa-value calculations and the anonymous referees. We gratefully acknowledge support from the Danish Strategic Research Council for this project.