Unmanned aerial vehicles (UAVs) are cost-effective remote sensing tools useful for generating very high-resolution (VHR) aerial imagery. Habitat maps generated from UAV imagery are a fundamental component of marine spatial planning, essential for the designation and governance of marine protected areas (MPAs). We investigated whether UAV survey altitude affects habitat classification performance and the classification accuracy of thematic maps from a tropical shallow water environment. We

more »

... ted repeated UAV flights at 75, 85, and 110 m, using a fixed-wing UAV on the Turneffe Atoll, Belize. Flights were ground truthed with snorkel surveys. Images were mosaiced to form orthomosaics and transformed into thematic maps through semi-automatic object-based image analysis (OBIA). Three subset areas (4000 m 2 , 17 000 m 2 and 17 000 m 2 ) from two cayes on the atoll were selected to investigate the effect of survey altitude. A linear regression demonstrated that for every 1 m increase in survey altitude, there was a ~1% decrease in the overall classification accuracy. A low survey altitude of 75 m produced a higher classification accuracy for thematic maps and increased the representation of mangrove, seagrass and sand. The variability in classified cover was driven by altitude, although the direction and extent of this relationship was specific to each class. For coral and sea, classified cover decreased with increased altitude. Mangrove classified cover was non-sensitive to altitude changes, demonstrating a lesser need for a consistent survey altitude. Sand and seagrass had a greater sensitivity to altitude, due to classified cover variability between altitudes. Our findings suggest that survey altitude should be minimized when classifying tropical marine environments (coral, seagrass) and, given that most fixed-wing UAVs are restricted to a minimum altitude of 70 m, we recommend an altitude of 75 m. Survey altitude should be a major consideration when targeting habitats with greater sensitivity to altitude variability.