Assessing the accuracy of acoustic seabed classification for mapping coral reef environments in South Florida (Broward County, USA)

Abstract: The Atlantic coast of Broward County, Florida (USA) is paralleled by a series of progressively deeper, shore-parallel coral reef communities. Two of these reef systems are drowned early Holocene coral reefs of 5 ky and 7 ky uncorrected radiocarbon age. Despite the ease of access to these reefs, and their major contribution to the local economy, accurate benthic habitat maps of the area are not available. Ecological studies have shown that different benthic communities (i.e. communities composed of different biological taxa) exist along several spatial gradients on all reefs. Since these studies are limited by time and spatial extent, acoustic surveys with the QTCView V bottom classification system based on a 50 kHz transducer were used as an alternative method of producing habitat maps. From the acoustic data of a 3.1 km2 survey area, spatial prediction maps were created for the area. These were compared with habitat maps interpreted from in situ data and Laser Airborne Depth Sounder (LADS) bathymetry, in order to ground-truth the remotely sensed data. An error matrix was used to quantitatively determine the accuracy of the acoustically derived spatial prediction model against the maps derived from the in situ and LADS data sets. Confusion analysis of 100 random points showed that the system was able to distinguish areas of reef from areas of rubble and sand with an overall accuracy of 61%. When asked to detect more subtle spatial differences, for example, those between distinct reef communities, the classification was only about 40 % accurate. We discuss to what degree a synthesis of acoustic and in situ techniques can pro