Monitoring changes in the spatial distribution and health of biotic habitats requires spatially extensive surveys repeated through time. Although a number of habitat distribution mapping methods have been successful in clear, shallow-water coastal environments (e.g. aerial photography and Landsat imagery) and deeper (e.g. multibeam and sidescan sonar) marine environments, these methods fail in highly turbid and shallow environments such as many estuarine ecosystems. To map, model and predict key biotic habitats (seagrasses, green and red macroalgae, polychaete mounds [Ficopamatus enigmaticus] and mussel clumps [Mytilus edulis]) across a range of open and closed estuarine systems on the south-west coast of Western Australia, we integrated post-processed underwater video data with interpolated physical and spatial variables using Random Forest models. Predictive models and associated standard deviation maps were developed from fine-scale habitat cover data. Models performed well for spatial predictions of benthic habitats, with 79-90% of variation explained by depth, latitude, longitude and water quality parameters. The results of this study refine existing baseline maps of estuarine habitats and highlight the importance of biophysical processes driving plant and invertebrate species distribution within estuarine ecosystems. This study also shows that machine-learning techniques, now commonly used in terrestrial systems, also have important applications in coastal marine ecosystems. When applied to video data, these techniques provide a valuable approach to mapping and managing ecosystems that are too turbid for optical methods or too shallow for acoustic methods.

Superseded by 78944

Optical, Geospatial, Radar, and Elevation Supplies and Services Panel (OGRE) 2011/12 Annual Report