Megafauna and demersal fish play an important role in benthic ecosystem function as they sequester carbon through the continuous redistribution of organic matter and control the population dynamics of smaller biota via predation, bioturbation and bio-engineering. Therefore, towed camera systems were deployed regularly at HAUSGARTEN to assess megafaunal community dynamics, yielding an ever growing number of images. The manual analysis of such large quantities of images is unfeasible as it is labour-intensive, time-consuming, and subjective. As with many similar studies elsewhere, the valuable footage can currently not be used to its full potential. Many megafaunal species, however, occur as singletons or are characterised by a patchy distribution calling for the analysis of large areas and large numbers of many images, respectively. 

To develop new methods to rapidly analyse these large quantities of images, we collaborate with the Biodata Mining & Applied Neuroinformatics Group at Bielefeld University. In the CORAMM project (Coral Risk Assessment, Monitoring and Modelling), we developed computer algorithms that can be trained to automatically detect corals on seafloor images. The system relies on experts to label areas on a set of images covered by a particular organism. To gather labels, we have developed a new web-2.0-based tool (Bielefeld Image Graphical Labeller and Explorer, BIIGLE). Given a sufficiently high number of labels, the computer system, can ‘learn’ how a feature is distinguished from the remainder of the image. Once trained, the system can automatically detect the features that it has learnt automatically in larger data sets. The system is currently developed further to enable a rapid analysis of seafloor photographs from the HAUSGARTEN time series. This will allow us to assess changes in the megafaunal assemblage.

Contact: M. Bergmann

Literature:

Purser, A., Bergmann, M., Lundälv, T., Ontrup, J. & T. Nattkemper (2009). Use of machine-learning algorithms for the automated detection of cold-water coral habitats - a pilot study. Marine Ecology Progress Series 397: 241-251.