When mathematics and the seafloor collide!
Antarctic sea-floor communities are unique, and more closely resemble those of the Palaeozoic than equivalent contemporary habitats. However, comparatively little is known about the mechanisms and interactions that structure these communities, or how they might respond to anthropogenic change.
This work is what happens when Bayesian networks and Antarctic benthic biology come together for a truly multidisciplinary approach. The three of us all have interests in marine community evolution and what shapes community structure but come at it from very different backgrounds. Working with people from different disciplines enabled us to approach some very traditional ecological questions with a new arsenal of tools.
The starting point was the data, hundreds of seafloor photographs taken around the South Orkney Islands, an isolated archipelago, North of the tip of the Antarctic Peninsula. The expedition was designed to investigate the biodiversity of the region and to assess the Marine Protected Area to the south of the islands. Transects of the sea floor were imaged at three different depths (500 m, 750 m and 1000 m) from 5 geographic regions. Dr Madeline Brasier quantified the photographs for a previous study comparing regional differences in diversity and abundance using a more traditional approach. This dataset was an ideal test case for our analyses because it was deliberately collected to sample a range of habitats at standardised depths using the same methodology.
Our study determines the community structure using Bayesian Network Inference analysis, which is a technique to statistically infer the causal relationships (or dependencies) between different variables. In the case of our image data the variables are the different groups of organisms and different environmental factors such as substrate type and depth. The strength of the relationship between two variables is calculated as a cumulative frequency distribution over all possible discrete states. Discrete Bayesian Network Inference Algorithms enable the description of causal relationships (over just mutual correlations) as well as non-linear interactions. We used these algorithms to determine the key community structuring organisms and environmental factors that underpin Antarctic benthic community dynamics. From this we were able to to infer likely changes and impacts caused by the removal or decline of these key taxa.
We found that sponges had the highest number of significant connections to other taxa and, therefore, the greatest influence on determining the composition of the seafloor community. These results were surprising because abiotic factors such as substrate and depth are usually considered to be the main driving factors. When we removed sponges from the network, the abundances of all major taxa reduced by a mean of 42%, significantly more than changes of substrate. Given that sponges are vulnerable to human impacts, e.g. trawling, longline fishing and climate change, these results show the importance of considering such habitat forming keystone taxa when considering marine spatial planning and management.
The limits of this first study were determined by two important factors; what was visible in the photographs (small animals and those burrowing in the sediment were not accurately represented) and the taxonomic level to which the photographs were quantified. Future use of this technique in Antarctica will examine the community in greater detail by identifying the organisms to the lowest taxonomic level possible and to use larger photographic datasets which cover an even wider range of habitats. The introduction of an AI species identification method would allow many more photographs to be processed and the power of the model to be increased. We are already on the lookout for suitable PhD Student to build upon this work.
This new tool for understanding the connectivity within an ecosystem has given us new insights into the importance of Antarctica's amazing habitat forming sponges and how removing one group can dramatically influence other parts of the system.