CCAMLR

Innovative multivariate statistical modeling techniques allow modelers to generate spatially comprehensive species distribution layers from discontinuous biological data, by fitting complex and scale-dependent relationship between species abundance and available environmental data. These species-specific layers can then be used in bioregionalisation, for instance by classifying directly on biological data without the need for environmental proxies.
We describe one method, called BRT (Boosted Regression Trees), by which we propose that CCAMLR may wish to generate species layers to inform the bioregionalisation of the Southern Ocean. We demonstrate the use of this method by generating 13 taxon-specific and aggregate data layers for pelagic zooplankton using a circumpolar dataset collected by Continuous Plankton Recorder (CPR) and twelve existing or newly derived continuous environmental data layers.
We also describe other available data that is appropriate for this method and is likely to be important for the CCAMLR bioregionalisation process, e.g. a quantitative circumpolar krill and salp database and various top predator databases. We also describe biological distribution estimates that we have compiled from other sources, using other methods. These include 115 marine mammal species layers generated using a semi-objective Relative Environment Suitability (RES) model, and 33 avian taxa distributions collated from available literature. While perhaps not as rigorous as distributions generated using statistical methods, we argue that these data constitute best available information at present, and should not be ignored in the bioregionalisation process.
Finally we identify potentially valuable sources of biological data that are currently unavailable but likely to become available in the near future, and advocate the use of ‘placeholder’ data layers built into the bioregionalisation process, to be replaced as better data becomes available. In this way CCAMLR can proceed using best available information at present and still incorporate improved data layers without revisiting methodological or procedural decisions such as those that will be reached by this workshop.

New Zealand has a considerable body of experience creating spatial classifications of the marine environment, and applying them for management. We assert that recent innovations in multivariate statistical modeling have made possible the combined use of spatially comprehensive environmental data and discontinuous biological data to generate rigorous, objective, data-driven classifications of the Southern Ocean sensitive to ecologically important contrasts, consistent with CCAMLR’s ecosystem management mandate. This paper considers a range of methodological options for data-driven marine classification, and reviews the results of three New Zealand classifications to draw methodological and practical lessons relevant to CCAMLR’s Bioregionalisation of the Southern Ocean.
We offer the following explicit recommendations to the CCAMLR Bioregionalisation process: 1) Use biological data; 2) Model species individually; 3) Generate a classification based on abundance, not presence-absence; 4) Use the most powerful statistical methods available, such as BRT and GDM; 5) Use a hierarchical algorithm; 6) Focus on an environment or community of interest; 7) Include information representing uncertainty.
We also highlight some of the inherent limitations of all attempts to represent spatially and temporally dynamic ecosystems using static representations such as produced by marine classifications. We identify important ecosystem processes that may not be captured by even the best classifications, and warn against uncritical or misinformed application of marine classifications in the management stage. Finally we highlight some practical steps to make marine classifications more ‘management-friendly’.

Phytoplankton production during the austral summer in the Southern Ocean is known to be limited by iron and light. Distributions of satellite-detected chlorophyll-a (Chl-a) show very complex and time-variable patterns that are hard to explain. We analysed covariance between satellite-detected and modelled variables and show that this covariance in time between the mixed layer depth (MLD), sea surface temperature (SST) and Chl-a can be used to map areas where different factors control phytoplankton production. Statistically significant spatial patterns in the covariance between MLD, SST and Chl-a show that the physical factors controlling phytoplankton production in the Southern Ocean change in a predictable manner. Well-defined areas exist where phytoplankton is light-limited in the summer due to insufficient stratification and where phytoplankton is clearly limited by nutrients (probably iron). The boundary between light limitation and nutrient limitation can be sharp and may be associated with the main hydrographic fronts (e.g. the Subantarctic Front).