No abstract

The Canadian Policy for Managing the Impact of Fishing on Sensitive Benthic Areas developed by the Department of Fisheries and Oceans Canada (DFO) in 2009 defines Significant Benthic Areas in DFO's Ecological Risk Assessment Framework as "significant areas of cold-water corals and sponge dominated communities". Kernel density estimation (KDE) was applied to research vessel trawl survey data to create modelled biomass surfaces for corals and sponges. From these, an aerial expansion method was applied to identify significant concentrations of these taxa across eastern Canada. The borders of the areas so identified were refined using species distribution models that predict species presence-absence and/or biomass, both incorporating environmental data. We present such predictive models produced using a random forest (RF) machine-learning technique. A suite of between 54 and 78 environmental predictor variables from different data sources were used. Occurrence models performed well in general with cross-validated AUC (Area Under the Receiver Operating Characteristic Curve) values over 0.8 in most of the cases. Biomass models provided diverse results depending of the taxa and region studied. The biomass models were compared with Generalized Additive Models (GAM), which produced comparable results to random forest, although the fewer assumptions required for RF made this method more convenient. These results have been used to identify significant concentrations of corals and sponges in eastern Canada, an essential first step in the identification of Sensitive Benthic Areas to ensure Canadian fisheries are conducted in a manner that supports marine conservation and sustainable resource use within and outside Canada's 200 nautical mile exclusive economic zone.

No abstract