This paper presents correlative species distribution modelling methods that could be used to predict the circumpolar distribution of habitat suitability for Antarctic toothfish. Potential environmental covariates that could be included in these models are presented for discussion and feedback, as are the types of CCAMLR data (i.e. presence of the species and/or attributes) and modelling algorithms that may be included in future work. Methods for dealing with issues such as sampling bias in model fitting, extrapolation in model predictions and spatial sorting bias in model evaluation are also suggested and highlighted for discussion. Preliminary results are provided to demonstrate the application of two methods (BIOCLIM and Maxent). For simplicity models included two environmental covariates (bathymetry and latitude) and default model settings were used. Evaluation of prediction accuracy included a select number of test statistic (i.e. AUC, max Kappa and point-biserial correlation) with some preliminary consideration of spatial sorting bias.
Mr Doug Cooper (CCAMLR Secretariat)