Use of predictive habitat modelling to assess the distribution and extent of the current protection of ‘listed’ deep‐sea habitats

Abstract Aim To demonstrate the application of predictive species distribution modelling methods to habitat mapping and assessment of percentage area‐based conservation targets. Location The NE A tlantic deep sea ( UK and Irish extended continental shelf limits). Methods M ax E nt modelling of three listed habitats ( L ophelia pertusa ( L innaeus, 1758) reef ( L p R eef), P heronema carpenteri (WyvilleThomson, 1869) aggregations ( P c A ggs) and S yringammina fragilissima ( B rady, 1883) aggregations ( S f A ggs)), with some pre‐selection of variables by generalized additive modelling. Models are validated using repeated 70/30 build/test data splits using AUC and threshold‐dependent assessment methods. Predicted distribution maps are used to assess the adequacy of existing area closures for the protection of listed habitats and to assess percentage representation of each community within existing MPA networks. Results Model performances are rated as fair ( L p R eef), excellent ( P c A ggs) and good ( S f A ggs). Current closures are focused on the protection of cold‐water coral reef and incidentally capture some S f A ggs suitable environments, but largely fail to protect P c A ggs. Considering the wider network of MPA s in the study region, approximately 23% ( L p R eef), 2% ( P c A ggs) and 6% ( S f A ggs) of the area predicted as suitable for each habitat respectively is contained within an MPA . Main conclusions To date, decisions on area closures for the protection of ‘listed’ deep‐sea habitats have been based on maps of recorded presence of species that are taken as being indicative of that habitat. Predictive habitat modelling may provide a useful method of better estimating the extent of listed habitats, providing direction for future MPA establishment and a means of assessing MPA network effectiveness against politically set percentage targets. Given the coarse resolution of the model, percentages should be taken as maximal figures, with habitat occurrence likely to be less prevalent in reality.