Modelling commercial fish distributions: Prediction and assessment using different approaches
|Title||Modelling commercial fish distributions: Prediction and assessment using different approaches|
|Publication Type||Journal Article|
|Year of Publication||2011|
|Authors||Jones, MC, Dye, S, Pinnegar, J, Warren, R, Cheung, W|
Species distribution models are important tools to explore the effects of future global change on biodiversity. Specifically, AquaMaps, Maxent and the Sea Around Us Project algorithm are three approaches that have been applied to predict distributions of marine fishes and invertebrates. They were designed to cope with issues of data quality and quantity common in species distribution modelling, and especially pertinent to the marine environment. However, the characteristics of model projections for marine species from these different approaches have rarely been compared. Such comparisons provide information about the robustness and uncertainty of the projections, and are thus important for spatial planning and developing management and conservation strategies. Here we apply the three commonly used species distribution modelling methods for commercial fish in the North Sea and North Atlantic, with the aim of drawing comparisons between the approaches. The effect of different assumptions within each approach on the predicted current relative habitat suitability was assessed. Predicted current distributions were tested following data partitioning and selection of pseudoabsences from within a specified distance of occurrence data. As indicated by the test statistics, each modelling method produced plausible predictions of relative habitat suitability for each species, with subsequent incorporation of expert knowledge generally improving predictions. However, because of the differences between modelling algorithms, methodologies and patterns of relative suitability, comparing models using test statistics and selecting a ‘best’ model are not recommended. We propose that a multi-model approach should be preferred and a suite of possible predictions considered if biases due to uncertainty in data and model formulation are to be minimised.