General models of marine animal habitats require a process-based approach to parameter selection and model design

Abstract

In Skov & Thomsen (2008) we concluded — based on a novel, process-based approach — that the most important factor governing the distribution of harbour porpoises at Horns Reef, North Sea, is tidally-driven upwelling. In their Comment, Tougaard & Wisz (2010, Mar Ecol Prog Ser 399:295-297) argue that the conclusion in Skov & Thomsen (2008) was poorly founded by the analyses, due to inclusion of data collected during the construction of a wind farm, the selection of explanatory variables based on data dredging, and the lack of assessment of statistical uncertainty and model validation. We demonstrate that the selection of the data was not compromised, neither by the presence of wind farm construction activities, nor by the initial use of a large number of model parameters. The integrated application of hydrodynamic models, geo-statistics and partial least squares regression assured that model parameters were selected using an ecosystem-based ap- proach. A formal internal validation of model performance using relative operator characteristics underlined the excellent robustness of our models, with AUC values ≥0.78.