Inconsistency/consistency of predictive mapping approaches for deep-water habitats: independent assessment recommends caution

Effective seafloor management requires knowledge of the spatial distribution of species and habitats. Surveying every square metre, or even square kilometre of the ocean floor with video or sampling techniques is practically and economically impossible. Even high-resolution acoustic maps, that could give at least information about the seabed morphology and sediments, are not always available. A possible solution may be found in predictive habitat mapping: statistical methods that use the (limited) existing knowledge of species/terrain interaction to predict species and habitat patterns in areas that have not been surveyed. However, our research has shown that the model outcomes have to be treated with caution: alhtough internal validation often indicates good performance, validation with independently collected data shows a different pattern. Model outcomes may also differ considerably depending on the approach. Using an ensemble of different models seems the way forward.