Testing the transferability of track‐based habitat models for sound marine spatial planning

Aim

Species distribution models (SDMs) are statistical tools aiming at mapping and predicting species distributions across landscapes. Data acquisition being limited in space and time, SDM are commonly used to predict species distribution in unsampled areas or years, with the expectation that modelled habitat–species relationships will hold across spatial or temporal contexts (i.e., model transferability). This key aspect of habitat modelling has major implications for spatial management, yet it has received limited attention, especially in the dynamic marine realm. Our aims were to test geographical and temporal habitat model transferability and to make recommendations for future population‐scale habitat modelling.

Location

Two contrasted regions of the North Western Mediterranean Sea: the cold and productive waters of the Gulf of Lion, and the warm and oligotrophic waters of Corsica.

Methods

We GPS‐tracked 189 Scopoli's shearwaters, Calonectris diomedea, at four breeding sites during the chick‐rearing period in 2011 and 2012 (418 foraging trips), and analysed their fine‐scale foraging behaviour. We then built colony‐specific habitat models (GAMMs) to test SDM geographical and temporal transferability and investigated the effect of extrinsic (environmental extrapolation) and intrinsic (trip characteristics) factors on transferability.

Results

Scopoli's shearwaters from our four study sites had comparable foraging strategies (as assessed from trip characteristics and isotopic diet tracers). Despite such similarities, SDMs revealed colony‐specific habitat associations. Geographical and temporal model transferability was better within than between regions.

Main conclusions

Crucially, our study illustrates how habitat–species relationships can vary between colonies located <200 km apart, and underlines the effect of spatio‐temporal extrapolation in habitat modelling. We therefore warn that defining adequate spatial scales for model predictions is critical to sound marine spatial planning and conservation.