Optimising Habitat-Based Models for Wide-Ranging Marine Predators: Scale Matters

Predicting the responses of marine top predators to dynamic oceanographic conditions requires habitat-based models that sufficiently capture environmental preferences. Spatial resolution and temporal averaging of environmental data layers is a key aspect of model construction. The utility of surfaces contemporaneous to animal movement (e.g. daily, weekly), versus synoptic products (monthly, seasonal, climatological) is currently under debate, as is the optimal spatial resolution for predictive products. Using movement simulations with built-in environmental preferences (correlated random walks, multi-state hidden Markov-type models) together with modeled (Regional Oceanographic Modeling System, ROMS) and remotely-sensed (MODIS-Aqua) datasets, we explored the effects of degrading environmental surfaces (3km – 1 degree, daily – climatological) on model inference. We simulated the movements of a hypothetical wide-ranging marine predator through the California Current system over a three month period (May-June-July), based on metrics derived from previously published blue whale Balaenoptera musculus tracking studies. Results indicate that models using seasonal or climatological data fields can overfit true environmental preferences, in both presence-absence and behaviour-based model formulations. Moreover, the effects of a degradation in spatial resolution are more pronounced when using temporally averaged fields than when using daily, weekly or monthly datasets. In addition, we observed a notable divergence between the ‘best’ models selected using common methods (e.g. AUC, AICc) and those that most accurately reproduced built-in environmental preferences. These findings have important implications for conservation and management of marine mammals, seabirds, sharks, sea turtles and large teleost fish, particularly in implementing dynamic ocean management initiatives and in forecasting responses to future climate-mediated ecosystem change.