Uniting Statistical and Individual-Based Approaches for Animal Movement Modelling

1. What they did

Predicting where animals will go when landscapes change needs models that account for their internal state, such as hunger and memory, which are hard to measure in the field. The authors combined a statistical model of habitat selection with an individual-based simulation model, using GPS data from 27 woodland caribou in winter in the managed boreal forest of Québec.

They replayed each real caribou’s movements in a simulated landscape to calculate hidden variables (energy gained from lichen against energy spent moving, energy stored over three days, and memory of previously visited patches), then fitted a step selection function using those variables plus canopy cover, altitude and distances to roads and cuts. They checked the model by cross-validation and by comparing simulated and real habitat selection and home ranges, then ran it on the current landscape and on one where 70% of each 225-km² block became mixed and deciduous forest.

2. What they found

• The model with memory, energy and environmental variables explained caribou movements best.
• Caribou moved back towards previously visited patches, chose steps with a high ratio of energy gained to energy spent, preferred lower canopy cover and higher altitude, and avoided roads and regenerating cuts.
• When their energy over the previous three days was high, caribou selected open, low-cover areas even more.
• The model was robust in cross-validation (rs = 0.94), and simulated caribou chose land cover types much like real ones; simulated home ranges were slightly smaller than real ones.
• In the hardwood encroachment scenario, simulated caribou selected fixed open areas, burned areas and water less, and selected mixed and deciduous forest, which they had avoided in the current landscape. Home range sizes did not differ between scenarios.

3. What they concluded

Using a statistical model to set the rules of an individual-based model, and generating hidden internal-state variables by replaying real movements, gives a robust model that can predict how animals will respond to landscapes not yet observed. The simulated change in habitat selection with hardwood encroachment could expose caribou to more wolves, because more mixed and deciduous forest is likely to mean more moose and therefore more wolves.

7. Implications for conservation (from the paper)

• Woodland caribou are threatened in the Canadian boreal forest, most likely because of human activities such as forest harvesting. Models like this can test the long-term effects of forest management plans before they are applied, including the risk of increased wolf predation through more moose.