Uncovering animal movement decisions from positional data Jonathan Potts, Postdoctoral Fellow, University of Alberta, September 2013
From decision to data
Movement
From decision to data Direct interactions
From decision to data Mediated interactions
From decision to data Environmental interactions
From decision to data
Movement: correlated random walk Example step length distribution: Example turning angle distribution:
The step selection function Fortin D, Beyer HL, Boyce MS, Smith DW, Duchesne T, Mao JS (2005) Wolves influence elk movements: Behavior shapes a trophic cascade in Yellowstone National Park. Ecology 86:
Example : Amazonian bird flocks Potts JR, Mokross K, Stouffer PC, Lewis MA (in revision) Step selection techniques uncover the environmental predictors of space use patterns in flocks of Amazonian birds. Ecology
Hypotheses
Maximum likelihood technique
Resulting model Step length distribution Turning angle distribution Canopy height at end of step Topographical height at end of step
Coupled step selection functions Potts JR, Mokross K, Lewis MA (in revision) A unifying framework for quantifying the nature of animal interactions Ecol Lett
Unifying collective behaviour and resource selection Potts JR, Mokross K, Lewis MA (in revision) A unifying framework for quantifying the nature of animal interactions, Ecol Lett
Collective/territorial models: from process to pattern Giuggioli L, Potts JR, Harris S (2011) Animal interactions and the emergence of territoriality, Plos Comput Biol, 7(3):e
Collective/territorial models: from process to pattern Deneubourg JL, Goss S, Franks N, Pasteels JM (1989) The blind leading the blind: Modeling chemically mediated army ant raid patterns. J Insect Behav, 2, Giuggioli L, Potts JR, Harris S (2011) Animal interactions and the emergence of territoriality. Plos Comput Biol, 7(3):e Vicsek T, Czirok A, Ben-Jacob E, Cohen I, Shochet O (1995) Novel Type of Phase Transition in a System of Self-Driven Particles. Phys Rev Lett, 75,
Coupled step selection functions Resource/step-selection models: Detecting the mechanisms Model 1 Model 2Model 3Model 4 Positional data
Detecting the territorial mechanism: the example of Amazonian birds
Amazon birds: space use patterns
Interaction vs. no interaction
Classical mechanistic modelling Use maths/simulations to show: Process A => Pattern B
Classical mechanistic modelling Use maths/simulations to show: Process A => Pattern B Observe pattern B
Classical mechanistic modelling Use maths/simulations to show: Process A => Pattern B Observe pattern B Conclude process A is causing B
Classical mechanistic modelling Use maths/simulations to show: Process A => Pattern B Observe pattern B Conclude process A is causing B Logical fallacy: A=>B does not mean B=>A
Classical mechanistic modelling Use maths/simulations to show: Process A => Pattern B Observe pattern B Conclude process A is causing B Logical fallacy: A=>B does not mean B=>A Guilty! Potts JR, Harris S, Giuggioli L (2013) American Naturalist
New approach Use maths/simulations to show: Process A => Pattern B
New approach Use maths/simulations to show: Process A => Pattern B Observe process A
New approach Use maths/simulations to show: Process A => Pattern B Observe process A See if pattern B follows
New approach Use maths/simulations to show: Process A => Pattern B Observe process A See if pattern B follows If not, process A is insufficient for describing data: i.e. need better model
New approach Use maths/simulations to show: Process A => Pattern B Observe process A See if pattern B follows If not, process A is insufficient for describing data: i.e. need better model Contrapositive: A=>B means not-B=>not-A Correct logic
Amazon birds: space use patterns
How close is a movement model to reality?
How close is a movement model to data?
Try to mimic regression approaches
Look at the residuals Zuur et al. (2009) Mixed effects models and extensions in ecology with R. Springer Verlag “Residual”: the (vertical) distance between the prediction and data
More complicated than regression
Earth mover`s distance: a generalised residual
How to use the Earth Mover`s distance Simulated movement in artificial landscape with two layers:
Earth mover`s distance and direction
Wagon wheels
Wagon wheels of Earth Mover`s distance: include direction
Dharma wheel
Dharma wheels of Earth Mover`s Distance
Using simulated data with a = 1.5, b = 0 x-axis: value of layer 1 y-axis: earth mover`s distance (EMD) Left: EMD from model with a = b = 0 Right: EMD from model with a = 1.5, b = 0
A scheme for testing how close your model is to “reality” (i.e. data) Suppose you have N data points
A scheme for testing how close your model is to “reality” (i.e. data) Suppose you have N data points Simulate your model for N steps and repeat M times, where M is nice and big
A scheme for testing how close your model is to “reality” (i.e. data) Suppose you have N data points Simulate your model for N steps and repeat M times, where M is nice and big For each simulation, generate the Earth Movers distances to give M dharma wheels
A scheme for testing how close your model is to “reality” (i.e. data) Suppose you have N data points Simulate your model for N steps and repeat M times, where M is nice and big For each simulation, generate the Earth Movers distances to give M dharma wheels Each spoke of the dharma wheel then has a mean and standard deviation (SD)
A scheme for testing how close your model is to “reality” (i.e. data) Suppose you have N data points Simulate your model for N steps and repeat M times, where M is nice and big For each simulation, generate the Earth Movers distances to give M dharma wheels Each spoke of the dharma wheel then has a mean and standard deviation (SD) Generate a dharma wheel for the data
A scheme for testing how close your model is to “reality” (i.e. data) Suppose you have N data points Simulate your model for N steps and repeat M times, where M is nice and big For each simulation, generate the Earth Movers distances to give M dharma wheels Each spoke of the dharma wheel then has a mean and standard deviation (SD) Generate a dharma wheel for the data If any spoke of the data dharma wheel is not of length mean plus/minus 1.96*SD from the simulated dharma wheel then reject null hypothesis that model describes the data well
Normalised earth mover`s distance
Acknowledgements Mark Lewis (University of Alberta) Karl Mokross (Louisiana State) Marie Auger-Méthé (UofA) Phillip Stouffer (Louisiana State) Members of the Lewis Lab
Movement and interaction data Mathematical analysis Simulations/IBMs Coupled step selection functions Conclusion “To develop a statistical mechanics for ecological systems” Simon Levin, 2011 Spatial patterns
Thanks for listening!