Report

How close is a complex movement model to reality? Jonathan Potts, Marie Auger-Méthé, Karl Morkoss, Mark Lewis Some history Some history One outcome: • Often conclusions are made based on relative fit between models • How about absolute fit? One outcome: • Often conclusions are made based on relative fit between models • How about absolute fit? • Suggestion: look at the residuals in step-length and turning angle distributions How do we extend the idea of a residual to more complex movement models? Generic movement kernel: probability of moving to x at a time τ in the future given that the agent is currently at y and arrived there on a bearing θ and is travelling through environment E is (|, , ) How do we extend the idea of a residual to more complex movement models? Generic movement kernel: probability of moving to x at a time τ in the future given that the agent is currently at y and arrived there on a bearing θ and is travelling through environment E is (|, , ) Study species e.g. topography e.g. food distribution Actual move Earth mover`s distance: a generalised residual Earth mover`s distance: a generalised residual () Ω , , | − 0 | • is the actual place the animal moves to Standardised earth mover`s distance Ω , | − 0 | 2 , − = Ω , Ω How close is your model to the data? Test the following null hypothesis: H0: “The data is indistinguishable from a stochastic realisation of the model” Power test on simulated data Researcher knows about layer 1 C(x) T(x) Researcher doesn’t know about layer 2 Power test on simulated data Researcher knows about layer 1 C(x) Researcher doesn’t know about layer 2 T(x) “Real” process: ∝ − − + −() Best model: ∝ − − + Power test on simulated data Potts JR, Auger-Méthé M, Mokross K, Lewis MA. A generalized residual technique for analyzing complex movement models using earth mover's distance. Methods Ecol Evol DOI: 10.1111/2041-210X.12253 Where is my model failing? Patterns in EMD Acknowledgements Mark Lewis (Alberta) Marie Auger-Méthé (Dalhousie) Karl Mokross (Louisiana State) Conclusion • Want to know how good your model is in an absolute rather than relative sense? Conclusion • Want to know how good your model is in an absolute rather than relative sense? • Use EMD for the best results.