[R-sig-eco] simm.levy in adehabitat

[Clément Calenge]

Hi Tim,


Tim Sippel wrote:
> Hi again-
> Clement, thanks very much for your response to my question a few days ago
> and revised code for the simm.levy function.  Ultimately what I am aiming to
> do is partition animal trajectories into distinct behavioural segments using
> the functions modpartltraj, bestpartmod and partmod.ltraj.  In the function
> mopartltraj one is required to define a reference model of animal movement
> (argument called 'limod') to compare to a real trajectory (arguement 'tr').
> The example 'limod' given with the functions assumes a fairly simplistic
> normal distribution with different means for different behaviours.  However,
> could the 'limod' argument be from a simulation of a movement process such
> as a levy walk (simm.levy) or multi-variate Ornstein-Uhlenbeck Process
> (simm.mou)?  
>   

Until now, we mainly used this function to partition a trajectory into 
segments based on the (disputable) hypothesis that the step length is 
drawn from a normal distribution. We chose to model the step length as a 
normal distribution because this is a common choice for the step length 
distribution in animal movement analysis (but see Morales et al. 2004, 
Ecology, 85, 2436-2445, for arguments against this choice). This 
approach gave us interesting results: for example, on the help page of 
the function modpartltraj, we partition the trajectory of a porpoise, 
and this partitioning shows that this trajectory, that I would have 
partitioned visually into 3 segments ("patchy" moves, then more "linear" 
moves, then again "patchy moves"), is more likely built by 4 segments 
("patchy" moves, then more "linear and slow" moves, then "linear and 
fast" moves and finally again "patchy" moves). This approach helps to 
see patterns in the data...

Actually, we did not try any other model (as noted on the help page, the 
method is still under research), but theoretically, the method can be 
used with other models, provided that each chosen model enable you to 
estimate a probability (density) for each step of the actual trajectory. 
Actually, the argument 'limod' should be a list of functions, each 
function giving the probability density of a step with given parameters 
(length, orientation, etc.) under a given model (see the help page of 
modpartltraj).

So it is theoretically possible (though untested) to pass to the 
argument 'limod' a list of functions giving the probability density of a 
given step according to the MOU or the Levy walk with specified 
parameters (these functions implementing the formulae found in the 
literature on these movements models). But these functions are not yet 
available in adehabitat and the user has to program them... 
Alternatively, these functions may theoretically rely on simulations to 
compute these probabilities (using some kind of kernel smoothing to 
estimate the probability density of the steps from the simulations of a 
given model, or something similar)... but IMHO this approach would first 
require to be carefully tested...
Hope this helps
Cheers,


Clément.

-- 
Clément CALENGE
Office national de la chasse et de la faune sauvage
Saint Benoist - 78610 Auffargis
tel. (33) 01.30.46.54.14