Large dataset of very different species - error calibration using prior?

[Thomas Lauber]

Hi Chris 

First, thank you so much for this incredible package and all the knowledge you put out here! 

My question is regarding error calibration and simulations: 

I have a very big dataset from Movebank, for which I want to understand (i) how much time does an animal spend in a habitat and (ii) how often does it cross habitat borders. For this, I want to simulate movement paths using ctmm and the given telemetry data.

For now, I 

• Filter for individuals with > 30 observations 
• Following the error vignette, I assign an error prior with uere() <- X for X=1 (10) for eobs (gps); then, I set the DOF to 2
  
• Stepwise fitting of a CTMM using error=TRUE
• Get the best performing model 
• Run 400 simulations (~5% error), given the best model and the telemetry data 
• Lastly, calculate the home range using AKDE 

Does this workflow make sense to you? And do you think its a good idea to do the fitting using my error prior, or would it be better to do it without (error=FALSE)? Unfortunately, I do not have calibration datasets and the quality of the data can obviously be very different. I am modelling movements from Elephants to tiny birds, so also no consistency here. 

I would really appreciate getting your thoughts on this! 

Again, great great work with this package! 

Best, 

Tom 

[Christen Fleming]

Thanks Tom,

That makes sense to me. Recent versions of the package should also provide a point estimate guess for the error prior, so you shouldn't have to adjust more than the DOF.

I think specifying a good prior is pretty safe relative to having no error model, if there is some chance that you could have small sampling intervals in the data.

Best,

Chris