Questions regarding the ctmm

[Caka Karlsson]

Good morning Chris

I'm Caka and as a beginner in these analyses, I must thank you for a very easy package to work with. 

I have a couple of questions which I would be really happy if you could help me answer. 

First: My GPS dataset has their positions in lat and long i.e., degrees. Is there any need to first transform them to Km before using the ctmm package in order to get the right areas? If so have you any suggestions on how to do that? 

Second, this might be a stupid question but, why do we first fit the data to a model e.g. an UO model (or the model the package recommends) and then fit an AKDE? why not fit an AKDE directly to the dataset? Also, I have noticed that I don't have to specify a bandwidth, why is that?

Lastly, my study species is a migratory species which shows a strong home range behaviour on its breeding ground. Tho, in its non-breeding ground, it explores a lot and shows less of a home range residency. I want to calculate the area the animal is using in its non-breeding ground. Is this still possible to do with ctmm, if I cut the dataset to only include this time period? I was thinking if the data could be fitted to a Brownian bridge model which does not have a home range and then calculate the AKDE from that or is this a completely wrong thought?

Thank you so much 

Kindly 

Caka

[Christen Fleming]

Hi Caka,

1. Your data are automatically projected on import and you don't need to worry about it.
2. When you run akde() the bandwidth is optimized as a part of that calculation. And when the bandwidth is optimized, an autocorrelation model (e.g., OU) needs to be leveraged. Fitting the autocorrelation model is the slowest part of the calculation and is also used for other analyses like occurrence() and speed().
3. You do want to segment the data into the different behaviors. When the behavior isn't range resident, you can select among non-resident models (e.g., BM and IOU) if you need more parsimony because of limited data, but you cannot calculate AKDEs from those models. You can calculate many other quantities, though, like occurrence() and speed().

Best,

Chris

[Caka Karlsson]

Thank you so much Chris for taking the time to answer these questions. It has been really helpful.

Thanks a lot 

Kindly 

Caka 

[Caka Karlsson]

Hey again 

Sorry but now some new questions have come up. 

When I´m looking at the variograms of the different stationary areas for my study species, the variogram according to me doesn't go to an asymptote.  This must mean that the animal is not showing a home range behaviour, right? 

Tho, when I fit the data to a model, the best-recommended one is an OUF anisotropic model. Does this mean that I¨m mistaken and that the animal is showing a home-range behaviour cuz otherwise, a BM or IOU model should be the best fit?

Also, I wonder if I'm using occurrence, how do I know the values of cor.min and dt.max? My data is irregular with a mean sampling interval of one hour, but I have up to a 24 hour gap.

Thank you so much for taking your time with this

Kindly

Caka

[Christen Fleming]

Hi Caka,

No asymptote means either non-resident behavior or not enough data to see it.

The default AICc model selection does not compare resident and non-resident movement models - they do not have compatible likelihood functions, and you cannot reliably use AIC/BIC/LRT to choose between them.

The default values for those occurrence() arguments are pretty good. The resolutions are what you need to play with for good visuals (versus memory and computation time).

Best,

Chris