variograms and model fits

[Sarah Hirsch]

Hello, 

I have been using the ctmm package to analyze the inter-nesting home ranges of 17 loggerhead sea turtles and have a couple of questions in regards to the variograms and model results.

(1) Most of the variograms do reach an asymptote, but I have four turtles whose variograms are much more subtle (example of two different turtles below). Would you say that these animals aren't resident animals and therefore a home range analysis is not appropriate for these individuals?

(2) When I run the model comparisons most turtles end up with a OUF anisotropic model as the top fit, but the variogram doesn't look great. Should I be concerned?

[Connor O'Malley]

Hey Sarah,

I have the same question regarding my mountain lion data. I would like a way to filter my data to only include residents so I can do estimations based on home range size. As far as I understand the ctmm package doesn't offer a way to do this besides visually inspecting the variograms like you're doing. So I look forward to hearing more on this from the experts! Thanks,
Connor 

[Christen Fleming]

Hi Sarah,

The top two variograms don't look very resident and the scales are much larger than the bottom variogram. It still might make sense to estimate their dispersal ranges, and present those estimates separately.

The bottom plot shows a pretty big discrepancy between the position autocorrelation timescale of the fitted model and of the empirical variogram. There could be a couple of culprits for this. Are these data sampled finely (on occasion) and/or do they have large location errors that need to be modeled? You said these were loggerhead sea turtles. This wouldn't happen to be Argos data, would it? That would definitely require an error model at 1-hour sampling.

Best,

Chris

[Christen Fleming]

Hi Connor,

There's no magic solution at present, unfortunately. There are a lot of different ways that individuals can be non-resident. If you have some a priori knowledge of dispersal events, then users have reported good success with segclust2d for objectively segmenting the data. If you have performed a ctmm range estimation analysis on multiple individuals and believe that some are resident and some are non-resident, then ctmm's new cluster() function can objectively classify those individuals, while propagating uncertainties, which is something that standard clustering methods can't handle.

Best,

Chris

[Connor O'Malley]

Great, thanks!

[Connor O'Malley]

Hey Chris,
I have another question along these lines. Is there a way to numerically quantify (as opposed to visual inspection) when the home range size asymptotes using the ctmm variograms? I want to run all my animals through a process like that and get an average for how many collar days it takes before the home ranges asymptote. Thanks so much!

Connor 

[Christen Fleming]

Hi Connor,

A model based approach would be the objective way to do this. At 3x tau[position], you are down to ~5% autocorrelation.

Best,

Chris