SVF showing an unexpected pattern

[Romain Fernandez]

Hello Chris,

Thank you very much to you and Inês for all the resources you’ve created about the ctmm package!

I’m currently working with GPS data collected from frugivorous bats, with one location recorded every 15 minutes. The tags are turned off between 6 a.m. and 6 p.m and individuals were tracked for different periods of time.

After watching the tutorials and reading the discussions in this group, I’ve been trying to interpret my variograms and determine whether I can use the ctmm package to estimate 50% and 95% kernels for my individuals, as well as their mean kernel using mean(..., sample = TRUE).

However, I was surprised by the pattern of my variograms: the curve crosses zero periodically, and this pattern differs between individuals. Maybe this pattern is related to their behavior, for example, individuals might return to the same place at the end of the night. I checked whether any periodicity was present in the data, but it seems negligible. Could there be another possible explanation?

With this pattern, it’s difficult to determine whether the individuals are resident, because the asymptote of the variogram is hard to identify. I’m not sure whether these data can be analyzed with the ctmm package, or if I did not use the functions of the package correctly. I fitted the best models and plotted them on the variograms, and I’m sharing the results here. Do these results look correct to you?

Thank you very much for any time you can give me.

Romain

[Romain Fernandez]

Sorry, I forgot to attach the files. You can download it here: https://we.tl/t-A6UoHfoBVR

[Christen Fleming]

Hi Romain,

This is normal for a central place forager. Most people crop out a buffer around the central place to isolate the foraging data. The periodic gaps in the data will still cause daily oscillations in the variogram. What is important in this case is that the mean asymptotes.

Best,

Chris

[Romain Fernandez]

Hi Chris,

Thank you for your answer. We can't use the buffer approach because we don't know the locations of roost colonies for this species. I have several questions about my first analysis with the ctmm package:

1. Should I include periodicity using ctmm(mean = "periodic") in my case? I notice several peaks in the periodogram for some individuals, possibly due to their central-place foraging behavior (e.g. individuals Goupille and Gnackgnack).  

2. I would like to create one model per season (dry and rainy) for each individual, when I have sufficient data. Can I filter months across all individual tracks to create seasonal models, provided the residency assumption holds for each individual?

3. Could I apply the same approach (as described in part 2) to construct one model per individual based on a single point type (foraging or resting points categorized from accelerometry data), in order to generate separate “foraging” and “rest” kernels?

4. Individuals “Goupille” and “Gnackgnack” were tracked for most of a year. Can I plot a single variogram and fit one model per individual, or should I create separate variograms and models for each year?

Thanks for your help.

Romain

[Christen Fleming]

Hi Romain,

1. You can see if model selection will support that.

2. Yes, and that looks very feasible for your data.

3. Yes, and for flying CPFs this can be almost necessary, because the two states are too different for the switching process to modeled with a single stationary model.

4. If behaviors are similar, it probably won't make a difference, as the variogram is only used to generate an initial guess for the model fitting. In general, I recommend using the individual variograms if they are of good quality and a pooled variogram if they are not.

Best,

Chris

[Romain Fernandez]

Hi Chris,

Thank you! For the question 4, how can I create a pooled variogram?

I have a couple of final questions. I plotted the variogram for one individual tracked over two years (left plot below). After filtering out data from May to November in both years, I obtained a truncated variogram (right plot below). I think this is expected, as the variogram takes both years into account.

1. What would be the best practice to generate a proper variogram visualization in this case and to test the residency hypothesis?  
2. Will my initial guess for the model fitting be incorrect, or is it only a visual issue with the variogram plot?

Thanks for your help.

Romain