Understanding oscillations in variogram

[Amelia J]

Hi there,

First off – thank you for this very helpful forum; I've learnt a lot reading through the questions and hope that my question can also provide value to other readers.

I have an individual (non-domestic Canis) that shows these oscillations up to ~60 days (shown at 50% and daily resolution):

For reference, below is his movement activity, organised by ‘time’. The spot on the far left is a mine site with anthropogenic food resource that the individual travelled to on multiple occasions:

1. I'm not sure how to put into words an interpretation for the oscillations, particularly in a biologically meaningful way. With the upwards sloping SVF it looks like either the individual has not yet reached a stable residency, or there is insufficient data.

2. Workflow to investigate the oscillations: 

I compared the DOF and – as I understand it – the Degrees of Freedom needed to estimate area for the OUF anisotropic model is 44.9. Using the sampling period/τ[position] equation, my data has 47 DOF. Does this suggest that I do have sufficient data to estimate home range area, and thus the upward slope likely indicates the individual hadn’t reached a stable residency by the time data collection stopped?

Finally, I checked the periodogram as the plots suggests some periodicity in the data. The peaks indicate that there may be irregularities in the sampling period more so than periodicity in movement.

My tentative conclusion is that the oscillations are due to sampling irregularities based on the periodogram, but I'm not sure what the next step would be here in that case. Perhaps remove the data from the mine site and retry analysis, even though that wont resolve the sampling irregularities?

Thank you!

Amelia

[Amelia J]

For what it's worth, upon reading other posts and the recommendation to segment the data using segclust2d, I separated the data into two segments and re-ran the variogram. 

This is state 1 (I did not cluster, just created segments): 

And this is state 2:

I've concluded based on this process that when segmented, the individual exhibits a range residency in one of the segments with a spurious daily periodicity arising from an autocorrelated sampling schedule. In the other segment, where the individual visits a mine site with periodicity (as per the variogram) there is also spurious daily periodicity but no range residency. 

Is this a fair interpretation of the work process to reach a conclusion, and in particular I remain unsure about whether my understanding of using the sampling period/τ[position] equation and the 'Degrees of Freedom needed to estimate area' is accurate.

Cheers,

Amelia

[Christen Fleming]

Hi Amelia,

You can run through the periodic model selection analysis in vignette('periodogram'). This can distinguish between behavioral and artifactual periodicities. I suspect the second cluster has a true periodicity related to back-and-forth motion between the left spot and the right area. The period is pretty long, but under a month. There might be some artifactual daily periodicty on top of that, but it doesn't explain the larger oscillations. You will want to eye-ball this period and provide an initial guesstimate. Newer versions of the package can estimate the period parameter, which is useful in cases like this, where its not something simple like the day or month.

Best,

Chris

[Amelia J]

Thanks Chris! I'll take a closer look at the model selection aspect of the periodograms :)

I have another question but it's somewhat unrelated so I'll post in a new post.