re: spatiotemporal autocorrelation an issue?

[Asia Murphy]

Hello, this is more of a theoretical question, but I just want to make sure my thinking is correct on this.

Let's say I have 100 traps that are 25 m apart and surveyed once a week for three weeks. Animal A is captured at Trap 1 and Trap 2 during Week 1 and Week 2, respectively. Normally, based on my species of interests' possible daily movements, I would thin out traps so I have 25 traps which are 100 m away so that the data isn't too spatially autocorrelated. In this case, I would have lost Animal A's Week 2 detection at Trap 2, because it's within <= 100 m of Trap 1.

But this time around I thin the traps out based on what week it is, so I keep Animal A's detection on Week 1 at Trap 1, and Animal A's detection on Week 2 at Trap 2, so now the two detections are within 25 m of each other, but separated by a week.

Would this cause the issues typical to ignoring spatial autocorrelation in SCR models? I feel that it would, but just want to check.

Thank you for your time!

[Jeffrey Royle]

hi Asia,

 In the context of SCR, I don't think any kind of arbitrary thinning of traps would cause any problems that aren't related to a loss of sample size which will have an effect on precision and possibly bias as well.  But any sort of autocorrelation in data that is related to spacing of traps will not cause any ill effects. In a sense, that autocorrelation is built into the model.

 I'm sorry if this doesn't answer your question exactly, will be glad to try again.

regards

andy

--
You received this message because you are subscribed to the Google Groups "oSCR" group.
To unsubscribe from this group and stop receiving emails from it, send an email to oscr_package...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/oscr_package/504d0902-2f49-4b64-89d0-b66b3d4a8c6cn%40googlegroups.com.

[Daniel Linden]

I started writing some thoughts but never pressed send.  I agree with Andy that trap spacing should not be used to thin data.  But temporal spacing can be important.

It is definitely a challenge dealing with technology where observations outpace our ability to model the data at fine spatial and temporal scales (e.g., camera traps, GPS).  Depending on your objectives, you can either make some data summarization choices to allow for inferences from the models you want to use, or you need to explore some advanced methods to make full use of the information in hand.

The treatment of animal movement in traditional spatial capture-recapture makes the assumption that individuals can traverse their entire activity area (e.g., home range) during each sampling occasion, so that the probability of using a location (e.g., trap or pixel) is simply a function of the location's distance from the activity center.  If your temporal sampling scale is small enough that the individual cannot possibly use certain parts of its activity area, then that is a problem.  The model would be attempting to calculate the likelihood of observations without knowing that some locations during some occasions have an effective probability of use that is 0.  Density estimation is pretty robust, but it becomes a bigger issue if you are exploring a resource selection function.

Ultimately your options depend on how the temporal and spatial scales of individual movement interact for your species.  We often benefit from the fact that daily movements for most wildlife species are far ranging enough that autocorrelation is not a problem for daily/weekly sampling occasions in capture-recapture.  If you had hourly GPS fixes and you wanted to integrate those into your SCR model, that's a different story.  Same for species that do not have a stationary home range.

On Mon, Apr 14, 2025 at 3:19 PM Asia Murphy <asiaj...@gmail.com> wrote:

--