re: what to do about multiple recaptures in SCR-RSF?

27 views
Skip to first unread message

Asia Murphy

unread,
Jan 27, 2025, 4:44:17 PM1/27/25
to oSCR
Hello,

I'm trying to run a dependent SCR-RSF model, because some of the collared animals were captured in my scat samples. However, there are a few individuals that were captured more than once. The cap.tel code in the vignette only allows for 1 capture for each collared animal. Do I combine individual-specific lists of the row numbers into one final list (cap.tel)?

Thank you!

Daniel Linden

unread,
Jan 27, 2025, 9:21:10 PM1/27/25
to oscr_p...@googlegroups.com
Hi Asia, 

I’m sorry for the confusion, but collared individuals can have >1 capture. The index in “cap.tel” is the row of the capture history matrix/array that contains the trap captures for each collared individual, however many there are. This lets the model code line up the likelihood of the telemetry locations with the correct likelihood of trap captures.

Happy to help further!




On Jan 27, 2025, at 4:44 PM, Asia Murphy <asiaj...@gmail.com> wrote:

Hello,

I'm trying to run a dependent SCR-RSF model, because some of the collared animals were captured in my scat samples. However, there are a few individuals that were captured more than once. The cap.tel code in the vignette only allows for 1 capture for each collared animal. Do I combine individual-specific lists of the row numbers into one final list (cap.tel)?

Thank you!

--
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/108b328b-828b-4de5-b600-7cffec46bf8cn%40googlegroups.com.

Asia Murphy

unread,
Feb 3, 2025, 1:46:41 PM2/3/25
to oSCR
Thanks, I really appreciate it!

I did have a few more questions, after re-reading the vignette, the published papers, getting a model running, going through previously asked question on this forum...

1. How can I change the detection function to half-normal? Or is it default?
2. What exactly does trimS do? I'm assuming it provides a limit to the distance the model will search for the activity center of an individual.
3. Follow up to above: if I have spatially-varying covariates on detection, why can't I use trimS at all, as mentioned on this forum before? Is it because it creates an artificial asymptote for covariates?
4. Why do you use cloglog as the encounter model in the vignette instead of binomial?

This isn't a question, just an fyi: the "see Details" link in the R help file for oSCR.fit, under the model heading, where it provides the special characters that can be used to fit inbuilt models, is broken. It goes nowhere.

Daniel Linden

unread,
Feb 4, 2025, 12:37:58 PM2/4/25
to oscr_p...@googlegroups.com
Hi Asia, the R package could definitely use some TLC.  Weird that the details are missing for the primary function (!) but thanks for identifying that.  We really ought to flesh out the bookdown that Chris started, even just to document some basics.  Gabby's cheat sheet is also a great resource: oSCR Cheat Sheet

As for your questions: 

1)  The detection function is half-normal by default, where theta = 2.  We actually do not have other functions built in, largely because in practice we have not seen a huge motivation to use other types.  We started to build functionality for the power model but it is incomplete in oSCR (hence the ability to set theta to any value between 1 or 2, but this does not yet result in a reasonable detection function).  This could be easily remedied with proper motivation...

2)  trimS sets the local window of the landscape over which activity centers are evaluated, to speed up computation.  We have variable guidance on what this should be set to, but 3x mmdm (mean max distance moved) is one possibility.

3)  I think you could still use trimS but you need to be aware that it could create some problems.  But you are right, it could unreasonably constrain the estimation of spatial relationships for detection.  Best to test out with simpler models or coarser state-space resolutions first and then build up.

4)  The cloglog has the same link function (log) as the resource selection function.  There is not otherwise an equivalence between the beta coefficients that estimate covariate relationships between the encounter model and the RSF model.  The difference between the binomial and cloglog is not meaningful for most applications of SCR anyway (given typically low encounter rates), so this choice should not be very limiting.

Reply all
Reply to author
Forward
0 new messages