Estimating density from camera trap data using occuRN and pCount

[Annika Zuleger]

Hello everyone,

sorry, if this question was asked before, but I couldn't seem to find the answer to it.

We are trying to get density estimates from camera trap data using pCount as well as occuRN (to compare those two methods with estimates from previous studies using distance sampling and literature values). 

As far as I understand it the abundance estimates we receive from the two models are per sampling unit, i.e. per camera trap location. Now we would like to have an estimate of the abundance per km² to be able to compare it to previous estimates. 

Our sampling design is a systematic 8 x 8 grid, with 500m spacing between cameras. So for each 1 x 1 km grid cell we would have 4 camera traps deployed in that grid. 

Any suggestions on how I can get from an average abundance of e.g. 3.36 [± 0.38] individuals per camera to an estimate of the abundance per km²?

If you need any further information on the sampling design, please let me know.

Thank you very much!!

Annika

[Marc Kery]

Dear Annika

 

Yes, the abundance estimates are per site, which in your case is some moderately well-defined area around each camera trap.

 

I am not a specialist of the camtrap literature, and I am sure that people have tried what you try before, so perhaps there is some reasonably precedent. However, personally I think that this is impossible without making several pretty heavy assumptions, especially about the magnitude and variance of activity ranges. The following paper does not exactly describe your case, but has a lot of wisdom in it: https://esajournals.onlinelibrary.wiley.com/doi/full/10.1890/ES11-00308.1

 

Any dissenting opinions from the camtrap community ?

 

Best regards -- Marc

--
*** Three hierarchical modeling email lists ***
(1) unmarked (this list): for questions specific to the R package unmarked
(2) SCR: for design and Bayesian or non-bayesian analysis of spatial capture-recapture
(3) HMecology: for everything else, especially material covered in the books by Royle & Dorazio (2008), Kéry & Schaub (2012), Kéry & Royle (2016, 2021) and Schaub & Kéry (2022)
---
You received this message because you are subscribed to the Google Groups "unmarked" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/3131efce-854a-4ea3-98b1-7a0df9e7af60n%40googlegroups.com.

[Ken Kellner]

I agree with Marc that it is very difficult if not impossible to determine the sampled area around a camera trap or other similar detector (at least using detection/non-detection data only), which makes getting an estimate of density problematic.

Because the "sampled area" is something you as a researcher can (and/or must) arbitrarily define, even if basing it on home ranges or something, there is a danger of ending up with an unscientific process of repeated modification of this value until you get a "reasonable" density estimate.

Here are a few additional papers that address this topic:

https://esj-journals.onlinelibrary.wiley.com/doi/abs/10.1002/1438-390X.12028
https://www.biorxiv.org/content/10.1101/2022.11.23.517687v1.abstract
https://conbio.onlinelibrary.wiley.com/doi/abs/10.1111/cobi.13517

Ken

> To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com<mailto:unmarked+u...@googlegroups.com>.
> To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/3131efce-854a-4ea3-98b1-7a0df9e7af60n%40googlegroups.com<https://groups.google.com/d/msgid/unmarked/3131efce-854a-4ea3-98b1-7a0df9e7af60n%40googlegroups.com?utm_medium=email&utm_source=footer>.

>
> --
> *** Three hierarchical modeling email lists ***
> (1) unmarked (this list): for questions specific to the R package unmarked
> (2) SCR: for design and Bayesian or non-bayesian analysis of spatial capture-recapture
> (3) HMecology: for everything else, especially material covered in the books by Royle & Dorazio (2008), Kéry & Schaub (2012), Kéry & Royle (2016, 2021) and Schaub & Kéry (2022)
> ---
> You received this message because you are subscribed to the Google Groups "unmarked" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.

> To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/ZR0P278MB08696B0F0178C58C4DF295F3EBD6A%40ZR0P278MB0869.CHEP278.PROD.OUTLOOK.COM.

[Quresh Latif]

Here are a couple more papers to add to the list. Sounds like these begin to
offer solutions, but to my understanding they still require defining an area
sampled by the camera:

Nakashima, Y., K. Fukasawa, and H. Samejima. 2018. Estimating animal density
without individual recognition using information derivable exclusively from
camera traps. Journal of Applied Ecology 55:735–744.

Nakashima, Y., S. Hongo, and E. F. Akomo-Okoue. 2020. Landscape-scale
estimation of forest ungulate density and biomass using camera traps:
Applying the REST model. Biological Conservation 241:108381.

Quresh S. Latif
Research Scientist
Bird Conservancy of the Rockies
Phone: (970) 482-1707 ext. 15
www.birdconservancy.org

-----Original Message-----
From: unma...@googlegroups.com <unma...@googlegroups.com> On Behalf Of Ken
Kellner
Sent: Tuesday, October 17, 2023 9:20 AM
To: unma...@googlegroups.com
Subject: Re: [unmarked] Estimating density from camera trap data using
occuRN and pCount

I agree with Marc that it is very difficult if not impossible to determine
the sampled area around a camera trap or other similar detector (at least
using detection/non-detection data only), which makes getting an estimate of
density problematic.

Because the "sampled area" is something you as a researcher can (and/or
must) arbitrarily define, even if basing it on home ranges or something,
there is a danger of ending up with an unscientific process of repeated
modification of this value until you get a "reasonable" density estimate.

Here are a few additional papers that address this topic:

https://link.edgepilot.com/s/a0c1ec6b/EhfVLrGEpkuYUVnpHh5yKQ?u=https://esj-journals.onlinelibrary.wiley.com/doi/abs/10.1002/1438-390X.12028
https://link.edgepilot.com/s/2fc880ce/TH0XE3VCe0SmmED63kfIWQ?u=https://www.biorxiv.org/content/10.1101/2022.11.23.517687v1.abstract
https://link.edgepilot.com/s/54121571/4yVylgBkq0SDsXTB-ULYWQ?u=https://conbio.onlinelibrary.wiley.com/doi/abs/10.1111/cobi.13517

Ken

On Tue, Oct 17, 2023 at 03:02:38PM +0000, Marc Kery wrote:
> Dear Annika
>
> Yes, the abundance estimates are per site, which in your case is some
> moderately well-defined area around each camera trap.
>
> I am not a specialist of the camtrap literature, and I am sure that people
> have tried what you try before, so perhaps there is some reasonably
> precedent. However, personally I think that this is impossible without
> making several pretty heavy assumptions, especially about the magnitude
> and variance of activity ranges. The following paper does not exactly
> describe your case, but has a lot of wisdom in it:

> https://link.edgepilot.com/s/01de6793/SUnuPWpcokO-rKk44i9FhA?u=https://esajournals.onlinelibrary.wiley.com/doi/full/10.1890/ES11-00308.1

> https://link.edgepilot.com/s/f57206c2/lx_zb3_2ik2wJPxnCqXIaw?u=https://groups.google.com/d/msgid/unmarked/3131efce-854a-4ea3-98b1-7a0df9e7af60n%2540googlegroups.com<https://link.edgepilot.com/s/4946a4bb/kHjTsI-fv0qwnhY6pmvNEQ?u=https://groups.google.com/d/msgid/unmarked/3131efce-854a-4ea3-98b1-7a0df9e7af60n%2540googlegroups.com?utm_medium=email%26utm_source=footer>.

>
> --
> *** Three hierarchical modeling email lists ***
> (1) unmarked (this list): for questions specific to the R package unmarked
> (2) SCR: for design and Bayesian or non-bayesian analysis of spatial
> capture-recapture
> (3) HMecology: for everything else, especially material covered in the
> books by Royle & Dorazio (2008), Kéry & Schaub (2012), Kéry & Royle (2016,
> 2021) and Schaub & Kéry (2022)
> ---
> You received this message because you are subscribed to the Google Groups
> "unmarked" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to unmarked+u...@googlegroups.com.
> To view this discussion on the web visit

> https://link.edgepilot.com/s/d03d1609/0xgw2-Avn0W5FefVsFmy6Q?u=https://groups.google.com/d/msgid/unmarked/ZR0P278MB08696B0F0178C58C4DF295F3EBD6A%2540ZR0P278MB0869.CHEP278.PROD.OUTLOOK.COM.

--
*** Three hierarchical modeling email lists ***
(1) unmarked (this list): for questions specific to the R package unmarked
(2) SCR: for design and Bayesian or non-bayesian analysis of spatial
capture-recapture
(3) HMecology: for everything else, especially material covered in the books
by Royle & Dorazio (2008), Kéry & Schaub (2012), Kéry & Royle (2016, 2021)
and Schaub & Kéry (2022)
---
You received this message because you are subscribed to the Google Groups
"unmarked" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to unmarked+u...@googlegroups.com.
To view this discussion on the web visit

https://link.edgepilot.com/s/8258870c/W0H5DFohu0apkf7r2uzoVg?u=https://groups.google.com/d/msgid/unmarked/ZS6mBwvNi0cHR0hZ%2540COYOTE.

[John C]

Hi all,

It has seemed to me that a reasonable way to get at density with CT's using something in unmarked might be to use something like distsamp, gmultmix, gdistsamp, etc. For "distsamp", maybe one could use motion triggered cameras with all detections summed over time and sorted into distance bins by camera, with a time-related offset to get at something close to an instantaneous density (N per area over a second or other small temporal unit), although I can't say whether the function would require some re-jiggering for a conical sampling area. For gdistsamp or gmultmix, could get the multinomial counts replicated over time/space with time-lapse images and distance bins (or multiple cameras taking simultaneous time-lapse images of the same [measured] viewshed), and derive D as lam*phi/area. No idea what sort of sampling parameters would be necessary to get something useful--probably a lot of 0 counts--but I kinda think these approaches might be conceptually preferable to some of the commonly used alternatives. Probably wouldn't want to place every camera on a road/trail, and would probably want to use this as a state predictor in some way.

Cheers,

John

[Jose Jimenez Garcia-Herrera]

Dear Annika, Marc, Ken, Quresh, John,…

 

This is another paper that is related to the topic:

 

https://doi.org/10.1016/j.gecco.2015.01.010

 

Best regards,

José

.
To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/ZR0P278MB08696B0F0178C58C4DF295F3EBD6A%40ZR0P278MB0869.CHEP278.PROD.OUTLOOK.COM.

[chris nagy]

We have been trying to find a way to count unmarked white tailed deer for a long time, and this year we tried the Space-to-Event (STE) and the Instantaneous Sample (IS) methods by Moeller et al described here:

https://esajournals.onlinelibrary.wiley.com/doi/full/10.1002/ecs2.2331

It is not occupancy-  or unmarked-based, so I am going a bit off topic, but it has worked well for us. I found its logic/assumptions to be much more reasonable and justifiable than the other unmarked animal density estimators we have tried (we've tried a lot).

It uses the area sampled/photographed by the camera trap rather than trying to estimate the distance to each individual animal in different photographs. This is a requirement for a lot of these estimators, but we found a way to make calculating this "viewshed" area easier by turning the area from a cone to a triangle. We used an obvious, straight and perpendicular boundary feature some distance away from the camera, for example, a fallen (and mostly straight) log or a fence edge, and setting the camera up 15 - 20 meters away facing that boundary. When looking at the pictures, it is easy enough to tell whether an animal (deer in our case) is within or beyond the boundary. Anything beyond the boundary is not counted. 

You will still need to know the viewable angles coming out of the camera, so you can calculate how wide the photograph is at a given distance, and then how much area the triangle in front of the camera is. We did this on our driveway with a camera, a bunch of small traffic cones, and some geometry. 

Important to note is that the method works best with time-lapse photos, not motion-triggered. If you use time lapse data, you remove nearly all the issues with detection, because the camera is taking a picture at a certain interval (eg, 15 minutes) no matter what. This essentially acts like a set of plots that you search every 15 minutes, and you count the number of animals you get per that total area. The set of time lapse searches across all your cameras acts as repeated searches. (This is the IS method, basically, the STE method is a little different)

As you can imagine, most of your time lapse photos will be empty. But with a fairly common species like WTD we got plenty of photos, to my pleasant surprise. For about 800 acres we used 25 cameras running for 2 weeks. But if you're doing a rarer species, you will likely need a ton of cameras.

To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/DB7PR01MB42666ACB8F2B2A0B72958AE592D6A%40DB7PR01MB4266.eurprd01.prod.exchangelabs.com.

[aw...@scenichudson.org]

Like Chris, we've been exploring IS and STE methods for using cameras to estimate abundance of white-tailed deer.  As Chris suggests, this method is really best suited for common species, not rare ones.  Moeller et al have published a couple very practical papers, one on estimating detection area for a camera, and a step-by-step paper for working through analyses with their R package, including accounting for data already collected using motion-detection instead of time-lapse triggered cameras.  I'd be curious to see comparisons of these vs some unmarked hierarchical models with different species and habitats, but the assumptions with these are beautifully straightforward.  I like Chris' idea of using a barrier object in the camera frame to simplify detection distance in camera views even more.

Moeller, A. K., S. J. Waller, N. J. DeCesare, M. C. Chitwood, and P. M. Lukacs. 2023. Best practices to account for capture probability and viewable area in camera-based abundance estimation. Remote Sensing in Ecology and Conservation 9:152–164.

Moeller, A. K., and P. M. Lukacs. 2022. spaceNtime: an R package for estimating abundance of unmarked animals using camera-trap photographs. Mammalian Biology 102:581–590.