multi-method models implemented in 'unmarked'?

[Alan Hitch]

Hello all,

I was wondering if single-season multi-method models have been implemented in the 'unmarked' package yet. I know they are in Presence and Mark.  I looked in the package literature but didn't find anything and just wanted to check here to see if I missed anything.

We surveyed for bats with mist net arrays paired with acoustic detectors and would like to model average detectability of each method for occupancy estimates.

Thanks for your help.

Alan Hitch

[Marc Kery]

Dear Alan,

would help to have a cite for the model. Do you mean Nichols et al. (2008) or Miller et al. (2011) ? If the former, then the answer is "no", if the latter then it is "yes" (this is called 'occuFP').

From what you say though I think it's the former. ----  But I wonder whether you can't just put all of your data into a single detection history for each site ? E.g., if you distinguish 10 occasions for mist-nets and 50 for the ARUs, then each detection history would have 60 occasions. You could then define an obsCov "method" that has value 1 for the first 10 occasions and value 2 for the remainder (or else code this two-group comparison as 0/1). Fitting that "method" covariate into p should give you an estimate of the per-occasion p's for mist nets and ARUs. Note that to make them perhaps as comparable as possible, you'd have to define the occasions in the same way for both, i.e., to have the same duration.

Or am I missing anything here ?

Best regards  --- Marc

---

From: unma...@googlegroups.com <unma...@googlegroups.com> on behalf of Alan Hitch <flaming...@gmail.com>
Sent: Monday, December 13, 2021 23:48
To: unmarked <unma...@googlegroups.com>
Subject: [unmarked] multi-method models implemented in 'unmarked'?

 

--
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/a49b77f0-1a1c-4e4e-b8f3-0312839fce8en%40googlegroups.com.

[Alan Hitch]

Hello Marc,

Thanks for the response.  You were correct I was referring to Nichols et al. (2008).  Thanks for the suggestion on combining the detection data and scoring each method as an obsCov.  I'll try that.  We structured the sampling design so both methods were surveying for the same amount of time.  They each have the same number of occasions.

Thanks again.

Alan

[gcsadoti]

Sorry, old question, but the multi-method/scale model of Nichols et al. (2008) can be run in RMark. Send me a pm if you need suggestions.

Giancarlo

[Sylvain Reydellet]

Hi everyone, 

I was wondering if there have been any news about the implementation of Nichols et al. (2008) multi-method model in unmarked. If not, @gcsadoti could tou give me any advice regarding the use of Mark ?

Thank you all

Sylvain

[Ken Kellner]

Hi Sylvain,

The goccu() function has been fairly recently added to fit the basic multi-scale model described in Nichols et al. (2008). I think whether this would work for multi-method data depends on your exact experimental design.

Ken

[Sylvain Reydellet]

Hi Ken, thanks for the so quick reply

I have 60 camera traps arranged in a grid, each cell is 1.5 x 1.5km. I'd like to define 4 or 6 cells as a single site. This site would then be occupied by 4 or 6 cameras, which would constitute my different detection methods.

Would it be interesting for the goccu() function ?

Thanks

Sylvain

[Ken Kellner]

Yes I think it would work. You'd have 4-6 of what unmarked calls "primary periods" per site, but rather than thinking of these as temporal replicates, they are spatial replicates (cells). Then within each "primary period"/cell presumably you'd have multiple detection periods.

unmarked will estimate an "availability" probability which would be the between-cell probability of detection, and "detection" probability which would be the within-cell detection probability.

Ken

[Sylvain Reydellet]

Okay I'll try this, thanks.

Even though I'm not sure to see how unmarked will deal with the "primary periods" that are really "primary site".

Sylvain

[Reed Scott]

Hi all,

I have been reading through this thread as well as the goccu() documenation, and had a question come up about the parameter theta. I am just trying to determine if availability varies by subsite. What I'm wondering is if there's some sort of built in covariate for subsite. EX: mod <- goccu(psiformula =~ 1, phiformula =~ subsite, pformula =~ 1, umf, numPrimary = 1) OR is this something where I would have to code in a variable for each subsite (i.e. assign Subsite1,Subsite2,Subsite3). Any help would be greatly appreciated!

-Reed

[Ken Kellner]

There's no built in variation - if you want the availability parameter to vary you'll have to specify the appropriate covariates yourself.

I think whether you can vary availability specifically by subsite depends on exactly what you mean by that. For example suppose each site is divided into subsites designated A, B, and C (that is, site 1 has subsites A, B, C and site 2 also has subsites designated A, B, C). This could be something like direction (north subsite, east subsite, south subsite) or order sampled (first subsite sampled, 2nd subsite sampled, etc.) In this case you have multiple data points for each level of 'subsite' covariate (A/B/C/A/B/C) and you can estimate the associated parameters.

On the other hand suppose you are uniquely identifing each subsite across all sites. For example at site 1 you have unique subsites coded A, B, C, and at site 2 you have unique subsites D, E, F and so on. In this case your covariate for subsite would have just a single data point for each possible level of the covariate (A/B/C/D/E/F). I do not think you can estimate parameters for the subsite covariate because you just have 1 data point per level. Plus, even so, you'd be estimating way too many parameters (#sites * #subsites per site - 1) for the unmarked optimization to realistically handle.

See below for a quick example simulation.

Ken

library(unmarked)

# Simulate some goccu data
M <- 100
S <- 3
J <- 5

y <- matrix(NA, M, S*J)

# Example with common subsite codes across sites
ysc <- data.frame(subsite = rep(c("A","B","C"), 100))

head(ysc)

blank <- unmarkedFrameGOccu(y=y, yearlySiteCovs=ysc, numPrimary = S)

cf <- list(psi=0, phi=c(0, -0.5, 0.5), det=0)

umf <- simulate(blank, psiformula=~1, phiformula=~subsite, pformula=~1,
coefs=cf)[[1]]

# recovers estimates
goccu(~1, ~subsite, ~1, umf)

# Now suppose every subsite code is unique
# simulate new dataset
ysc <- data.frame(subsite = factor(1:(M*S)))

head(ysc)

blank <- unmarkedFrameGOccu(y=y, yearlySiteCovs=ysc, numPrimary = S)

# Need a parameter for all subsites (-1 for the intercept)
cf <- list(psi=0, phi=c(0, rnorm(M*S-1)), det=0)

umf <- simulate(blank, psiformula=~1, phiformula=~subsite, pformula=~1,
coefs=cf)[[1]]

# doesn't work both due to too many parameters and lack of information
goccu(~1, ~subsite, ~1, umf)

> >>>>>>>> ------------------------------
> >>>>>>>> *From:* unma...@googlegroups.com <unma...@googlegroups.com> on

> >>>>>>>> behalf of Alan Hitch <flaming...@gmail.com>

> >>>>>>>> *Sent:* Monday, December 13, 2021 23:48
> >>>>>>>> *To:* unmarked <unma...@googlegroups.com>
> >>>>>>>> *Subject:* [unmarked] multi-method models implemented in

> >>>>>>>> 'unmarked'?
> >>>>>>>>
> >>>>>>>> Hello all,
> >>>>>>>>
> >>>>>>>> I was wondering if single-season multi-method models have been
> >>>>>>>> implemented in the 'unmarked' package yet. I know they are in Presence and
> >>>>>>>> Mark. I looked in the package literature but didn't find anything and just
> >>>>>>>> wanted to check here to see if I missed anything.
> >>>>>>>>
> >>>>>>>> We surveyed for bats with mist net arrays paired with acoustic
> >>>>>>>> detectors and would like to model average detectability of each method for
> >>>>>>>> occupancy estimates.
> >>>>>>>>
> >>>>>>>> Thanks for your help.
> >>>>>>>> Alan Hitch
> >>>>>>>>
> >>>>>>>> --
> >>>>>>>> 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/a49b77f0-1a1c-4e4e-b8f3-0312839fce8en%40googlegroups.com

> >>>>>>>> <https://che01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgroups.google.com%2Fd%2Fmsgid%2Funmarked%2Fa49b77f0-1a1c-4e4e-b8f3-0312839fce8en%2540googlegroups.com%3Futm_medium%3Demail%26utm_source%3Dfooter&data=04%7C01%7Cmarc.kery%40vogelwarte.ch%7C4e447f38474049dd922808d9be8ef649%7C251ae8d19a904569a35b9d2bc41188d2%7C1%7C0%7C637750344501938479%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=IcsxFCAGE%2Fq9Dkjuf0z7ZUeAQU6IcG8XINwQUMo1gD4%3D&reserved=0>
> >>>>>>>> .
> >>>>>>>>
> >>>>>>>
>
> --
> *** 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 visit https://groups.google.com/d/msgid/unmarked/6e50475a-e31f-4505-8e5a-229164e9cf7an%40googlegroups.com.

[Reed Scott]

Hi Ken,

Thanks so much for your reply! That did the trick for my data. Although I had one quick clarifying question about numPrimary. It seems that what purpose numPrimary serves depends on the model being used (i.e. a single season multi-scale model vs a dynamic multi-scale model). In this case I am running a single season model with 3 subsites at each site, so do I set numPrimary = 3? And would this be implemented differently if this was a dynamic model? 

Thanks!

Reed

To view this discussion visit https://groups.google.com/d/msgid/unmarked/Z1IT7l05d_WbVGHK%40BADGER.

--

Reed Scott (he/him/his)

Email: rscot...@gmail.com

Phone: 716-381-1467

[Ken Kellner]

Reed,

You're right, for goccu numPrimary designates the number of subsites per site in this case, whereas for the dynamic model (colext) numPrimary is the number of years/time periods. It is probably not ideal that we use the same terminology for both when it can mean different things for different models.

Ken

> > https://groups.google.com/d/msgid/unmarked/Z1IT7l05d_WbVGHK%40BADGER.
> >
>
>
> --
> Reed Scott (he/him/his)
> Email: rscot...@gmail.com
> Phone: 716-381-1467
>

> --
> *** 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 visit https://groups.google.com/d/msgid/unmarked/CAEvTuUqt%2BoRAtDRLvct-WFmU4cdNRXu6GO1AxhEfZvxQUxyH8w%40mail.gmail.com.