Suitability of N-mixture Models and/or Courses about using Unmarked?

[Julia Braga Morais]

Hi group!

As part of my analysis, I’m exploring the use of N-mixture models, but I would really appreciate your advice on their suitability given the structure of my datasets, as in reality I only have 1 site to monitor. Here's a brief overview of the three monitoring methods I’m comparing:

1. Transect-based spotlight counts

   • One site divided into 4 transects

   • 10 surveys per year (with double counts by citizen scientists—potentially treated as 20)

   • Monitored over 3 years

2. Nest occupancy surveys

   • 295 sites (nests) 

   • 5 repeated surveys per year

   • Conducted over 2 years (occupancy by citizen scientists for 1 year only)

3. Thermal video surveys

   • One site - 1 long boat transect

   • 15 survey nights per year, over 2 years

   • Video footage can be reviewed multiple times, allowing repeated counts of the same event

My key question is:
Are N-mixture models appropriate for this kind of structure, particularly when spatial replication is limited (as in the first and third methods)? Also, I’m curious whether repeated counts from different observers or from video replays can meaningfully improve detectability estimates under these models.

Additionally, if you’re aware of any courses, workshops, or labs where I could deepen my understanding of hierarchical models for ecological monitoring (especially N-mixture and occupancy models), I’d be very grateful for any recommendations.

Thank you very much for your time and consideration—I truly appreciate any guidance you can offer.

Best regards,
Julia Morais
PhD Candidate
Monash University

[Jeffrey Royle]

hi Julia,

 I would not recommend N mixture models for (1) simply because there are too few sites.   I think study (3) it would be potentially ok if you chop the transect into smaller spatial units.  Since the replicates are repeated views of the same video, you don't have to worry about the closure assumption and using sub-units of the transect make a lot of sense (so "yes" to your key question there).  Then you have sampling over time to worry about too which is a different matter. You might consider treating this temporal replicates as "units" too and then doing what we call "stacking the data" (look around for what that means...). 

 We may be doing a workshop in Oregon this fall, so keep following the unmarked list for information.

regards

andy

--
*** 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/a0f19a96-a720-484d-a30b-e06590e20f4an%40googlegroups.com.

[Julia Braga Morais]

Dear Andy,

Thank you so much for your reply—it truly means a lot to hear directly from you. Your insights were incredibly helpful, and I appreciate the clarity you brought to each of the scenarios I’m working with.

Regarding Method 1, I actually have a bit more structure than I initially described. Each transect was surveyed in pairs, with counts recorded separately from the left and right sides. I had been summing these for simplicity, partly because we were tracking a biological observational covariate (the proportion of birds in a particular behavioural state), which was calculated per transect rather than per side. But now, thinking in terms of spatial replication, I realize I could potentially treat these as 8 separate sites instead of 4. Do you think that would improve the model's suitability in this case?

I’ll also explore “stacking the data” as you suggested—it seems like a very promising approach to deal with the temporal aspect, as perhaps doing a time-for- space substitution would require more years of surveys.

Lastly, I’d be very keen to attend the Oregon workshop if there’s any possibility of joining virtually. I’ll definitely keep an eye on the unmarked list for updates.

Thanks again for your generosity in sharing your time and advice.

Best regards,
Julia Morais
PhD Candidate

Monash University – Australia

To view this discussion visit https://groups.google.com/d/msgid/unmarked/CAE%2BwgF39CDSiBUzKgd5%3DDAOkx3b_bsi9Dkj9-K17A95xOmPkKA%40mail.gmail.com.

[Jeffrey Royle]

hi Julia,

 Thank you for your kind words.

 Re: the use of transect halves as replicates.... this is a potentially good idea although it raises some modeling questions related to potential dependence among the units and the effect of that. I would be cautious about that in general but especially with a very small sample size, and 8 isn't too many anyhow!  Sample size requirements for these models are not too different from occupancy models and I generally tell people they might work well with 50 or more sites depending on the objective, complexity of the model and so forth.  There are a lot of simulation studies published (see the original paper from 2004 and also the AHM1 book by Marc Kery and myself).  I just don't think you have enough data in that case study ...  But you should explore the model as a way of learning about its performance and limitations -- simulation is a great exercise to learn about a model and also to learn about your system.

 Re: the workshop, -- we're just in the very early stages of planning right now but, to be honest, I doubt highly that we would have an online platform. We don't have a lot of support to do these workshops and that requires technical support at a significant level and, in addition, it would diminish the in-person vibe and also a lot of people would choose that option instead of in-person, which further diminishes the workshop atmosphere.  

regards

andy

To view this discussion visit https://groups.google.com/d/msgid/unmarked/CADHtTxKGGac1-1ybBVry%2BBwoukvpQTqfbNeG%3Dmq5Zd-z%3DKpE_A%40mail.gmail.com.

[Julia Braga Morais]

Hi Andy,

Thank you so much for getting back to me again!

Apologies for circling back, but I had one more (possibly naïve) question about replication. You mentioned that for the video method, temporal replicates could potentially be treated as sampling units—essentially stacking them as "sites" within a year. Would something similar be valid for the small dataset from the spotlighting method too?

In that case, citizen scientists conducted repeat counts at the same 8 chopped sites, about 20 minutes after the expert survey, and from a different direction. This was repeated across three years. The transects are continuous, and the species is penguins—so they’re mobile, but not dramatically so, and although they respond mildly to observers, we include a 20-minute pause between surveys/observers to let them settle back into baseline behavior.

I was wondering if it would be reasonable to treat those second surveys as additional spatial units—i.e., stack them in the same way, making it 8 sites × 2 observer directions × 3 years = 48 spatial units. Would that be conceptually okay as a way to explore the model, or do you think the dependence and movement between adjacent units (especially in continuous space) would invalidate that approach? My goal at this point isn’t necessarily to produce robust ecological inference for management action, but rather to learn how the model behaves with data that stretch or challenge assumptions a bit—and also to compare different sampling methods for this unusual urban colony.

As for the workshop—totally understand, and I really appreciate the importance of the in-person format. I’ll do my best to attend! Though I have to say, suggesting an online version felt a bit like modeling: maybe not perfect, but sometimes it’s valid to try and see what happens! In the meantime, I’ll be trimming the video files and getting everything prepped.

Thanks again for your patience and generosity in answering these questions. I'm just trying to learn and understand the tools better, and your insight really helps.

Best regards,
Julia

To view this discussion visit https://groups.google.com/d/msgid/unmarked/CAE%2BwgF0TUNQxqCXZm%2BftZsS44-9ahTJnijZfJRbh7ami9jApVg%40mail.gmail.com.

[Jeffrey Royle]

hi Julia,

 Yes I think you can use a "time for space" substitution (temporal replicates as sample units) in general but it's important to think carefully about the interpretation and also you must still have true replicate samples of each 'unit'. I'm not sure I would feel comfortable recommending it for your spotlighting survey, there seems to be a lot going on there, but it might work out depending on what your questions are. I guess from the standpoint of exploring the limitations of the method and understanding efficacy, you might as well try it out and see how things work out.

 Good luck!

regards

andy

To view this discussion visit https://groups.google.com/d/msgid/unmarked/CADHtTxKhU5n3ReQ07%3D5pZRRRwkcD489gQprm7wA%3Dr390rBjutQ%40mail.gmail.com.

[Anam Ahsan]

Does any one know how to run multiple species single season model 

Anam 🍀

To view this discussion visit https://groups.google.com/d/msgid/unmarked/CAE%2BwgF2ZZOD%2BxiiRM2bTXzayQ2C5%3DwFeSp3FG1q_rHq8nbwu1A%40mail.gmail.com.

[Marc Kery]

Dear Anam,

for multispecies Nmix model, you'd normally use one of the Bayesian engines, JAGS, NIMBLE or even Stan. However, it should be possible to fit this model with function pcount() in unmarked, by stacking data for all species and adopting species-level random effects.

Best regards  --- Marc

---

From: unma...@googlegroups.com <unma...@googlegroups.com> on behalf of Anam Ahsan <anam.a...@gmail.com>
Sent: Tuesday, April 29, 2025 19:13
To: unma...@googlegroups.com <unma...@googlegroups.com>
Subject: Re: [unmarked] Suitability of N-mixture Models and/or Courses about using Unmarked?

 

To view this discussion visit https://groups.google.com/d/msgid/unmarked/0D047EDD-0763-4D23-89B1-2F253A409D9E%40gmail.com.

[Anam Ahsan]

Hi 

I need help in occupancy estimation 

Like I find the top model 

With three site covariates . I also calculated confidence interval Amon’s det and state covariates also calculated the beta coffieent . No I don’t know how to find psi values for this model . Like do we get one psi value or different psi values Todd different site covariates. 

Please help to know how to exactly get psi values . 

What is the code 

Anam 🍀

On Apr 25, 2025, at 11:08, 'Julia Braga Morais' via unmarked <unma...@googlegroups.com> wrote:



To view this discussion visit https://groups.google.com/d/msgid/unmarked/CADHtTxKhU5n3ReQ07%3D5pZRRRwkcD489gQprm7wA%3Dr390rBjutQ%40mail.gmail.com.

[Anam Ahsan]

Please any one help me to find how to exactly find psi values for a model . It’s urgent 

Anam 🍀

On May 24, 2025, at 23:15, Anam Ahsan <anam.a...@gmail.com> wrote:

Hi