Defining a "visit" in TTD and RN for ARU data

[Logan Clark]

Hey Folks,

I have a question about how I should thin/organize my ARU data (to define a "visit") to fit time-to-detection and Royle-Nichols models in unmarked for abundance.

I have ARUs at 100+ sites listening for ruffed and dusky grouse. Each has a 12min cycle of recording of 4mins on and 8mins off (I originally chose 4 mins because ruffed grouse drum every ~4mins). ARUS turned on at 5am and off at 11am. Each ARU has ~2,000 4min recordings before it filled its SD card. Peak activity for both species is is 5-630am and really drops off after that. 

If I chose a day as a visit and each 4min file as a binned time interval for TTD, it would take very few, if any, intervals before a grouse was detected at nearly any site with at least one grouse. 

If I chose the 4min file as a visit, and divided it into 8 (30sec) intervals, i feel that would be more representative of site abundance for TTD. But I also feel that  2,000 visits and zero inflation would make for a bad fit. 

I know there are many other options for defining a visit and am hoping some folks on here have some insights.

I have calendar date, time after sunrise, and avg noise level (from python librosa package) for each 4min file.  

FYI I am relatively new at this stuff and am a MS student. 

Thank for reading

--

Logan B. Clark, AWB

cell: (231)-409-3655

[Marc Kery]

Dear Logan,

first of all, I CC HMecology back in, since that way more people can benefit and more can help, too. Couple comments back:

• What the data are and what a model estimates are two different things. For instance, the RN model takes a binary detection/nondetection response and yields an estimate of underlying abundance. This sounds like magic and often it works, but sometimes also less well (in the sense of sometimes the resulting abundance estimates make sense and sometimes not). The RN model depends more heavily on its assumptions than does an abundance model that uses more information, such as the binomial Nmix model for replicated counts or multinomial Nmix for spatially replicated capture-recapture data.
• What we call TTD models exist both for an underlying abundance- and for an underlying presence/absence state. Trying to estimate abundance from TTD data is more risky than simply estimating presence/absence, again in the sense that more can go wrong when assumptions are less well met (i.e., estimates can be unreasonable).
• unmarked function occuTTD() can be used to model either underlying presence/absence or abundance. The change can be effected by chosing linkPsi = 'cloglog' (which yields a model for underlying abundance) or linkPsi = 'logit', which results in a model for occupancy.
• About 2000 visits per site for TTD modeling: to be honest I don't even know whether occuTTD() accepts repeated measurements for a closed, or single-season, model. Have to check. If not, may have to go to BUGS software.

Best regards  --- Marc

---

From: Logan Clark <clark...@gmail.com>
Sent: Wednesday, January 29, 2025 21:07
To: Marc Kery <marc...@vogelwarte.ch>
Subject: Re: Defining a "visit" in TTD and RN for ARU data

 

Marc thanks for the response!

I am not necessarily trying to compare RN to TTD estimates. I anticipate likely having to define a visit differently between the two. I am using those two because Fiss et al. 2024 showed that they work for ARUs and are comparable to N-mixture estimates from human observer point counts. They chose to have their visits sort of match their point count visit structure, which I do not plan on doing. My task is to compare and contrast the efficiency of ARUs vs human observers for spring monitoring grouse, and I'm hoping to thin the data as little as possible to leverage the temporal strength of ARUs.

I also conducted human observer point counts at my ARU sites and estimated abundance in unmarked's n-mixture function pcount(). I would now like to derive abundance estimates from my ARU data also in unmarked. 

I thought RN model response was a simple detection/non-detection per visit?

So for the TTD model you don't foresee any issue with 2,000 visits per site? And time in seconds to the first detection as the response (i.e no binning)?

I will check out the Goldstein paper now.

Really appreciate your response here,

Logan

On Tue, Jan 28, 2025 at 3:38 AM Marc Kery <marc...@vogelwarte.ch> wrote:

Dear Logan,

you seem to want to compare the two models for abundance estimation, right ? And you seem to want to process your data identically for both ?

I don't think the latter is necessary. The natural way to get the TTD data is just to take the difference in the time stamps between first detection and start of recording. There is no need to bin here; this would just lose accuracy.

However, for a method that uses discrete-time detection data, such as the classical RN model, you have to make the usual choice of what consists an occasion. In making this choice, you must balance the wish to not lose information (which happens when you have more than 1 detection during an occasion, but that then gets recorded as a 1) on the one hand, and to not have too many occasions on the other, for numerical reasons, and because a too fine temporal discretization may lead to problems with temporal autocorrelation in the detections. For ideas about how to discretize continuous recording data, see the recent paper by Goldstein et al.

Best regards  — Marc

---

From: hmec...@googlegroups.com <hmec...@googlegroups.com> on behalf of Logan Clark <clark...@gmail.com>
Sent: Monday, January 27, 2025 20:00
To: hmec...@googlegroups.com <hmec...@googlegroups.com>
Subject: Defining a "visit" in TTD and RN for ARU data

 

--
*** Three hierarchical modeling email lists ***
(1) unmarked: for questions specific to the R package unmarked
(2) SCR: for design and Bayesian or non-bayesian analysis of spatial capture-recapture
(3) HMecology (this list): 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 "hmecology: Hierarchical Modeling in Ecology" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hmecology+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/hmecology/CAD%2B5yyb6XjR4UQJKGS5OJEhd7t%2B1W6yi8AwTC%2BqSMJFgLe9Jjg%40mail.gmail.com.