Dynamic multistate occupancy advice

[Tyler Pilger]

Hello,

I am seeking advice on using a multistate dynamic occupancy model framework. I have 39 sites in a river that were surveyed under a robust design. Surveys on 2 consecutive days in a month form secondary periods when occupancy state is assumed static. Sites were surveyed again after a few weeks, and these “monthly” survey events form the primary periods between which occupancy state changes. In years 1 and 2, there were four primary events spanning February to June, and in year 3 there were five primary events during these months. The target species is migratory and enters the river each year and, we hypothesize, increases the proportion of occupied sites as more individuals immigrate and move further upriver.

 We think that two different size groups (small and large) have different initial occupancy and colonization rates. The states are unoccupied (0), occupied by small and large fish (1), and conditional on occupied, occupied only by large fish (2). These states have nothing to do with reproduction, small fish being present is not a result of spawning but due to differential migratory behavior.

 Environmental conditions varied across years, which could lead to initial occupancy for states 1 and 2 to vary across years (or not). Similarly, colonization rates may vary across years as well as by state.

Initially, I ran separate dynamic occupancy models for each size class and each year using colext function from unmarked. However, a reviewer was not satisfied with this approach and suggested that I do not separate the dataset to dataset. Hence, the multistate framework to account for different size groups. But, I am unsure how to set this up so that I can test if initial occupancy or colonization probabilities vary across years. Any suggestions are welcome, as are recommended readings. I have not been able to find an analogous example since most studies only look at primary periods within a single year or treat years as the primary periods.

Thanks in advance.

[Marc Kery]

Dear Tyler,

 

here are some tentative thoughts; other thoughts, including dissent, very welcome 😊

 

I would see the benefit of a multistate model mostly in the case where you think there is either some interaction among the small and the large class of the fish or when you’re interested in the transition rates among the states among primary periods, or in rates of state assigment errors.

 

However, you seem to want to compare some of the parameters among the two classes of fish, and you don’t seem to be interested in interactions, transitions or assignment errors. Thus, one simplistic analysis would simply be a dynocc model stratified by size class. Thus, you stack the two data sets on top of each other and add a factor for “large” and “small” and then test for effects of that factor on whatever parameter you’re interested.

 

The only snag is that this analysis would do some pseudo-replication, i.e., ignore possible dependencies in the data due to the fact that each physical site (per year, visit) occurs twice in your response data. In the simplest case, this might perhaps just make your SEs/CIs a little too small and you could obtain adequate uncertainty intervals by some sort of non-parametric bootstrapping ? Or alternatively, you might fit random site effects into every parameter. This should now work in the latest versions of unmarked owing to the power of TMB.

 

I am not really sure about this, but I do think that multistate modeling is overkill for your research questions.

 

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/2ca562e6-d0fb-49a3-97e9-d58abb4ffdd4n%40googlegroups.com.

[aw...@scenichudson.org]

I may be off base here as I'm not very familiar with multi-species models, but if the interest is in the interaction between large and small fish, could you treat them as separate "species" and use some of the multi-species models that Rota et al. added to unmarked? -Alex

[Tyler Pilger]

Hello Marc,

Thank you for the advice. I was overthinking the problem which led me to think I needed a more complex model. I was able to run the dynocc model by stacking by years and size class and then used the nonparametric bootstrapping to acquire appropriate confidence intervals for each year by size combination. I was not able to run the random site effects model because when I did that with the colext function I got an error saying "This function does not support random effects". I had the latest version of unmarked installed (1.3.1) from CRAN but another version was published to CRAN on July 8 so maybe it works with that version.

For posterity or in case others stumble across this thread, I found more information about the "stacked" method for stratifying by year on this thread https://groups.google.com/g/unmarked/c/OHkk98y09Zo/m/YMOPqWXOBQAJ and in chapter 2 section 2.3.3 in AMH vol 2. This section also provides code to implement the stratified nonparametric bootstrap.

Best,

Tyler