Using UBMS to account for missing year (primary) occasions in Dynamic Occupancy Models?

[cscjsm...@gmail.com]

Dear Unmarked Community,

      I am running a dynamic occupancy model with data from 2019, 2022, 2023 and 2025. I am aware that the unequal intervals/missing years in the primary occasion makes the colonization and extinction rates difficult to interpret and a bit problematic. I have read several posts that suggest trying to fit a model in Winbugs or JAGS, as these are more flexible and deal with the missing years better, but I cannot find any examples online (surprisingly). I remembered recently that UBMS is available to use Unmarked in a Bayesian framework and wondered if it was flexible enough to address this problem.

I tried fitting a model with NA's for the 2020, 2021 and 2024 years, and it produced reasonable results, including for a 'year' model (using each year as a factor to estimate colonization/extinction rates for each year). I am mostly interested in modeling covariates on extinction and colonization (less interested in the rates), and have been running into the problem that the main covariate of interest is a yearly-site categorical covariate, and I get an error about unable to run models when yearly-site covariates are missing data (for 2020, 21, 24).  

I am interested to know:

1. Is there a more elegant way than using NA's for the missing years to model unequal intervals in UBMS? Any suggestions on modifying priors?

2. Is there any way to model a yearly-site covariate with missing data in UBMS?

Thanks so much for creating an awesome Bayesian extension of Unmarked that is user-friendly for us ecologists who are intermediate with stats.

Chris

[Ken Kellner]

Hi Chris,

ubms is probably a little too aggressive in erroring for missing yearly-site covariates. I *think* that you should be ok to just fill in random values for the covariates in the missing years, and as long as the corresponding detection/nondetection data are NAs, they'll be ignored.

However I would be cautious in using ubms when you have entirely missing periods. I'm not sure that the way ubms handles this is exactly correct in all cases (see https://github.com/ecoverseR/ubms/issues/85). I haven't had time to explore this in depth yet.

If you don't absolutely need a Bayesian analysis, I would consider using unmarked instead, which should handle the missing values better.

Ken

> --
> *** 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/a997a04b-302e-4cb2-be31-7c1979daf778n%40googlegroups.com.

[Ken Kellner]

I was wrong about this, if an entire primary period is missing than the corresponding yearlySiteCovs will still be used, so inserting random values is a bad idea (in both ubms and unmarked). So there isn't an easy solution to this problem. Thanks Jim Baldwin for pointing out my mistake.

Ken