Dynamic occupancy models and abundance model equivalent

[Michelle Gibson]

Hi,

I'm new to unmarked and am wondering which type of model to use for my data given what I want to do with it. I've read through the unmarked overview, website resources, and some of the original publications but I'm still a bit unsure.

Aim:

What I want to be able to do is test whether species track environmental conditions over space and time, specifically relating site-based environmental variables with species abundance/occupancy. I'd also like to know the extent of species' movements as a measure of extinction or colonization probability, and of course I want to account for imperfect detection. For these reasons, the 'colext' function seems like the way to go, although I have abundance data so could also an equivalent abundance model.

The study system/data:

The study region is arid and the surveys began following an extreme rain event, so there are some massive influx/efflux species dynamics, especially opportunistic species, which I'm particularly interested in. For example, in the wettest year, the opportunistic species can be distributed so that there are thousands of individuals at just a few sites and none in the next year, so are very patchy, whereas resident species are usually found in lower numbers across many sites. The study period spans multiple years, so that fact paired with highly mobile species leads me to assume this data would require an 'open' or 'dynamic' model. The data were collected each year during winter from 2012-2015 at 150 fixed-location sites (multi-season, repeat surveys), so that each of the 150 sites is surveyed once a year. Sites are located 16km apart, and at each site, 8 different line transects and 7 different point counts are conducted in succession, although I'm currently only working with the line transect data (see figure below): 

It took approximately 1-1.5 hours in total to survey a site, so presumably occurrence state doesn't change. We used distance sampling methods to record all bird species that we see and/or hear (no distance cut off), # of individuals, and exact distances from transect line. We also noted double counts of individual birds that are recorded at more than one line transect at a single site. 

Questions:

1.) Since dynamic occupancy models ('colext') require replicate surveys at a site, I was wondering whether each line transect at a site could be counted as an independent replicate (with each year as a season) given that we accounted for double counted individuals? If not, is there an alternate way to still use colext with my data? 

2.) Finally, given that I have abundance data, it would be nice to do some kind of equivalent model to a dynamic occupancy model using species counts. It looks like there are three options for dynamic abundance models in unmarked: 'gmultmix', 'gdistsamp', and 'pcountOpen'. pcountOpen might be good as it is designed for repeated count data and includes parameters for recruitment and apparent survival, similar to colext, but I don't understand enough about the other two models to rule them out. 

Any advice you can give on using 'colext' to model occupancy and using one of the three abundance model options for my data would be greatly appreciated!

Thanks!
Michelle Gibson

[Michelle Gibson]

Sorry, looks like the figure didn't work. See attached 

[John Clare]

Hey Michelle,

1. ) I think the answer is maybe (or 'it depends'). The new Applied Hierarchical Modeling book (vol. 1; Kery and Royle) goes into some detail regarding using spatial replicates in lieu of temporal replicates and the use of transects for sampling (chp 10), and that might be a good place to start looking.  It kind of depends upon what is being sampled (degree of species movement/level of dependence between transects, detection radius off of the transect, and so forth).
2. ) Gmultmix and Gdistsamp implement models that are kind of analogous to multi-method multi-scale occupancy models, in that there is a 3 level hierarchy with an added parameter between population size and detection probability that corresponds to availability (individuals temporarily leave a site, or hide underground and can't be detected, etc.). Dail-Madsen type distance sampling has been developed and is also covered in the AHM book (chp. 9); this might be the best way to approach the problem (you might be justified in pooling all transects, if nothing has been double counted, and avoid having to deal with the sampling structure).  

John

[Michelle Gibson]

Hi John,

Thanks a lot for your reply. I actually have the AHM book so will read more about it.

Cheers,
Michelle

--
You received this message because you are subscribed to a topic in the Google Groups "unmarked" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/unmarked/qO8MZnuy6dM/unsubscribe.
To unsubscribe from this group and all its topics, send an email to unmarked+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.