Which method for multi-point, multi-visit, multi-year sampling?

559 views
Skip to first unread message

Max Henschell

unread,
Jan 26, 2016, 3:54:18 PM1/26/16
to unmarked
Hello unmarked community-

I am investigating the effects of human disturbance on bird communities in protected areas. I would like to estimate density/abundance for each species recorded at 27 locations over 3 years. I collected point count data 3 times per year at 3 point counts within each of the 27 locations. During point counts, I recorded distance and subdivided my 10 minute count into 2 5 minute periods, recording new individuals in the second sampling period. I also have several detection covariates (e.g. date of sample, time of sample, wind speed, temperature) and a few site-level covariates (e.g. forest structure, recreational trail width, amount of trail use) that I would like to test whether they are associated with different densities/abundance at each location.

My biggest question is which unmarked method would be most suitable for my data. I think I have a few options since I have both distance (gdistsamp?) and removal sampling (gmultmix?) methods. The multi-point, multi-visit, multi-year aspect is most unclear to me as to which method(s) to investigate and how to proceed.

It looks like stacking multi-year data is common practice with gdistsamp, so "sites" become "sites x year", correct? Is there a way to incorporate the multiple counts/visit into this? Combine counts within a visit into one "super-visit" perhaps? Or could I use 9 secondary sampling periods(3 points x 3 visits/year), and 3 primary sampling periods (3 years)?

If I go the gmultmix route, I am unsure on how to organize the data. Perhaps 18 secondary sampling periods(2 time periods x 3 points x 3 visits/year), and 3 primary sampling periods (3 years)?

Any advice/guidance would be greatly appreciated!

Max Henschell

Jeffrey Royle

unread,
Jan 26, 2016, 6:44:34 PM1/26/16
to unma...@googlegroups.com
hi Max,
 I recommend the use of gdistsamp with the "stacked" data structure.  In this case I would pool all of the data from the consecutive replicates for each visit and make sure to use an offset which is number of replicates (or adjust things afterwards).  I would not "stack" visits that are consecutive in time because they should be extremely highly correlated.

 Note: For the next release of unmarked we plan to include a Dail-Madsen version of the gdistsamp (called distsampOpen). This will handle the general data structure that you have and allow for fairly complex dynamics.

 For analysis of these data in gmultmix there should be examples of data organization in the package help files , but the basic idea is you need to have "nsites x nreps" data matrices and then cbind these together:   (Y1, Y2, ... , Ynyears).

 In volume 2 of the AHM book (due next year....) we plan to provide some more detailed examples of gmultmix()

 In a future release of unmarked we plan to provide a Dail-Madsen version of gmultmix also.
 
regards,
andy


--
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.
For more options, visit https://groups.google.com/d/optout.

Max Henschell

unread,
Jan 28, 2016, 10:07:13 AM1/28/16
to unmarked
Hi Andy-

Thank you for your suggestion. I'll proceed with gdistamp.

Is there an ETA for the next unmarked version with distsampOpen?

Max Henschell

Jeffrey Royle

unread,
Jan 28, 2016, 7:38:20 PM1/28/16
to unma...@googlegroups.com
the earliest for our next release which may or may not include distsampOpen is probably early April. 

regards
andy


--

Nicole Michel

unread,
Feb 12, 2016, 5:53:29 PM2/12/16
to unmarked
Hi Andy and Max,

I've got a follow-up question on this. I'm also working with data collected that includes removal (three uneven time periods), distance sampling (three distance bands, the last unlimited radius), and two within-year repeat visits (one month apart during the breeding season, so likely violating the closure assumption). It looks like unmarked could handle at most two of these components at once (e.g., gmultimix for removal + repeat visits, gdistsamp for distance + repeat visits) but not all three, is this correct? Will distsampOpen be able to accommodate data with both time and distance bands?

I've read your (Andy's) 2014 Auk paper with Courtney Amundson and Colleen Handel combining removal and distance sampling in JAGS. Have you considered expanding this to repeat within-year visits, either with or without closure, or have any suggestions of how best to go about this? If there's no one "magic bullet" yet which would you recommend: running removal+distance separately on each visit, running gmultimix (discarding distance info), or running gdistsamp (discarding removal info)?

Thanks in advance!

Cheers,
Nicole

Jeffrey Royle

unread,
Feb 12, 2016, 6:34:14 PM2/12/16
to unma...@googlegroups.com
hi Nicole,
 if we ever get distsampOpen working, it will only handle distance sampling data with a population dynamics model, not mixed DS and multinomial data (note: this is something that would be good to have at some point but we don't have the time to do it; if anyone in the community has an interest in writing such extensions then this is something we'd be interested in).

 Regarding the Amundson et al. paper, I haven't seen any example of TR/DS with with repeat visits within a year. However, I think with all the examples in chapter 9 of the AHM book (Kery and Royle 2015), it would be easy to put this together (if you're a decent BUGS programmer).  If you do this (or if anyone does it) please let us know, we'll include it in volume II of the book.

 Short of doing that, I'm not sure what I would recommend in terms of collapsing the data.

 General note:  it seems to me that any two multinomial protocols could be combined into one big multinomial protocol and then the gmultmix function would work.   DS is of course a type of multinomial protocol. Can it be done? With some skillful hacking I think. And of course a custom piFun.

regards,
andy

Reply all
Reply to author
Forward
0 new messages