Occupancy modelling with data from spatially clustered sampl

Forum for discussion of general questions related to study design and/or analysis of existing data - software neutral.

Occupancy modelling with data from spatially clustered sampl

Postby goncalo-cs » Sun Nov 23, 2014 9:04 am

Hi everyone, I am currently working on a project focused on occupancy patterns of a carnivore species.

Our data comes from camera-trap surveys conducted originally for other studies, with different objectives.
In total we have detection histories for 169 sites, distributed in 11 study areas. While distance between different areas can be more than 100km, sites within each area are spatially clustered and are not independent as distance between camera-trap stations is around 1km, a short distance considering the vital area of our target species.

Originally we used single season occupancy models to identify main factors (environmental covariates) influencing occupancy probabilities considering all the stations in the analysis. However, we are unsure on the validity of our approach since we do not include any explicit spatial processes on the modelling procedure.
We are aware that due to the lack of independence among stations within the same study area we are actually looking at the probability of use rather than probability of true occupancy. Nonetheless, we do not fully understand the consequences of clumping together two very different sampling scales.

Therefore we would really appreciate any opinion on the best way to proceed with this kind of data. Are single season occupancy models a valid approach or is there any better method that accommodates spatial relations between trapping stations? I was suggested to include an additional factor with 11 levels (one per area), however I don't think my data can support that many parameters. Another option might be to try to include a random effect of study area in WinBUGS, but I fear I might have the same problem.

Thank you in advance,
Gonçalo Curveira Santos
goncalo-cs
 
Posts: 1
Joined: Tue Nov 18, 2014 9:12 am

Re: Occupancy modelling with data from spatially clustered s

Postby ferny » Wed Nov 26, 2014 6:22 pm

I would also be interested in this as a site I work on has I think a similar problem.

Goncalo, FWIW my understanding is this:
1) you can have, as a rule of thumb, up to 1 parameters per 10 sites you sample - so I don't know whether another model or covariate might be useful to you, but that would be the guide as to whether you can do it or not.

2) I think independence of sites is a fundamental assumption of the models. I've come across models which allow various assumptions to be broken but none which allow for this one. By and large I think it is a problem which needs to be dealt with in the study design phase (which wasn't possible for you given that you've collated data from different studies) and I'm not aware of a post-hoc solution (which absolutely is not to say there isn't one, I'd also love to hear if there is!). I don't know what the consequences of non-independence would be on the interpretation of your data.
ferny
 
Posts: 7
Joined: Wed Oct 23, 2013 3:41 pm

Re: Occupancy modelling with data from spatially clustered s

Postby katyg » Tue Aug 22, 2017 10:45 am

ferny wrote:I would also be interested in this as a site I work on has I think a similar problem.

Goncalo, FWIW my understanding is this:
1) you can have, as a rule of thumb, up to 1 parameters per 10 sites you sample - so I don't know whether another model or covariate might be useful to you, but that would be the guide as to whether you can do it or not.

2) I think independence of sites is a fundamental assumption of the models. I've come across models which allow various assumptions to be broken but none which allow for this one. By and large I think it is a problem which needs to be dealt with in the study design phase (which wasn't possible for you given that you've collated data from different studies) and I'm not aware of a post-hoc solution (which absolutely is not to say there isn't one, I'd also love to hear if there is!). I don't know what the consequences of non-independence would be on the interpretation of your data.


I have a similar situation with 4 grids separated by a minimum of 150 m ranging to 3.5 km (measured from closest edge of grid). 10 camera stations within each grid are spaced 35-70 m apart. This is part of a study assessing effects of wind energy development on black bear beechnut habitat use. Behavioral responses, movement, and habitat use are already being assessed through GPS collared individuals. The primary purpose of the camera data is to evaluate black bear activity within 1 km of the wind energy development footprint pre- and post-development. Camera data using this setup has already been collected for the pre-development phase, from Oct-Dec 2016 and Apr.-May 2017, when bears are known to forage in beechnut stands. My question moving forward pertains to the extremely spatially autocorrelated data inherent in this study design. Given the focus on black bears, there is no hope for independence of sites with camera stations spaced at 35-70 m, even if bears are only using these sites for a couple of months. I cannot think of a way around this huge violation, except for a complete redesign of the camera grids and spacing, at which point I would be left with just the post-construction data to make use of. This might not be a big deal, since I could always compare data from 2 grids set up in proximity to the wind energy development and 2 grids away from it (in similar habitat of course). Alternatively, could I use the pre-construction data as an index of activity (I am assuming this is similar to use rather than true occupancy) and compare activity levels pre- and post-construction? I am assuming the huge independence of sites violation still makes it impossible to reach biologically meaningful conclusions regarding activity levels though.....I would appreciate any ideas on this! I am trying to think of useful ways that this camera study can supplement the GPS collar data, but am stumped by the spatial autocorrelation, and have trouble justifying another camera spacing that may suffice if the current camera design setup is completely overhauled. given extremely variable food driven movements of black bears (ranging from ~2 - >100 sq. km).
katyg
 
Posts: 2
Joined: Mon Aug 21, 2017 5:29 pm


Return to analysis & design questions

Who is online

Users browsing this forum: No registered users and 1 guest

cron