Psi estimate of 1 for specific sites out of total sites?

questions concerning analysis/theory using program PRESENCE

Psi estimate of 1 for specific sites out of total sites?

Postby samundra » Wed Dec 21, 2016 7:52 am

Hi Phidot group,

I am working on snow leopard occupancy estimate using camera trap data of Nepal. I have 33 sites which are grids of 16 sq km and have one detector (camera trap) each in one site. Out of 297 occasions I have 21 occasions of capture. I used three covariates which are encounter rates of wild prey and livestock and the potential habitat of each sites in sq km. The top model came out to be a combination of Wild prey and potential habitat for occupancy and potential habitat for detection. The only problem I'm facing with is, in some of the sites out of 33 generated psi estimate of 33 sites, I have psi estimate of 1 with SE 0 and CI 0-1.

I looked at the data as well and generally those sites which have higher detection have psi value of 1 generated.

Is this a concern that I should be worried about? What should I do? Your words will be very valuable for me as I have looked over many literature and books but I still couldn't find this issue. Please forgive me if this has been discussed before as I looked into the earlier discussed questions as well but couldn't find the exact question and only related ones.

Thank you for your patience and time,
Best wishes,
Samundra
samundra
 
Posts: 1
Joined: Tue Oct 25, 2016 1:59 am

Re: Psi estimate of 1 for specific sites out of total sites?

Postby jhines » Fri Jan 06, 2017 9:36 am

Hi Samundra,

With sparse data, the complexity of models you can run is very limited. In many cases, most of the apparently occupied sites (sites with detections) will be at the high (or low) end of values of a covariate. The model then converges on values of beta's where the intercept is very low (extremely negative) and slope very high. This results in no occupancy (psi=0) for sites with values of the covariate below some threshold value and high occupancy (psi=1) for sites with covariate values above the threshold. Standard errors for estimates near a boundary (zero or one) are undefined, so you can disregard the SE and confidence interval in the output.

The take-home message is that the model is probably too complicated for your data. With only 33 sites, models with more than 2 covariates are going to be difficult to fit reliably. I'd suggest a model with constant detection probability. Do you think that potential habitat affects the probability of detection? I can see why it makes sense for occupancy, but I'm not sure how it could affect detection at a site, given that the species is present. In some cases, a covariate makes sense to affect both occupancy and detection. For example, distance to urban areas might make a species less likely to occupy a site, but might make individuals easier to detect, given they are there. In this case, distance to urban areas affects both occupancy (negatively) and detection (positively).
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA


Return to analysis help

Who is online

Users browsing this forum: No registered users and 5 guests

cron