Brian
bgerber wrote:To me, a site in my case would be the camera detection zone located in a specific microhabitat. What I would like to determine for the largest carnivore is if they are occupying much of the National Park and if micro/macro habitat characteristics influence occupancy.
That's fine if you want to do it that way, although you'd have to interpret your psi estimate as 'use' (ie the species was within this point at some time during the sampling, but not necessarily all the time) so keep that in mind when you start thinking about microhabitat covariates. Plus, unless you're planning on having 100's of camera stations, you'll probably only be able to pull out large-scale effects. Finally, are your cameras going to be placed truly randomly, or only on game trails? This could also influence the biological interpretation of any apparently important covariates.
bgerber wrote:Do you see an issue in which cameras vary from 200m apart to 1500m apart? If I want to make an inference about the occupancy of a carnivore for the entire national park and place cameras randomly in such a way that some cameras are clumped and some are very far from another camera, is there not a bias towards the clumped cameras? Thus, I would bias my occupancy estimate high or low depending on if the clumped site was occupied or not occupied.
Presuming those distances are relatively large compared to the detection range of a camera, then no. As you're defining your site as the detection range of the camera, then if there's some clustering due to random chance then so be it, that arrangement is as equally likely as any other. How do you know that any cluster will either be all occupied, or none? Couldn't the boundary occur somewhere within the cluster?
bgerber wrote:I have pilot data that was collected in this way. I have a total of 43 camera sites, but I keep thinking that I need to analyze a subset of the sites, based on a potential homerange. If I use all 43 sites, I can't seem to adequately reconcile how the spatial autocorrelation doesn't bias my inference for the entire park.
If you have random placement then I don't think it's too much of an issue, here's my logic. No claims it's not flawed. Forgetting about detectability for a moment, if you knew whether or not each site on the landscape was occupied (potentially with spatial correlation), then through them all in a hat and randomly drew out 1 site at a time and noted whether it was occupied, then you should get an unbiased estimate of the fraction of sites that are in the hat that are occupied. If you knew the degree of spatial correlation, then perhaps you could come up with a more efficient sampling scheme than simple random sampling, but that may then be sensitive to the assumed level of correlation. As for detectability, provided that the outcome of a survey at any site does not depend upon whether the species was detected or not at another site, I think you're fine. About the only way I can imagine spatial correlation would be a problem here (off the top of my head) would be for a highly territorial species such if you have 2 sites within an individuals home range, there's no way you could get a detection (of the species remember, not necessarily the same individual) at both sites in the same night. With camera traps though a simple fix would be define your 'survey' as longer period such that there is some chance of getting a detection at both places, ie make it long enough for the animals to move around a bit, eg a 5-day period perhaps.
Sorry for the lengthy reply.
Cheers
Darryl