CMR and Occupancy data

[darryl]

I was using the terms in a conceptual sense rather than something formal. Any devices that are inside an individuals individual's 'home range' will have a non-zero chance of detection (to a practical level of significant digits). Any device that has non-zero probability of detection for 2 or more individuals could be implied to be within an area of home range overlap for those individuals. In your formulation of SECR you estimate home range centres then a spatial scale parameter on the detection function which is likely related to home range size. So like I said, you may not be directly estimating home range-stuff directly, but it was my folksy explanation for why using SECR in this context may be advantageous over 'normal' closed population methods. Apologies for any confusion. I agree that the very notion of a 'home range' is somewhat fuzzy and ephemeral.

[murray.efford]

Maybe it sounds like I'm being pedantic, but it helps to get the concepts straight. The notion of bounded home ranges held people back for years. It's important conceptually that the composite methods do not attempt to estimate occupancy (overlap of ranges with sampling points) - they just make use of a model that predicts whether any animal will be observed at a point in a given set of samples. It is quite OK for all animals to have non-zero probability of detection in all detectors. And while I'm being pedantic - the SECR methods do not routinely estimate home range centres - they integrate over all possible locations for each home range. Estimating the centre is an optional extra.
OK. I'll shut up now.
Murray

[sixtystrat]

So Darryl, does this mean you are "fer it" or "agin it"?
Seriously, I think I get the gist of this. It sounds like the major issue is to come up with a sampling scheme that will work. My original plan was to estimate the bear populations in areas where they were relatively abundant and use the bait stations in peripheral areas where densities were lower but where range expansion and growth could occur. In Louisiana, we have already estimated density using SECR on 2 study areas (0.15 bears/km2 on one and 0.66 bears/km2 on the other), which probably represents the range for areas with established bear populations (of course the real range would go all the way to 0). In Florida, I can design a sampling scheme pretty much however I want. What kind of sampling design were you envisioning? I guess I did not understand Murray's comment about GRTS (?) subsampling. Thanks!
Joe

[murray.efford]

I think Darryl's original point about sampling, with which I agree, was that you could screw up if you estimated detection (functions) in a nonrepresentative subset of the region and extrapolated that to other sites. The way around this is to use a formal probability-based sampling design that ensures representativeness for both the intensive and extensive phases. A systematic grid with random origin is effective in this way, but has some drawbacks. 'Generalised random tesselation stratified' sampling (!) is a compromise that lies between a simple random sample and a systematic sample - it's popular in some US agencies. I suspect it doesn't really add much, but it's easy to implement (e.g. library spsurvey in R) and looks good, so why not? Exactly what is appropriate for the bears would need more thought... The original Conroy et al. idea was a sort of adaptive sampling, which is a little different and requires some fairly strong model assumptions (it seems unlikely to work, even in SECR form, if there is strongly density-dependent variation in detection).
Murray

[darryl]

In principle I think what you're suggesting is doable (especially with SECR) but the devil is going to be in the detail of the sampling in terms of how well it will work out, particularly if the number of primary areas where you doing more intensive work is going to be small, spacing between devices, etc. Getting into those details, however, is probably a bit beyond the scope of the forum and realizing that the free advice provided here is worth every cent you paid for it.

[sixtystrat]

Understood. Free advice appreciated nonetheless!
Joe