help with N

[sixtystrat]

I am estimating N for a isolated but growing bear population in Tennessee using DNA from hair samples. The estimate I get from MARK is 254 bears but when I estimate N with secr, the number is unreasonably high. The population is isolated so I think I am estimating too many bears in the peripheral areas of the mask when the population is really concentrated in the center of the grid. Of course, I can manipulate the mask to produce whatever N I want but that is obviously a poor way to go. I have run a linear density surface and that model is well supported but the estimates of N are still way too high (373). This is a reintroduced population so it is finite. I want a total N so I can estimate growth rates given the number or animals reintroduced long ago. How can I get a more realistic estimate of N? Thanks!
Joe

[murray.efford]

Joe
There are logically two alternatives : your density estimate is wrong (too high) or you are extrapolating the density surface over an inappropriate area (this error is likely to be compounded if you are extrapolating a trend in density - best to avoid trend models here). Or maybe these are two sides of the one coin: If the population is limited to a circumscribed area and the detectors were placed selectively in that area then clearly you should not extrapolate the density obtained to a wider area (sampling was not representative of that wider area). You could call this a design problem. (Not such a big problem if the habitat really is circumscribed and mapped, but that doesn't seem to be your case). If all bears were exposed to detectors, perhaps the non-spatial estimates are the most reliable. That's not very satisfying, but realistically I don't see how else you can handle it.
Murray

[sixtystrat]

Thanks Murray. So should I have set up detectors in areas beyond where I thought bears would occur and then fit a density surface model?

[murray.efford]

Hi Joe
I was speculating on your actual design and design options, so I'm not keen to give strong advice. If the occupied habitat has a known, hard boundary (Efford and Fewster 2013 Fig1A) then you have a choice of spatial and non-spatial estimators of N, although the non-spatial ones are slightly biased. Adding a fringe of detectors in unoccupied but potentially occupied habitat (Fig 1C) is a safe way to estimate N by non-spatial methods, and spatial methods for N also perform well but with slight positive bias. In neither case would I advocate using a density trend model to estimate N.
Hope this makes sense
Murray

[sixtystrat]

Yes that makes sense. Thanks again.
Joe