Individual accumulation curves

[lourens]

Hi All,
I have a question that might not be related to secr specifically, but rather field methods. After around 90 days of camera trapping for leopards accumulation curves for individuals did not reach a asymptote,suggesting that i failed to detect all individuals. Is this a problem for secr? My results have quite large CI, making inferences difficult.
How do i deal with this issue?
Any suggestions would be welcomed.
thanks-Lourens

[murray.efford]

Hi Lourens

The general answer to your question might be 'relax - that's what capture--recapture is for'.

My specific answer for spatially explicit capture-recapture runs like this:

Conceptually, there is an indefinitely large number of leopards out there that might ultimately appear at your cameras, so we do not expect an asymptote and estimates of density should be unbiased regardless. Many times there will be one, either because the curve gets very nearly flat once the cameras have exhausted the 'easy pickings' of the immediate population, or because the biological population really is finite.

The 'exhaustion' effect depends on the shape of the detection function. It is greatest with short-tailed detection (e.g., halfnormal) and least if the function has a long tail (e.g. exponential or hazard rate). This relates to patterns of movement.

It is also possible that turnover in the population will cause an otherwise asymptotic curve to steepen. This is something to be aware of, and may justify cutting a long camera trapping series into 2 or more sections.

Murray

Here is some code to illustrate my claims above (yes, I know the numbers are unrealistic!)

Code: Select all

library(secr)
grid <- make.grid(spacing = 40)   ## default grid is 6 x 6
CH.EX <- sim.capthist(grid, pop = list(D=20, buffer = 1000), detectfn = 'EX',
                   detectpar = list(g0=0.05, sigma=50), noccasions = 100)
## nonasymptotic accumulation curve
plot(1:100, counts(CH.EX)[['M(t+1)']][-101], ylim = c(0,800), xlab = 'Occasion', ylab = 'M(t+1)', type='l')

## select just 20 occasions and fit model
fit <- secr.fit(subset(CH.EX, occasions = 1:20), buffer = 300, detectfn = 'EX')

options(digits=3)
predict(fit)
#    link estimate SE.estimate     lcl     ucl
# D       log  17.5341     1.32924 15.1164 20.3386
# g0    logit   0.0553     0.00376  0.0484  0.0631
# sigma   log  50.8838     2.25264 46.6568 55.4938

[lourens]

Hi Murray,
Thanks so much, I saw you give a similar answer in the general MARK forum some time ago. I agree, and also thought along those lines, but it is nice just to get confirmation.
Lourens