Using call-up surveys in occupancy models

[neilmidlane]

Hi

I have been using call-up surveys to estimate lion density in a national park. I would now also like to apply the results of this survey to an occupancy model. The survey entails playing a buffalo distress call to attract lions to a point where they can be observed. These surveys can be calibrated to determine likely response distance, as well as probability of response. This then gives a response probability in a given circular area. Could this then be deemed to be detection probability? For example, if the circular area is 50km2, where probability of response is 50%, and occupancy model grid size is 200km2, would it be valid to calculate p(detection) as 50% x 25% = 12.5%?

If so, could this p(detection) be input into PRESENCE upfront so that the software doesn't use the detection history to determine p(detection)?

Thanks
Neil

[murray.efford]

Neil
I think you need to back off from this idea. The detection probability from a lured point survey (I assume you're following the Buckland et al. method) applies to each individual, whereas occupancy detection probability refers to the probability of detecting at least one individual if the species is present (although Royle-Nichols models provide an interface between the two). More fundamentally, the probability does not scale by area as you suggest. It is possible in principle to fix parameters in these models (I'm not sure about Presence) but it becomes hard to compute the correct SE. It also seems to me that you haven't defined a meaningful occupancy parameter (see paper on this due out in Ecosphere in a couple of weeks).
Murray

[neilmidlane]

Hi Murray

Thanks for the response, maybe I didn't explain myself clearly enough. An example of the calibration of this methodology can be found in Ferreira, S.M. & Funston, P.J., 2010. Estimating lion population variables: prey and disease effects in Kruger National Park, South Africa. Wildlife Research, 37(3), p.194-206. The method involves using 2 vehicles, one with the lions and a second one a known distance away. The second vehicle plays a recording to attract the lions, moving closer until a response by the lions is noted by the first vehicle. The distance of the response is measured, and the number of lions arriving at the second vehicle, as a percentage of the group size, is also recorded.

Here is a brief excerpt from the paper explaining the results of the calibration:

"We could only estimate a response for lion groups without cubs – the response decayed with distance. The fitted inverse sigmoid model predicted that lions responded up to 4.3 ± 0.9 km (mean ± SD; 95% CI: 2.5-6.1 km) away from call-up stations (Fig. 2a). This means that call-up stations sampled 57.7 ± 24.9 km2 (mean ± SD; 95% CI: 8.9-106.6 km2). We assumed that the area sampled by a call-up was the same for groups with cubs than that for groups without cubs. Approximately 73% of the 28 lion groups without cubs within the sampling area responded to a call-up (0.734 ± 0.076, mean ± SE)."

Given this, I can't understand why it wouldn't be appropriate to state that "there is a 73% (0.734 ± 0.076, mean ± SE) probability of detecting a lion within an area of 57.7 ± 24.9 km2 (mean ± SD; 95% CI: 8.9-106.6 km2) surrounding a call-up station. Following that, the p(detection) in the cell would be a function of the circle's area as a proportion of the cell's area.

I'm not sure if the above gives any more clarity and perhaps changes your opinion at all? Your advice/comments would be welcomed.

Thanks
Neil

[murray.efford]

Drat - timed out again. I don't have F&F. Suggest you look at Buckland et al. 2006 J Appl Ecol 43:377-384, which is probably a more sophisticated treatment of a similar problem (note their calibration uses one observation per animal, not a whole sequence). I think your calculation requires restrictive assumptions (single station near centre of each grid cell; maximum detection radius less than grid cell half-side) and ignores the difference between group-level detection and detection of occupancy. My heads hurts when you cite 73% detection within an uncertain area!
Murray