High detection probability - need repeat visits?

[shar3840]

I am developing a long-term monitoring protocol for pika to be used in several national parks. The objectives are to monitor the status and trends in site occupancy, possibly with 1-2 habitat covariates. Combining visuals, calls, and fresh sign, this species has been shown to be highly detectable (usually p > 0.90 but may be as low as 0.80). Given that we need to keep this protocol simple and easy for the various parks to implement over just a week or two, should we still conduct repeat surveys to estimate detection probability? I know some might say we should always account for imperfect detection but essentially we could survey 100 sites once or 50 sites twice. It’s obviously a tradeoff but I am not sure if the correction is worth the reduced sample size.

Any thoughts, comments, or suggestions would be greatly appreciated.

-Mackenzie

[bacollier]

I am developing a long-term monitoring protocol for pika to be used in several national parks. The objectives are to monitor the status and trends in site occupancy, possibly with 1-2 habitat covariates. Combining visuals, calls, and fresh sign, this species has been shown to be highly detectable (usually p > 0.90 but may be as low as 0.80). Given that we need to keep this protocol simple and easy for the various parks to implement over just a week or two, should we still conduct repeat surveys to estimate detection probability? I know some might say we should always account for imperfect detection but essentially we could survey 100 sites once or 50 sites twice. It’s obviously a tradeoff but I am not sure if the correction is worth the reduced sample size.

Any thoughts, comments, or suggestions would be greatly appreciated.

-Mackenzie

Mackenzie,
You should probably still estimate detection probabilities, but perhaps you could move to a removal design rather than a standard design so that you would only go to a site until you have a positive detection. There are some tradeoff's to using a removal design, but it might work for you. Have a look at the paper below, it might answer a few of your questions.

MacKenzie, D.I., and J.A. Royle. 2005. Designing efficient occupancy studies: general advice and tips on allocation of survey effort. Journal of Applied Ecology 42: 1105-1114.

Bret

[Doherty]

I wonder if you have some flexibility in your field design to obtain multiple observations?

You could visit sites multiple times (I wonder if you think this is your only option?).

Could you have multiple observers visit sites at the same time? Maybe for safety reasons when working in the mountains you already have people working in pairs? This could lead to multiple observations from a single "visit"?

If you have a single person visiting sites would it be possible to think of your estimation problem on shorter time scale? Could a single person make multiple scans over a short period of time (each scan = 1 visit.. although there is probably going to be a dependence issue here to be careful of)... or use a "removal" design - outlined in Darryl et al.'s book.. but again - rather than focusing on different days... focus on shorter periods of time (e.g., 10 minute periods, an hour).

You will know the field situation/logistics better than I do... but I am thinking you probably have some flexibility in the field design that might help you better meet underlying assumptions of the estimation methods.

Paul

[shar3840]

Those are great suggestions and I have been strongly considering either the removal method or multiple observers. However, these still require quite a bit more effort than single visits and it is possible that only 1-2 people will be available for surveys in each park. I recognize the value of these methods for many studies, but I need to be able to justify if this is really necessary, given such a high detection probability already. I am considering running some simulations and playing with some pilot data to see what improvements are made to the estimates and SEs.

Thanks again for the ideas.

-Mackenzie

[darryl]

Hi Mackenzie
A lot of simulation and analytic work has already been done and we've found whenever p<1, you're going to get a smaller SE on your estimate of occupancy by having at least some repeat surveys even if that means going to fewer sites, see either the paper Bret suggested or Chapter 6 of MacKenzie et al. 2006. Then there's also the issue of bias, you can't correct for it unless you know what detectability was. This is even more important if you're interested in habitat relationships, but detectability may also vary by habitat type.

Just one note about doing multiple surveys on a single visit. If there's a possibility of a temporary absence (pika not there today when you are, but could be there tomorrow), you won't be able to differentiate between that and a longer term absence (not there all summer). This may or may not be a problem depending upon exactly how you want to interpret 'occupancy'.

Cheers
Darryl