Occupancy vs Detection Probability

[Jonathan Aronson]

Hey there

I have run an analysis with unmarked in R. I will post my results soon but first I wanted some help understanding some theory. I have a simple question - what is the difference between the probability of detection and the probability of occupancy?

The way I understand it is that the probability of detection is the probability of detecting/finding/catching/seeing/hearing/sampling your study organism at a particular site. The probability of occupancy is the probability of your study organism actually being at a particular site even though you didn't detect/find/catch/see/hear or sample it. 

I study bats and investigate activity at sites using acoustic sampling. Often, species are not recorded on site because of limitations of the equipment even though they may actually be present on site. Therefore, if my knowledge is correct, the probability of detection would tell me the probability that I may record bats using my acoustic equipment based on covariates and the probability of occupancy would tell me the probability that a bat would be a site based on covariates?

Am I getting the right picture here?

Thanks

Jon

[Dan Linden]

Yes.  You can find this information with a Google search.  The following a nice introduction (with key references):

http://fresc.usgs.gov/products/fs/fs2005-3096.pdf

[Gavin Jones]

You can look at detection probability as a "nuisance" parameter. Ultimately, the parameter of interest is occupancy, but in order to estimate occupancy with accuracy, we have to take into account how effectively the species of interest can be detected. This is why detection probability should be approached first--it actually impacts your estimates of occupancy.

For example, a site surveyed for three seasons with three surveys in each season might appear:

101  000  001

If we could detect the species with 100% accuracy, we'd say that the site went extinct from year 1 to year 2. However, we know from seasons 1 and 3 that the species was present at the site, but not always detected (indicated by zeros). Clearly, detection probability is not 100%. Therefore it is possible that during year 2, the species was actually present (occupancy), but undetected (detection). This probability is directly incorporated into the calculation of occupancy probability. 

Therefore, occupancy probability is an informed estimate of occupancy after taking into account detectability. Detection probability is a nuisance parameter that is determined from the 1's and 0's of your survey history.

This is a pretty basic concept--it might be helpful for you to read some of the Mackenzie papers from the early 2000s, and the text Occupancy Estimation and Modeling.

Mackenzie et al. 2002

Mackenzie et al. 2003