Occupancy of discrete habitat patches

When we are dealing with well-defined, discrete habitat patches, such as ponds which may or may not be occupied by frogs during the breeding season, the idea of "occupancy" is straightforward. (See the page on 'Frogs in ponds'.)

To estimate occupancy, we need to estimate detection probability, the probability that the species will be detected if it is present. We can then estimate the proportion of ponds occupied by frogs, allowing for non-detection and false absences.

We estimate detection probability with multiple survey occasions at each site. But this depends on the closure assumption, that the species is present (or absent) throughout the survey. (Individual animals do not need to remain at the site, but some members of the species must be present on each occasion.)

In general, the closure assumption is met if the 'site' is much bigger than the home range of the animals, so that most animals are moving around within the site, but not entering or leaving.

Use of discrete habitat patches

Suppose we were interested in herons instead of frogs. We use a similar protocol, but this time we look for herons. Herons move around from pond to pond, and at many ponds will be sometimes present and sometimes genuinely absent. The closure assumption doesn't apply. What are we measuring now?

The appropriate concept here is use (or 'utilization' if you prefer longer words). What proportion of ponds are used by herons? In principle, we now have two probabilities:

1. Pr(present at site on survey occasion | site is used)
    
2. Pr(detected | present at site on survey occasion)

Multiple-observer protocols allow us to estimate the second of these, provided the observers visit the site at the same time. But that may not be what we want.

Most protocols, with survey occasions at different times, do not allow the two to be distinguished, and we estimate the combined probability, which is the product of the two above:

Pr(detected | site is used)

We can then use this to estimate the proportion of ponds used by herons, allowing for the fact that herons move between ponds and non-detection even when they are there.

If herons spend little time at each pond, probability 1 above will be low, and so will the overall probability of detection. Then you will need lots of observations at each pond to get a reasonable estimate of use. A better scheme would be to look for heron tracks - provided these can be reliably distinguished from other animals. In that case, you should plan your surveys so that you are not looking at the same tracks each time.

Notice that, in this case, the sites generally lie within the home range of the animals, not the other way round.

Occupancy/use of large tracts of habitat

How do these ideas apply when we cannot distinguish small patches of habitat, but we suspect that the species of interest does not use the whole of a large tract of habitat?

Let's look first at a small-scale problem:

Now imagine that the diagram is a map of a forest and the grey blobs are the home ranges of a our target.  We want to estimate the percentage of the habitat that falls within a home range. As with the blobs of lichen, we can look at sample locations and see what percentage fall within the home range of an animal.

Unlike lichen, however, we can’t be sure that the species is absent from a sample location when we fail to detect them: we need to estimate the detection probability and adjust the occupancy estimate accordingly.

Occupancy/use of individual habitat patches

Sometimes we are interested in specific habitat patches rather than an overall estimate of occupancy or use. For example, we may know that gibbons were present in a particular valley before logging, and we have not detected then since logging was carried out.

The appropriate question now is, "What is the probability that this is a false absence?"

To estimate this, we need an estimate of detection probability. We can't calculate this for the patch in question, if we have not detected the species. Instead we would have to use data collected using the same protocol in similar habitat elsewhere. Once we have an estimate of detection probability, , we can calculate the probability that we could go to a site where the species was present n times and not detect it as:

Pr(false absence) = 1 - (1 - )ⁿ

If the probability of false absence was tiny - you might set the bar at p = 0.05 (5%) - you would be reasonably confident that the species was in fact absent from the habitat patch.

Main points

• The interpretation of 'proportion occupied' is straightforward for discrete patches of habitat where closure applies, ie. the species is consistently present or absent on all survey occasions.
   
• Closure implies that home ranges are smaller than habitat patches ('sites').
   
• If animals move between habitat patches (ie. home range is bigger than the patch), closure does not apply. In this case, it may be possible to estimate patch use.
   
• Survey protocol must be adapted to estimate the probability of detection given that a patch is used, rather than an animal is actually present.
   
• For animals with well-defined home ranges, we can estimate the proportion of a large tract of habitat used by sampling a large number of randomly-placed sites.
   
• For a specific site where a species has not been detected, we can estimate the probability of a 'false absence' provided we can obtain an estimate of detection probability.