Occupancy Models - Introduction to Ecological Sampling - page 126

based on the given data, it can be accounted for within the probability state-
ment to remove any bias caused by the imperfect detection. That is, the prob-
ability statement for this detection would be

Pr( h = 000) = ψ− (1)(1 p − p )(1) − p +−ψ (1 ) .

i 1 2 3

The addition of the two terms in the probability statement is how the ambi-
guity in the observation is accounted for; each term represents the possible
explanation for how the species could never be detected at a unit, either a
false or a genuine absence, respectively.
Probability statements can be developed for all s sampling units that are
surveyed, and in combination can either be used within a maximum likeli-
hood or Bayesian framework to obtain results in the usual way. Note also
that equal sampling effort at each unit is not required so that the number of
surveys per unit need not be consistent (see MacKenzie et al., 2002).

EXAMPLE 9.1 Blue-Ridge Two-Lined Salamander

MacKenzie et al. (2006) presented results from data collected on the
blue-ridge two-lined salamander ( Eurycea wildrae ) in the Great Smoky
National Park, Tennessee. Thirty-nine 50-m transects were surveyed
on five occasions between April and mid-June 2001. Each transect con-
sisted of one natural cover transect where objects such as rocks and logs
were lifted to check for salamanders and a parallel artificial cover object
transect where cover boards (pine boards) were placed every 10 m and
searched to reduce the impact on the environment. MacKenzie et al.
(2006) pooled the data from the two types of transect to obtain a single
detection or nondetection for each survey of a transect. Further details of
the original study can be found in the work of Bailey et al . (2004).
Blue-ridge two-lined salamanders were detected at least once at 18 of
the 39 transects. This would correspond to an uncorrected estimate of
the probability of salamanders being present on a transect of 0.46 with a
standard error (SE) of 0.11. Using the methods described to account for
the possibility of false absences, assuming the detection probability was
the same for all five surveys, the estimated probability of salamanders
being present on a transect is 0.59 with an SE of 0.12. The nearly 30%
increase of the uncorrected estimate is a result of detection probability
being estimated to be low at 0.26 (SE = 0.05) per survey; hence, the prob-
ability of not detecting the salamanders on a transect where they were
present after five surveys would be 0.22 [i.e., (1 - 0.26) 5 ]. That is, there
was a 22% chance of not detecting salamanders along a transect after
five surveys even when they were present.

9.3.2 Multiple-Category Model

The two-category model of MacKenzie et al . (2002) was extended by Royle
and Link (2005) and Nichols et al. (2007) for situations when two categories