Environmental Engineering Reference
In-Depth Information
philosophy and techniques for this method, and is essential reading for those
wishing to analyse distance data. Here we summarise the basic underlying con-
cepts that must be understood in order to design and implement an effective dis-
tance sampling survey (Box 2.3).
Precision in distance-based estimation depends on having both a reasonable
number of replicates (lines or points) to minimise sampling error, and a reasonable
number of sightings, to minimise error in the detection model. Although it is
possible to estimate sampling error without replicates by assuming a theoretical
sampling distribution, this approach is not ideal because the assumed distribution
is likely to be wrong, and it provides a very weak basis for inferring anything about
the wider population. As a rule of thumb, at least 10 independent lines or points
are needed to get a reasonably precise estimate of the encounter rate. The number
of sightings,
n
, is approximately related to precision by:
3
CV(
ˆ
)
2
n
CV(
D
ˆ
)
with coefficient of variation, , expressed as a proportion (Buckland
et al
.
2004). Thus to achieve a coefficient of variation of 25%, around 50 sightings will
be needed. Considering the relatively low power to detect differences at this level
of precision (Section 2.2.2), this number of sightings can be regarded as a rule of
thumb minimum. Where it is difficult and time consuming to set up transects, and
encounter rates are low, it might be sensible to make
repeat visits
to transects in
order to boost the number of sightings. However, visits to a single transects should
not be treated as independent replicates for the purposes of calculating
sampling error (see Section 2.2.3 on pseudoreplication). As far as possible, any
additional effort should be channelled into new transects rather than repeat visits
to existing transects.
When
placing transects
, the ideal randomised or systematic approaches
(Section 2.2.3) may be impossible to achieve in practice. It may be that some parts
of the study site are simply too hard to get to, or that travelling anywhere in the site
is too difficult to place samples far from access points or from one another. For
transects, the costs of travel might be minimised by:
●
Running long transects with short sections 'off effort', thus dividing the long
transect into sampling units;
●
Eliminating time off effort entirely by defining sampling units as the arms of
a zigzag travelling pattern.
While neither of these approaches provides strictly independent sampling units,
it is usually reasonable to treat them as if they are, so long as coverage of the entire
study site is reasonably high. Another approach sometimes used in difficult
habitat is simply to take a
path of least resistance
, avoiding dense cover or even
using paths or roads. This approach is almost certain to yield biased abundance
estimates, and should be avoided.
