Geoscience Reference
In-Depth Information
discussed in this chapter is an improvement in sample efficiency. Adaptive
sampling allows survey effort to be targeted to where any plant or animal of
interest has been found, and this targeting leads to improvements in efficiency.
A number of different designs are introduced in this chapter, and the focus
has been on how adaptive selection can be added to existing well-known
designs. Adaptive cluster sampling is a technique leading from simple ran-
dom sampling, stratified, systematic, or two-stage sampling, where additional
survey effort is allocated to the immediate neighborhood of the sample units
within which an individual is found (or where some measure from the sample
unit exceeds a threshold value). The surveyor prespecifies the pattern of the
neighborhood that is searched and the threshold value that triggers adaptive
allocation. These survey protocol features have an important influence on the
final sample size and sample efficiency. Various modifications to Thompson's
(1990) original design have been suggested to constrain the final sample size.
Other adaptive designs discussed include designs for allocating effort in
stratified and two-stage sampling. In two-phase stratified sampling, addi-
tional survey effort is allocated to certain strata on the basis of first-phase
within-strata estimates. In adaptive two-stage sequential sampling, addi-
tional survey effort is allocated to primary units based on whether the pri-
mary unit estimate exceeds a threshold value.
Complete allocation stratified sampling uses the features of stratified sam-
pling, where the population is categorized into strata. In the first phase, if
any individuals are found within a stratum, the entire stratum is searched.
Matching the size and shape of the strata as closely as possible to the size
and shape of the likely clusters in the species of interest will ensure that the
final survey effort is well targeted to searching where the species occurs.
Even without this perfect match between the strata and species clusters, the
survey method is still an efficient design.
There are many other adaptive designs than those discussed here. Other
adaptive sampling designs can be found in the literature, often initiated by a
very practical and real sampling problem. For example, Samalens et al. (2007)
developed a sample plan for detecting beetle bark infestation. Additional sur-
vey effort was deployed along the edge of forest stands that were near piles
of logs where the beetles were breeding. Yang et al
.
(2011) designed a forest
survey based on adaptive cluster sampling for which, instead of searching the
neighborhood when a sample unit (a forest plot) met the predefined condi-
tion, the size of the plot was increased. Adaptive line transect sampling was
suggested for line transects to increase effort by following a zigzag pattern in
sections of the line where a threshold abundance had been met (Pollard et al
.
,
2002). They discussed surveys of harbor porpoises and how having a zigzag
survey path that did not cross itself, that had no gaps, and was easily followed
was an important consideration for shipboard surveys.
As interest in adaptive sampling continues to grow, the number and com-
plexity of different designs will grow. To assist this expanding field of designs,
Salehi and Brown (2010) suggested the use of the terms
adaptive searching
and
