Geoscience Reference
In-Depth Information
about because there are few individuals in the population or because the
population covers a large area (McDonald, 2004).
Adaptive sampling designs involve the survey team changing or adapting
the procedure used to select sample units as information comes to hand dur-
ing the course of sampling. The key point here is that the protocol of what
to sample and where may change as the survey evolves, and the field crew
needs to be able to adapt to this change. The statistical properties of all the
designs discussed are well documented, and they all fall within the realm of
probability-based sampling.
There are two categories of adaptive sampling: adaptive searching and
adaptive allocation (Salehi and Brown, 2010). Adaptive searching includes
designs for which the neighborhood of certain sample units is of special
interest (e.g., a rare plant is found in the sample unit and the search shifts
to focus on the surrounding units); adaptive allocation involves allocation of
additional survey effort to the general area of interest, such as a stratum that
is thought to contain rare plants.
Interest in adaptive sampling peaked with the development of adaptive
cluster sampling (Thompson, 1990), although adaptive designs were sug-
gested well before this (e.g., Francis, 1984). I begin by discussing adaptive
cluster sampling and then move into some of the other adaptive allocation
designs to illustrate the range and versatility of adaptive sampling.
3.2 Adaptive Cluster Sampling
Adaptive cluster sampling was first introduced by Thompson (1990). The
design was developed for sampling populations that are rare and clustered.
It is similar to cluster sampling, in which a cluster of units is selected, and
either the entire cluster or a portion of it is sampled. The typical example
used to explain cluster sampling is surveying children in a school. Classes
are natural clusters of children, so a selection of classes is chosen, and chil-
dren within classes are surveyed. In adaptive cluster sampling, clusters are
selected, but the difference is that the size, location, and total number of clus-
ters are not known.
In its simplest form, adaptive cluster sampling starts with a random sam-
ple. Prior to sampling, a threshold value
C
is chosen, and if any of the units
in the initial sample meet or exceed this threshold,
y
i
≥
C
, then neighboring
units are sampled. Any unit that meets or exceeds this threshold is consid-
ered to have “met the condition.” If any of these neighboring units meets this
condition, their neighboring units are selected and so on. In this way, as sam-
pling continues for any cluster that is detected in the initial sample, the edge
of the cluster is delineated, and the cluster itself will be sampled. The final
sample is the collection of clusters that were detected in the initial sample,
