Spatial Analysis and Decision Assistance
Secondary Sampling Design
SADA provides different strategies to determine future sample locations.
Depending on the geospatial interpolator
chosen, the following strategies are available.
Adaptive Fill
This approach is designed to spatially fill the holes among existing
data with new data points by suggesting locations that are the
farthest from any other data point. This
method is the simplest sample design to use and is independent of the geospatial
interpolator chosen; however, it gives no regard to the magnitude or variability
of data or to the user's decision rule.
Estimate Rank
This approach fills new samples into unsampled locations that are modeled
to have high concentration levels relative to the existing data. This
approach can be useful for verifying the extent of hotspot regions and
is available for any of the interpolation schemes. It gives no weight,
however, to the variability of data. Consequently, data may be placed
in an area that is high in concentration values but is rather well characterized.
Variance Rank
Variance rank fills new samples into unsampled locations that have high
estimation variances. This approach will fill data into locations that
may not be well characterized from a modeling perspective. Since this
approach gives no weight to the magnitude of concentrations, samples may
appear where data are sparse but where corresponding concentrations are
very low relative to the decision rule. This approach is available only
with ordinary kriging.
Percentile Rank
This approach is almost a merger of the two previous approaches. It gives
weight to both magnitude and variability and reduces the tendency to place
data in well-characterized hot spots and in sparse areas with very low
detected or nondetected values.
Uncertainty Rank
This approach is the only one that is connected to the decision rule.
It places new sample locations in areas where there is the greatest uncertainty
about exceeding the cleanup goal (block scale). This approach is useful
for delineating the boundaries of an area of concern.
Secondary Constraint
In each of the last four approaches, the exact sample locations are optimized
in a mathematical sense. For example, in estimate rank, the unsampled
point with the highest concentration level is chosen as the first priority
sample location. This location may, however, be located extremely close
to another data point (in most cases this is the highest sample data point).
While this satisfies the mathematical rule, it may not satisfy the user.
Typically, samples should serve at least two purposes: 1) meet one of
the approaches described above and 2) provide a good spatial spread to
the sampling scheme.
Therefore, SADA allows a secondary constraint approach where the user
specifies a minimum distance required between any new sample locations
and any previously sampled data. New sample locations are then optimized
with respect to both purposes. In the last four figures shown, a constraining minimum
distance of 240 meters was used.
| 
