|
Spatial Analysis and Decision Assistance
Geospatial Analysis
Several tools are provided in SADA for performing a geospatial analysis.
These tools include methods for measuring spatial correlation among data,
modeling spatial correlation, and producing concentration, risk, probability,
variance, and cleanup maps. Among these tools are five geospatial
interpolators:
- ordinary kriging,
- indicator kriging,
- inverse distance,
- natural neighbor,
- and nearest neighbor.
Ordinary kriging and indicator kriging are based on Stanfords open source GSLIB. For information on GSLIB and its open source license please click here. The following sequence of images shows how a data set might progress through
geospatial analysis to a final decision map.
First, an area of concern is selected using the polygon tool. Here, we take a defined area
north of the road.
A variogram summarises the relationship between differences in pairs of measurements and the
distance of the corresponding points from each other by measuring the variance. Typically,
the variance (y-axis) increases with distance apart of measurements (x-axis) due to
expected spatial correlation that occurs between points that are closer together.
Terms commonly used in correlation modeling include:
- Range - the distance between locations beyond which observations appear independent (the variance
no longer increases.)
- Sill- where the variogram levels off, the limiting value of variance along the y-axis.
- Nugget effect- the y-intercept of the variogram, represents the random variation in
data points or may be associated with sampling error.
- Lag distance- is a distance within which any two samples (a pair) is taken for variogram calculation.
Different variogram model types are available, including:
- Spherical
- Exponential
- Gaussian
The variography calculations in SADA are based on Stanford's open source GSLIB. For information on GSLIB and its open source license please click here.
Application of ordinary kriging yields a map of estimated values for the site.
These concentrations can also be converted to risk levels, in this case, residential risk from incidental
soil ingestion.
Kriging also provides an estimate of kriging variance at each location. While this type of variance is not
a measure of local estimation accuracy, it is dependent on the covariance model and the data configuration
(while being independent of the data values). These full distibution of the kriging estimate and
variance can be used to determine the probability that interpolated values are below/above a critical threshold.
Similarly, by specifying a confidence level, blocks in the grid that are above the specified confidence level can
be determined as being within an area of concern.
|
