Appropriate
selection of an interpolation method to apply is a key aspect of the current
work. However, the issue is two-fold insofar as there is an ideal situation
and a practical situation. Ideally, a comparison between several methods would
be conducted, including both univariate and bivariate geostatistical methods
and simpler univariate methods. From the corresponding literature, only the
methods most commonly applied and those methods which performed best in the
context of the particular study can be determined. However, it is impossible
to transfer the conclusions from these studies to the current work and conclude
that these methods would also perform best in the current study. Every study
area is a unique product of a host of variables. To determine which interpolation
method is best suited for application to precipitation data in the GVRD, several
potential methods would actually have to be applied and then tested for differences
in their relative performance by means of a validation procedure. However, for
application to the current work, the models chosen have been restricted by available
software, time and especially the availability of domain expertise within the
project.
Three models were tested in the current work – IDW, Ordinary Cokriging
and Linear Regression. These particular models were chosen because they are
common models and they provide a useful comparison amongst a diversity of potential
methods. IDW only accounts for the distance between stations, that is, the spatial
structure. Cokriging accounts for both the spatial structure of the data as
well as including information derived from a second attribute (elevation). The
linear regression model accounts for the effect of elevation but contains no
explicit reference to the spatial structure of the station data.
