Geoscience Reference
In-Depth Information
These spatialized factors form part of the interpolation functions as external drift
(in the case of kriging) or as independent variables (in the case of statistic
interpolations). The factors are archived as information layers in the GIS. These two
types of interpolation not only rely on the use of the independent variables, but also
on the location of the climatological stations, which is essential.
2.3.2.1.
Kriging with external drift
Kriging is a very reliable technique from a statistical viewpoint, because it is
based on the analysis of covariances. One of the main criticisms that can be applied
to the method of kriging is that the spatial variations of the phenomena that are
analyzed are not only a function of spatial autocorrelation, but a function of other
processes. In many cases, variables, which are weakly correlated with a
phenomenon that is being analyzed, will have a strong influence on the spatial
variation of the phenomenon in question. For example, slope and the topography of
the Earth's surface have a considerable effect on temperature. The kriging method
retains this information with the help of external drift.
In comparison to normal kriging, kriging with external drift improves the
accuracy of estimations that have been made, especially in zones where there are not
enough samples [HUD 94; TAB 05].
2.3.2.2.
Statistical spatial analysis
Theoretically, correlations that are solved by the process of least square
regressions are not appropriate when there is autocorrelation, in other words when
there is spatial dependence between the connected variables. It is still not possible to
avoid autocorrelation. Most attention is needed when such variables are to be
integrated.
It is strongly recommended, if not necessary, to carry out a statistical test on
variables [LHO 05].
This type of determinist approach can be explained in more detail with the help
of Figure 2.4, which is an example of interpolation that uses spatial statistics. This
approach relies on the use of diverse information, which is digitally archived and
which is then combined together [FEY 95; JOL 03]. This approach is then used to
locate information (from the information that is available) which would be suitable
in helping to solve the problems associated with interpolation.
The principle of this approach involves calculating (by multivariate regression)
the links that are established between the endogenous climatological variable that is
to be interpolated, and the independent climatological variables that are stored in the
GIS as information layers. The process of statistical modeling makes it possible to
recreate continuous fields of endogenous variables.
