Geoscience Reference
In-Depth Information
- the location of the meteorological stations;
- the density of the stations from where the different climate variables are
recorded (the interpolation of rainfall amount or temperature, which are taken from
a large number of measurement stations, is much easier than the interpolation of
duration of sunshine);
- the geometric precision of the tools that provide information on exogenous
variables (information produced by SPOT or Landsat TM is of a better quality than
the information produced by CLC), and information on the spatial resolution of the
cells that make up the processed images.
Another important point, is the fact that the type of information available largely
shapes the choice of interpolation method to be adopted in order to optimize the
results of the interpolation process. Research that is under way and, as yet,
unpublished shows that the results of interpolation are identical on a global scale,
regardless of the method of interpolation that is used (regression, kriging). This is
only true when the information that is available is complete (e.g. information
relating to rainfall and temperature). The same cannot be said for the interpolation
of duration of sunshine, or for any other variable for which the information is
provided by a small number of climatological stations. In this case, the regression
method is not capable of solving the problem, what kriging effectively does.
The environmental characteristics of the area to be studied (such as topography
or land cover) and the spatial information that models these characteristics also
largely influence the choice of the interpolation method to be used. The unpublished
research mentioned in the previous paragraph has revealed that temperature and
precipitation in the West of France (Brittany, Normandy, Vendée), which have a
regular gradient and thus quite a high autocorrelation (precipitation and temperature
scored 0.6 and 0.58, respectively, on Moran's
1
index), were accurately predicted by
the kriging method. Conversely, the region of the Provence-Alpes-Côte-d'Azur
(PACA; in the Southeast of France) had quite low autocorrelation levels
(precipitation and temperature scored 0.35 and 0.3, respectively, on Moran's
1
index). This low autocorrelation was caused by the mountainous topography which
favor the emergence of relatively independent micro-climates. An interpolation
method made up of two analysis phases provides the best results (measured by the
standard deviation of the residual autocorrelations) [CAR 03; FUR 95; JOL 03;
LHO 05]:
- the first phase involves estimating the climate variable by using the multiple
regression method (the independent variables are altitude, slope, degree of
topographic roughness, etc);
- the second phase involves estimating the residual autocorrelations produced
during the first phase. The kriging method is not necessarily the best interpolation
method used, contrary to what many authors believe. It would be the case if the
level of autocorrelation of the residuals was high. However, when the
autocorrelation of the residuals is weak the cubic polynomial provides the best
results.
