Geoscience Reference
In-Depth Information
that exists between the duration of sunshine (the dependent variable), and slope (the
independent variable) does not make much sense in terms of space. This is due to
the fact that the correlation between these variables is calculated from such a small-
range data series (0 to 5°).
It is difficult, if not impossible, to estimate the duration of sunshine by using
either linear or multiple regressions. In order to carry out such an estimation it is
best to use the kriging approach as this approach provides the best results. This is
due to the bad spatial distribution of the climatological stations that record the total
duration of sunshine. These stations are generally synoptic and are found in
uniform, homogenous sites that increase the level of autocorrelation. Let us notice
that a kriging with an external drift does not provide any better results than the
normal kriging method. Here the problem is more informational than related to the
interpolation method used.
Several factors need to be determined before any process of interpolation can
take place, for example the number of climatological stations where the climate
variables are to be observed and recorded, the location of the climatological stations
and the density of the climatological stations in a given area. The type of
interpolation method which can be used depends partly on these factors.
2.4.2.
Information source and quality of interpolation
In the field of spatial statistics the most difficult problem to overcome is
collecting the independent variables. In the first part of the chapter we saw that there
are two major types of geographic information that are readily available and they
include DEMs and information resulting from remote sensing. The issue raised in
this section refers to the effect of the information source on the quality of
estimations. To help answer this question, data from two distinctly different sources
of information will be compared, i.e. CLC and SPOT. Interpolation works well if
the independent variables introduced into the regression models are of a quantitative
nature. Variables, such as altitude, slope, and roughness of terrain, do not pose any
problem. However, variables that provide information on land cover are more
difficult to deal with. Measuring the distance that these variables are from an object
is a useful and practical method to use when it comes to measuring the thermal and
pluviometric influence that a particular object might have on its surrounding area. In
some cases it is necessary to rely on the fractal dimension of certain objects, such as
wooded areas or developed areas (areas with lots of sites on them), in order to
estimate the influence that the spatial distributions of these formations will have on
climate variables [FUR 94; JOL 03].
Another variable that is often used is the NDVI, which is easy to integrate within
the different interpolation models. It should be pointed out that the NDVI is a simple
numerical indicator that not only integrates data, such as type of vegetation, but also
the state in which the vegetation can be found, such as lack of water, disease,
infestation of insects, etc. The index reflects the importance that vegetation has in
