Geoscience Reference
In-Depth Information
h e range of the observations
z
can be checked by
min(z)
ans =
3.7199
max(z)
ans =
7.8460
For linear geostatistics, the observations
z
should be Gaussian distributed.
h is is usually only tested by visual inspection of the histogram because
statistical tests are ot en too sensitive if the number of samples exceeds about
100. One can also calculate the skewness and kurtosis of the data.
histogram(z)
skewness(z)
ans =
0.2568
kurtosis(z)
ans =
2.5220
A l at-topped or multiple-peaked distribution suggests that there is more
than one population present in the data set. If these populations can be
related to particular areas they should be treated separately. Another reason
for multiple peaks can be preferential sampling of areas with high and/or low
values. h is usually happens as a result of some
a priori
knowledge and is
known as a cluster ef ect. Dealing with a cluster ef ect is described in Deutsch
and Journel (1998) and in Isaaks and Srivastava (1998).
Most problems arise from positive skewness, i.e., if the distribution
has a long tail to the right. According to Webster and Oliver (2001), one
should consider root transformation if the skewness is between 0.5 and
1, and logarithmic transformation if the skewness exceeds 1. A general
transformation formula is:
