Consider the line , in the three-dimensional space
, as shown on
Figure 7.2. Consider the one-dimensional random function
) defined on line
; this random function is
second-order stationary, with a zero expectation,
and a one-dimensional covariance
Let , be the projection of any point
onto the
line
, and consider
the three-dimensional random function defined by
. This random function
is second-order stationary, with a zero expectation
and a three-dimensional covariance equal to
In practice, to produce a realization of
, the
value
, simulated at
the point
, on the line
, is given to all points
inside the slice (or the
band) centered on the plane
perpendicular
at
to
, as
illustrated in Figure 7.2. The thickness of this slice is the
equidistance between the simulated values on the line
.
We then consider lines
corresponding to the
directions of the unit vectors
uniformly distributed over
the unit sphere. On each line
a realization
of a
random function
is generated
isomorphic to
, the
random functions
being independent.
Using the method outlined above, a three-dimensional realization
, is
made to correspond to each one-dimensional realization
.
A final value is then assigned to each point by taking the sum of the
contributions from the
lines:
The resulting realization
is a realization of a
three-dimensional random function
which is of course, of second-order stationary and has a zero
expectation.
The covariances are treated in practice in the way that the
three-dimensional covariance
is fixed and the one-dimensional
covariance
to be simulated on each of the
lines is thus
given by the derivative
It is thus always possible to determine the one-dimensional covariance
to be imposed on the random function
simulated on the lines
. By turning these lines and their
corresponding bands in
, the required three-dimensional simulation
, with the specified isotropic covariance
is obtained.
We remark that the number of lines chosen in practice is often 15.