A random function is said to be stationary, in the strict sense, if
its spatial law is invariant under translation. More precisely, the two
-component vectorial random variables
and
are identical in law (have the same
-variable distribution law)
whatever the translation vector
.
However, in linear geostatistics, as we are only interested in the previously defined two first-order moments, it will be enough to assume first that these moments exist, and then to limit the stationarity assumptions to them.