Estimation of a non stationary covariance modelΒΆ
Let be a multivariate normal process of dimension where . is supposed to be a second order process and we note its covariance function. We denote the vertices of the common mesh and the associated values of the field . We suppose that we have fields. We recall that the covariance function writes:
(1)ΒΆ
where the mean function is defined by:
(2)ΒΆ
First, we estimate the covariance function on the vertices of the mesh . At each vertex , we use the empirical mean estimator applied to the fields to estimate:
at the vertex :
(3)ΒΆ
at the vertices :
(4)ΒΆ
Then, we build a covariance function defined on which is a piecewise constant function defined on by:
where is such that is the vertex of the nearest to and the nearest to .