contents next up previous
Next: Single Variable Cdf Estimator Up: Function Objects Previous: Kriging Constraint

Subsections


Covariance set

Cokriging of variable Z1 accounting for variables Z2,..., ZM requires the covariances between all the variables: Ci, j($ \bf h$), i, j = 1,..., M. These covariances are specified through a Covariance Set. Different models can actually be used to determine these covariances. The full cokriging approach (LMC approach) is very demanding because it requires the complete knowledge of all covariance functions Ci, j($ \bf h$). Modeling cross-covariances from data is a difficult task, because all the covariances can not be modeled independently from one another. The Markov models MM1 and MM2 alleviate the difficulties of full cokriging by using only the collocated secondary variables: the cokriging of location u depends on the neighboring values of Z1($ \bf u$ + $ \bf h_j$) and only the collocated variables Zi($ \bf u$), i = 2,..., M (instead of many neighborhoods of variables: Zi($ \bf u$ + $ \bf h^i_j$)). Hence the covariances Ci, j($ \bf h$), i, j = 2,..., M need only be modeled for distance $ \bf h$ = 0. Moreover, the covariances C1, j, j = 2,..., M are (approximately) proportional to C1, 1 or Cj, j, depending on the Markov approximation used (MM1 or MM2).



Associated Types



Refinement of

Assignable



Notations

a   an object of a type that models Covariance set
U   a type that models Location
u1,u2   objects of type U
i, j   objects of type convertible to unsigned int



Valid Expressions



Models


contents next up previous
Next: Single Variable Cdf Estimator Up: Function Objects Previous: Kriging Constraint
nicolas
2002-05-07