Page 247 - Modern Spatiotemporal Geostatistics
P. 247

228      Modern Spatiotemporal Geostatistics —   Chapter  11

        EXAMPLE  11.11:  A  useful class of nonhomogeneous/nonstationary covariances
        is determined  by the  decomposition relationship  (Christakos,  1992)



        where p^/^s,  t)  and p^/p.(s',  t')  are suitable  polynomials  in  space  and time.
        Furthermore, models of homogeneous/stationary  covariances c y  can be derived
        from


        where UQ  is a linear space/time differential operator.  An  interesting generalized
        spatiotemporal  covariance derived from  Equation  11.50 is as follows










        where the coefficients a 0, a^, b p, c^, and dp/^  must satisfy certain relationships
        derived  from  the  permissibility  conditions.  The  first  three  terms  in  Equation
        11.51 represent space/time  nuggets; the fourth term  is purely polynomial.  The
        last  term  which  is logarithmic  in the  space lag is obtained  only  in 2-D.

        EXAMPLE  11.12:  Yet another interesting  set of nonseparable covariance models
        in R n  x  T  can  be defined  from  separable  covariances.  In general, a sum of
        separable  covariances  is a  nonseparable  covariance;  a  model  belonging to  this
        class  is  (see also  Eq.  10.27,  p. 204)




        where  6j,  c it  and &  are suitable  coefficients.  Note  that  a  superposition  of
        separable  terms  enables  one  to  take  into  account  correlations  that  are  not
        captured  by a single separable term.


        And    Still  the  Garden    Grows!

        We conclude this chapter by expressing the view that the great charm of  BME
        analysis  lies  in  its  almost  unlimited  versatility  and  generality.  Any  possible
        combination  of  scalar  or vectorial  natural  processes,  single-point  or multipoint
        maps,  Euclidean or  non-Euclidean spaces,  homogeneous or nonhomogeneous
        spatial  patterns, stationary or  nonstationary temporal trends,  linear or  nonlin-
        ear  predictors, etc. arising in  practical problems can be examined starting  from
        essentially the  same few  basic  BME equations.  In addition,  most of the  existing
        classical  techniques fit  naturally  into  the  BME  framework, within  which  they
        acquire additional  strength  and significance.  And still the  garden grows!
   242   243   244   245   246   247   248   249   250   251   252