Page 194 - Modern Spatiotemporal Geostatistics
P. 194

Modifications  of  BME  Analysis              175

        Multipoint      BME     Analysis

        In  Chapters  6-8,  we  mainly  considered single-point  mapping,  i.e.,  we  were
        seeking an estimate  Xk of Xk at a single point p k, given the  physical knowledge
        bases  (3  (general)  and  5  (specificatory).  The  multipoint  mapping approach,
        on  the  other  hand,  is seeking  the  estimates  Xfej  °f  Xfc 3  a *  several  points p k.
                 •
        (.7  =  1,  • • , p) simultaneously, given the  physical knowledge available (Fig.  5.4,
        p.  121).  When practically  possible, multipoint mapping could be a considerable
        improvement  over  single-point  mapping.  Indeed,  while  multipoint  mapping
        offers  several interrelated  estimates at  a time,  single-point  mapping  gives one
        estimate  at  a  time,  independently  of  its  neighboring  estimates.  The  former
        involves  a  multivariate  pdf  that  is  more  informative  than  the  univariate  pdf
        used  in the  latter.

        Multipoint    BME estimation

        In  principle,  the  BME  approach can  provide multipoint  estimates  by  replacing
        the  single  Xk with the  vector




        and  the  vector                    with the  vector




            Then,  the  basic  BME  equations  of  multipoint  space/time  mapping  are
        derived as follows.  The  Lagrange multipliers  /j. a  are calculated from the  solution
        of the following system of  (p-point) equations





        The  multipoint  (p-point)  posterior  pdf  is given by




        where                 and,  as usual, the  forms of  A  and y s depend on  the
        specificatory  knowledge  considered  (see also  Example 9.7).  We  now  have  a
        system of  p  BMEmode  equations






                                                                 •
        Solution of the system (Eq.  9.29)  provides the estimates x fe  =  ( X k t , • • ,  Xk p).
        The number p of points to  be estimated  simultaneously may depend on theoret-
        ical, computational,  and physical, as well as on decision-making considerations.
   189   190   191   192   193   194   195   196   197   198   199