Page 139 - Modern Spatiotemporal Geostatistics
P. 139

120      Modern  Spatiotemporal  Geostatistics —  Chapter 5
         Specificatory Knowledge Base").  Approaches leading to  different arrangements
         of  the  knowledge  bases incorporated  in the  prior  and meta-prior  stages  can be
         considered,  a fact  that  adds to  the flexibility  of the  BME  analysis.  Computa-
         tional  aspects can  also  play  a  role  in  decisions about  the  physical  knowledge
         to  be  used  at  each stage.  In some cases,  physical knowledge-based  constraints
         may  arise for  which  pdf  maximizing  of  the  expected  information  (Eq.  5.2)  is
         difficult  to  obtain  (see,  e.g.,  Shimony,  1985).  It  is then  possible that  these
         constraints  can  be put  in  a form  that  is easily  incorporated  at  the  meta-prior
         stage.

         The   Integration     or  Posterior   Stage

         In  the  prior  stage,  the  operator  9£  processed  general  knowledge  Q.  The  Q
         knowledge as well as the specificatory knowledge S considered in the  meta-prior
        stage  are incorporated  into  the  mapping  process  by means of  a new operator
         introduced  by the  following  postulate.
         POSTULATE 5.4:   At the integration stage,  the updated pdf of the map
        should  be  expressed  in  terms  of  an  operator  %  which  integrates  the
        general  knowledge-based  operator  9£  with  the  specificatory  knowledge
         S considered  at  the  meta-prior  stage, i.e.,




        where  !?C =  ^7 U 5,  and  A  is a  normalization  parameter.

            At  this  point,  Equation  5.35  presents an abstract  form  of % which  indi-
        cates  its  dependence on  the  general knowledge operator  %,,  the  specificatory
        knowledge  S, and the mapping  vector x map  =  (Xhard. X soft,  X k)\  the opera-
        tor  *y s  is associated with  the  mapping  point p k.  Conceptually,  Postulate  5.4
        is a direct  consequence of  Proposition  4.1  (p.  95),  in which  case the  operator
        y s  and  the  normalization  parameter A  account  for  the  probability  functions
        /<? [Xmap(^)] and MXdata(>S)]> respectively. TheXfclXdataC^) stands for the
        possible values Xk  of the map in the context  specified by Xdata(^)  anc '  Q-  Prob-
        ability  models such  as Equation  5.35 should  always  be understood  to  apply  in
        the  appropriate context, which defines the current state of  physical  knowledge.
        Postulate  5.4  suggests  a  natural  synthesis  of  the  general  knowledge  of  the
        prior  stage  and  the  specificatory  knowledge  of  the  meta-prior  stage.  In  the
        multipoint  case  (see Chapter 4,  p.  89),  Postulate  5.4  is  concerned with  the
        multivariate pdf f^(Xk)- Xk = (Xfcu- • • )XfcP)- m the single-point case, on the
        other  hand, we are dealing with  several  univariate pdf's f^(xki),  i  =  1, • • • ,p
        (see  Fig.  5.4).
            Certain ^-operators  have been introduced  earlier  in this chapter  (p.  106).
        In  order,  then,  to  obtain  an explicit  analytical  form  for  y s,  the  form  of  the
        specificatory  knowledge S  must first  be described explicitly  (in  addition  to  the
        hard facts, a geostatistician  may find  it  useful to  look to the judgments  of those
   134   135   136   137   138   139   140   141   142   143   144