Page 151 - Modern Spatiotemporal Geostatistics
P. 151

132       Modern  Spatiotemporal  Geostatistics —  Chapter 6

            BME  analysis is easily modified  to  handle various other  cases of  practical
        importance  arising  in  physical applications  in which  soft  data  are available  not
        only  at  the  data  points  p aoft,  but  also  at  the  estimation  point p k  itself  (see
        Example 3.17, p. 86). Such a situation is presented in the following proposition.
        PROPOSITION 6.5:    Assume that  the  specificatory  knowledge  includes
        the  hard  data  (Eq. 3.30) and  the  probability  law  fs(x sojt^  Xk)>  where
        So  : X SOft  € I  and Si  ' Xk € 4  so that  S=  So U Si-  Then, the  posterior
        operator  has the  form




            Similar  expressions for the f)f-operator can be derived in cases where other
        forms  of soft  data  are available at  the  estimation  points.
        EXAMPLE 6.6:  From  Equation  6.16 one obtains  the  posterior  pdf f K(xk)  =
        ^"^[Xtnap!   PfeL  where




        is  the  normalization  parameter.  In  the  special  case  that  fs(x softi  Xk)  =
        fso  (X i0ft)fsi  (Xk),  Equation  6.16 yields  the  recursive  relationship  f K(xk)  =
          1
        ^~ /5i(Xfe)/3Co(Xfe)-  where  f Ko(xk)  is  the  posterior  pdf  associated  with
                 and
            Depending  on the  practical  application,  various combinations  of  the  pre-
        ceding  results are possible.  Furthermore,  additional  classes of ^-operators can
        be created that  incorporate  many  other  kinds  of  soft  data,  including the  intu-
        ition,  beliefs,  and  subjective  assessments of  laymen.  However,  these  kinds  of
        soft  data  must  be fitted  into  an organized  and coherent  system of  knowledge
        before they  can attain significance or applicability.  Indeed, it  is the rich  network
        of  physical  knowledge  bases created  by generations of  scientists  and engineers
        that  gives  meaning to  such  soft  data  as is occasionally  provided  by laymen.


        Discussion
        Certain  of  the  previous  results  can  be summarized  in  terms  of  the  following
        useful  expressions of  the  BME  posterior  operator  and pdf



            with

        where  A  is the  normalization  parameter;  and the B,  D  and E s  determine  the
        form  of  the ^-posterior  operator  (Table  6.1).  This  form  depends, of  course,
        on the  specificatory  knowledge available.
   146   147   148   149   150   151   152   153   154   155   156