Early uncertainty assessment: application to a hydrocarbon reservoir appraisal

Guillaume Caumon and A. G. Journel. ( 2004 )
in: Geostatistics Banff, Proc. of the seventh International Geostatistics Congress, Kluwer, Dordrecht

Abstract

Assessment of uncertainty of global resources from sparse appraisal data is a difficult challenge. While many algorithms have been defined to compute one single “best” estimate a* of the unknown global value a, assessing the uncertainty calls for the definition, necessarily subjective, of a randomization process. Most error assessment algorithms, including bootstrap resampling, consider a randomization of the global estimate a*. We suggest a joint randomization of both the unknown a and its estimate a* within a Bayesian framework, given alternative plausible geological scenarios. This allows for: considering a prior probability distribution for the unknown target value a based on analog studies obtaining the data likelihood by spatial bootstrap instead of using some arbitrary analytical distribution assessing the value of data in reducing the prior uncertainty, a prerequisite to decide on new data acquisition strategies. The proposed procedure does not call for data independence nor Gaussian assumptions which are seldom met in practice. It accounts explicitly for alternative geological interpretations of the quantitative data available, a critical source of uncertainty too often ignored. The method is applied to a complex synthetic fluvial reservoir.

Download / Links

BibTeX Reference

@INCOLLECTION{Caumon04Banff,
    author = { Caumon, Guillaume and Journel, A. G. },
    editor = { Leuangthong, Oy and Deutsch, C. V. },
     title = { Early uncertainty assessment: application to a hydrocarbon reservoir appraisal },
 booktitle = { Geostatistics Banff, Proc. of the seventh International Geostatistics Congress },
   chapter = { 0 },
      year = { 2004 },
 publisher = { Kluwer, Dordrecht },
       doi = { 10.1007/978-1-4020-3610-1_56 },
  abstract = { Assessment of uncertainty of global resources from sparse appraisal data is a difficult challenge. While many algorithms have been defined to compute one single “best” estimate a* of the unknown global value a, assessing the uncertainty calls for the definition, necessarily subjective, of a randomization process. Most error assessment algorithms, including bootstrap resampling, consider a randomization of the global estimate a*. We suggest a joint randomization of both the unknown a and its estimate a* within a Bayesian framework, given alternative plausible geological scenarios. This allows for:
considering a prior probability distribution for the unknown target value a based on analog studies
obtaining the data likelihood by spatial bootstrap instead of using some arbitrary analytical distribution
assessing the value of data in reducing the prior uncertainty, a prerequisite to decide on new data acquisition strategies.
The proposed procedure does not call for data independence nor Gaussian assumptions which are seldom met in practice. It accounts explicitly for alternative geological interpretations of the quantitative data available, a critical source of uncertainty too often ignored. The method is applied to a complex synthetic fluvial reservoir. }
}