Library Subscription: Guest
Begell Digital Portal Begell Digital Library eBooks Journals References & Proceedings Research Collections
International Journal for Uncertainty Quantification

Impact factor: 1.000

ISSN Print: 2152-5080
ISSN Online: 2152-5099

Open Access

International Journal for Uncertainty Quantification

DOI: 10.1615/Int.J.UncertaintyQuantification.2018019782
Forthcoming Article

Comparison of Risk Analysis Methodologies in a Geostatistical Context: Monte Carlo with Joint Proxy Models and Discretized Latin Hypercube

Susana Santos
University of Campinas
Ana T Gaspar
University of Campinas
Denis J Schiozer
University of Campinas


During the development of petroleum fields it is fundamental to assess risk. Several methods are presented in the literature but its choice must agree with the level of uncertainty and with the type of uncertain parameters, to ensure low computational costs and reliable results. We compare two risk analysis methodologies applied to a complex reservoir model comprising a large set of geostatistical realizations: (1) generation of scenarios using the Monte Carlo sampling technique combined with joint proxy models (JMM), and (2) generation of scenarios using the Discretized Latin Hypercube sampling technique combined with geostatistics (DLHG). For a reference response, we conducted a risk analysis using the classic Monte Carlo simulation (MC) with a very high sampling number. We compared the methodologies looking at the: (1) accuracy of the results, (2) computational cost, (3) difficulty in the application, and (4) limitations of the methods. The results show that both methodologies are reliable but reveal limitations in the JMM. Due to the way this method captures the effect of a geostatistical uncertainty, the number of required simulation runs increased exponentially and became unfeasible to consider more than 10 realizations. The DLHG method showed advantages in such a context, as it obtained more precise results from less than half of the simulation runs, and without needing proxy-models, as the risk curves are computed directly from the simulation results. In addition, this method has a fast straightforward application, and used all geostatistical realizations (in the order of 10²).