2009-2010 IAMG Distinguished Lecturer Series

Roussos Dimitrakopoulos, the IAMG Distinguished Lecturer for 2009, is professor and holds the Canada Research Chair (Tier I) in “Sustainable Mineral Resource Development and Optimization Under Uncertainty – BHP Billiton”, at the Department of Mining and Materials Engineering, McGill University in Montreal, Canada.   Roussos serves as the Editor-in-Chief of the journal of Mathematical Geosciences published by Springer and is also Director of McGill’s COSMO Laboratory.  Previously he was Professor and Director of the Bryan Research Centre, Univ. of Queensland, Australia. He holds a PhD from École Polytechnique, Montreal, and a MSc from the University of Alberta, Edmonton. He has been working in stochastic spatial simulation and optimization since 1983, and the last decade on risk-based optimization in mine planning and valuation. Roussos has been Senior Geostatistician with Newmont Mining Co., Denver, and Senior Consultant with Geostat Systems Int’l. He has taught and worked in North America, Australia, South America, Europe, the Middle East, South Africa and Japan.
URL: http://people.mcgill.ca/roussos.dimitrakopoulos/

Institutions interested in having Prof. Dimitrakopoulos visit should contact the DL Committee Chairman, Sean McKenna, at samcken@sandia.gov
 

Lecture 1

An Overview of Modern Stochastic Conditional Simulations:  Fast and efficient, point and block support, Gaussian and non-Gaussian including high-order, sequential simulations       

Modeling the spatial uncertainty of natural phenomena may require large size simulations (grid sizes up to 108) and a new ‘line’ of sequential approaches with low computational costs can be used.  After giving examples of the ‘size’ issue, this presentation provides a general overview of sequential decomposition of a pdf for simulating very large fields at point-support scale.  Subsequently, the approach is expanded to the direct simulation at the block-support scale. The differences in computational performance is documented in examples and further discussed for the case of efficient multivariable simulations.  The last part of the presentation considers an expansion of sequential approaches beyond the second-order methods currently employed, and shows how the sequential framework is developed to high-order, non-Gaussian, non-linear simulation.        

Lecture 2

An Introduction to Stochastic Simulation: Basic concepts made easy and examples     

Modeling the spatial uncertainty of natural phenomena using geostatistical or spatial stochastic simulations is commonly used.   This presentation aims to introduce the non specialist to:  (a) basic concepts presented in an intuitive way, through examples; (b) the type of problems addressed with respect to natural spatial or spatial-temporal phenomena; (c) introduce the concept of random number generation; (c) the generation of correlated numbers and conditional distributions; (d) the ‘intuitive’ sequential Monte Carlo sampling; and (e) using the above to solve different problems (environment, mining, reservoirs).       

Lecture 3

High-order Geostatistics:  Simulating complex, non-Gaussian geological and environmental phenomena

Geo-science and engineering related phenomena such as characteristics of mineral deposits and attributes of petroleum reservoirs, pollution levels, the earth’s surface temperature, and so on, represent complex natural systems distributed in space. Their spatial distributions are currently predicted from finite measurements and second-order spatial statistical models. The latter models are limiting, as geo-systems are commonly highly complex, non-Gaussian and exhibit non-linear patterns of spatial connectivity. Non-linear and non-Gaussian high order geostatistics is a new area of research based on higher-order spatial connectivity measures termed spatial cumulants.  

In this presentation, definitions of high-order statistics are first given, then, the inference with spatial templates and interpretation of anisotropic cumulants are introduced. Several examples are presented to elucidate the concepts stressing the physical interpretation of cumulant maps.  Subsequently, new research results on ‘high-order’ conditional simulations are shown.  A new simulation method is outlined and is founded upon spatial cumulants in the high-order space of Legendre polynomials. The method does not require any data pre-processing or transformations, it is shown in the examples presented to reproduce complex spatial geometries, bimodal data distributions, and the high-order cumulants of the data used.  The presentation concludes with the ‘down stream’ effects from the use of simulation approaches to engineering problem solving.   
 
Lecture 4

An Extended View of Mining Geostatistics:  Integrating short- and long- term mine production forecasting under uncertainty and application in a major gold mine

Do our models work?  If they do, what could they encompass?  How do our predictive models compare to reality?  What type of problems surface in the world of engineering?  These are the types of questions addressed here, through a specific example from the world of mining and metal production.  The presentation explores stochastic optimization for mine production scheduling as a space and time problem, integrated with stochastic simulations of orebodies with data updating capabilities, and simulation of non-available “future data”.  A large gold mine and tests conducted demonstrate that problems exist, how stochastic solutions perform, and how this adds value to the operation. 

Lecture 5

Mining Geostatistics Revisited: Limits of the current paradigm, non-linearity of the chain of mining, extended stochastic solutions, applications and monetary value  

Conventional approaches to estimating reserves and optimization for mine planning and production forecasting result in single, often biased forecasts. This is largely due to the non-linear propagation of errors in understanding orebody attributes from a limited finite number of drilling data., throughout the chain of mine planning and mining  A ‘redefinition’ of mining geostatistics is considered to include two interacting and potentially fusing elements: stochastic simulation and stochastic optimisation. These two elements provide an expanded mathematical framework that allows modelling of orebody uncertainty and its direct integration to mine design, planning and valuation of mining projects and operations. The pertinent mathematical models and multiple examples show the key characteristics and value of this redefined geostatistical modelling framework. 

last updated 2009-06-21