Value of Information in the Earth Sciences

Course Description

We constantly use information to make decisions about utilizing and managing natural resources. How can we quantitatively analyze and evaluate different information sources in the Earth sciences? What is the value of data and how much data is enough?

The purpose of the course is to give participants an understanding of the multidisciplinary concepts required for conducting value of information analysis in the Earth sciences. The value of information is computed before purchasing data. It is used to check if data is worth its price, and for comparing various experiments.

The course will outline multivariate and spatial statistical models and methods (Bayesian networks, Markov models, Gaussian processes, Multiple point geostatistics), and concepts from decision analysis (decision trees, influence diagrams), and then integrate spatial statistical modeling, geomodeling and decision analysis for the evaluation of spatial information gathering schemes.

Unlike the traditional value of information analysis, this course focuses on the spatial elements in alternatives, uncertainties and data.

A coherent approach must account for these spatial elements, and clearly frame the decision situation - we demonstrate a workflow for consistent integration and apply this in a series of examples. In this course we discuss and show examples of the value of imperfect versus perfect information, where the likelihood model of geophysical measurements is less accurate. We also discuss the value of total versus partial information, where only a subset of the data are acquired.

Course Objectives

Upon completion of the course, participants will be able to:

  • Frame a spatial decision situation with alternatives, experiments and spatial geomodelling;
  • Use a workflow to conduct value of information analysis in spatial situations;
  • Interpret and compare the value of information of different spatial experiments.

Course Outline

  • Motivation for value of information analysis in the Earth sciences;
  • Decision analysis and the value of information:
    • decision making under uncertainty, value functions, utility, decision trees, influence diagrams, value of perfect information, value of imperfect information,
    • run simple demo example / project on computer;
  • Statistical modeling and spatial modeling:
    • Bayesian networks, Markov models, Gaussian processes, non-Gaussian spatial processes. An important element here is conditioning to data (Bayes rule) and the spatial design of experiments, which will be important for the value of information analysis later,
    • run demo / project on computer;
  • Value of information analysis for spatial models:
    • framing of spatial decision situations and opportunities for spatial data gathering,
    • partial and total spatial information / imperfect and perfect spatial information,
    • coupled or decoupled spatial value function,
    • develop a workflow for value of information analysis in spatial applications,
    • run demo / projects on computer;
  • Examples of value of information analysis in various energy transition applications: petroleum, mining, CO2 sequestration, hydrology, groundwater and wind energy production:
    • description of decision situations, statistical modeling, data gathering opportunities,
    • run demo / project on computer.

Participants’ Profile

The course is designed for students, researchers and industry professionals in the Earth and environmental sciences who has interests in applied statistics and/or decision analysis techniques, and in particular to those working in petroleum, mining or environmental geoscience applications.

Participants should have knowledge of basic probability and statistics, and mathematical calculus. Although it is not essential, it helps to know basic multivariate analysis and decision analysis or optimization. The participant must be willing learn statistical topics and earth science applications, and appreciate the multidisciplinary approach to solving quantitative challenges.

About the Instructor

Jo Eidsvik is Professor of Statistics at the Norwegian University of Science and Technology (NTNU), Norway. He has a MSc in applied mathematics from the University of Oslo (1997) and a PhD in Statistics from NTNU (2003). He has industry work experience from the Norwegian Defense Research Establishment (1998-1999) and from Equinor (2003-2006). He has been a visiting professor at the Statistics and applied mathematical sciences institute (SAMSI) in 2009- 2010 and at Stanford University in 2014-2015. Eidsvik has teaching experience in a variety of statistics courses at the university level, including Statistics, Probability, Applied regression analysis, Stochastic processes, Spatial statistics, Computational statistics. He has been head of the graduate study program in Industrial Mathematics (~50 students every year) and the undergraduate program in physics and mathematics (~100 students every year) at NTNU. He has supervised 45 MSc students and 7 PhD students. He has written about 50 papers in statistical and earth sciences journals.