Print this Page

Short Course 4: Extracting, Visualizing and Interpreting Structure in Geochemical Data

Geochemical data collected from government and exploration industry surveys are derived from un-modified (bedrock) to modified (weathered, transported) media that are comprised of minerals. Minerals are ordered structures governed by atomic forces which impose “structure” (mineral stoichiometry) in the way the elements are associated. Rocks, till, soils and other forms of sample media are mixtures of minerals and thus represent a complex array of structured mineral material. In order to discover and interpret the structure in geochemical data, a range of statistical methods are required.

Because geochemical data are compositional, the issue of closure can be dealt with using logratios, as advocated by John Aitchison. This one-day short course will examine the theory and application of logratio analysis to geochemical datasets for the intention of extracting, visualizing and interpreting structure.

Particular attention will be paid to the role of selecting ratios of minerals as well as ratios involving amalgamations of minerals, where these are either pre-specified or identified statistically.  The objective is to simplify the understanding of compositional data so that the geochemist can interpret the essential relationships between the minerals based on a few key ratios.

We will examine the effect of measurement error including detection limits and structural zeroes through interval censoring models and zero replacement approximations. We will also show how amalgamation can greatly reduce the zeroes problem in many cases and how correspondence analysis can be especially useful with dealing with the zeroes problem.

We will also cover the situation where the objective is to understand differences between samples, after accounting for correlation across space or time.

Different methods of visualization of a compositional data set will be highlighted, as a means of facilitating the interpretation and understanding of its structure.  Univariate, bivariate and multivariate graphical displays adapted to the compositional nature of the data will be presented.

All methods presented will be coded in R and demonstrated during the short course, and participants will have access to the R scripts for their own analyses.

 

Date: Monday 3 September 2018

Time: 9 am – 5 pm

Venue: Conference Centre of the Clarion Hotel, Plato Meeting Room

Cost: 120 USD

Course Schedule:

09:00-09:30   Introduction to geochemical data analysis: basic principles, objectives.

09:30-10:30    Aitchison’s approach to compositional data analysis: closure, subcompositional coherence and logratios; logratio analysis.

10:30-11:00     Tea/Coffee

11:00-12:00     Variable selection as a way to simplify understanding of geochemical data; the role of amalgamations. Initial demonstration of R software.

12:00-13:00     Important side issues: the problem of zeros; detection limits; measurement error; correspondence analysis as an alternative approach to dimension reduction in the presence of many zeros.

13:00-14:00     Lunch

14:00-14:30     Some graphical tools to aid understanding of geochemical data: univariate, bivariate and multivarate. Software demonstration.

14:30-15:30      Geochemical data linked to time and space: how to account for spatial and temporal correlations.

15:30-16:00     Tea/Coffee

16:00-17:00     A final case study and discussion of results.

Course Instructors:

John Bacon-Shone is  currently Director of the Social Sciences Research Centre, Associate Dean and Associate Director (Knowledge Exchange), The University of Hong Kong (HKU). He has worked for HKU for more than 35 years, including three years on secondment to the Hong Kong Government assisting with drafting of the Chief Executive’s annual Policy Address and writing speeches and policy documents on health care, migration, population growth, opinion polls, education and technology. At HKU John has been responsible (1) for all quantitative research methods courses in the Graduate School; (2) for knowledge exchange between HKU and the broader non-research community; and (3) for human subjects research in HKU as chairman of the human research ethics committee. Since 2014, he has been the moving force in developing HKU’s policy and procedures for research data management.  His research interests include statistical computing, survey methodology, compositional data analysis, biostatistics, gambling, data archiving, privacy, sociolinguistics and policy research and he was responsible for introducing Computer-Aided Telephone Interviewing to Hong Kong.

 

Michael Greenacre is Professor of Statistics at the Universitat Pompeu Fabra, Barcelona. He specializes in applied multivariate analysis and has written six books on correspondence analysis and related methods, as well as four co-edited books on data visualization.  Amongst the courses he gives is one on Data Visualization in the Master of Data Science in the Barcelona Graduate School of Economics. He has given short courses in 15 countries on all five continents, including an annual course for the last 12 years on multivariate data analysis for ecologists in Tromsø, Norway, and a workshop on compositional data analysis for biochemists at the Alfred Wegener Institute in Germany. Author of over 70 research publications in international journals, including several on compositional data analysis, some of his books are also published online at www.multivariatestatistics.org.  Personal web page: www.econ.upf.edu/~michael

 

Eric Grunsky is a professional geoscientist (P.Geo. British Columbia) whose career has included field mapping and applied research at: the Geological Survey of Canada, Ottawa, the Division of Exploration and Mining, CSIRO, Australia, the Alberta, British Columbia and Ontario provincial geological surveys. His research has focused on the application of multivariate statistical methods and spatial statistics applied to geochemical data. He was the recipient of the Krumbein Medal (2012) and the Felix Chayes Prize (2005) by the IAMG, in recognition for his work in applied geochemistry. Eric has published extensively in peer-reviewed journals and government reports. He has also presented numerous short courses for the interpretation of geochemical survey data. He is the Secretary General for the International Association for Mathematical Geosciences (IAMG) [2016-2020], a fellow of the Association of Applied Geochemists (AAG) and a member of the Geochemical Society. Eric is a Professor at the China University of Geosciences, Beijing, China; an Adjunct Professor at the University of Waterloo, Canada and a member of the Science Advisory Board for the Canadian Mining Innovation Council Footprints Project.

Permanent link to this article: http://www.iamg2018.org/index.php/short-course-4-extracting-visualizing-and-interpreting-structure-in-geochemical-data/