Print this Page

Short Course 1: Machine Learning for Geoscience Modelling: Introduction and Advanced Topics with Case Studies

At present, there is a considerable increase in geo- and environmental data coming from different sources, including numerous earth systems monitoring networks (meteorology, climate, pollution, natural hazards) and remote sensing (satellite imagery, drones, LIDAR, etc). In order to be used efficiently in earth system modelling and prediction, these data have to be processed and understood. Recent trends in earth system prediction modelling and decision making requires more adequate and consistent prediction on uncertainty, which requires new ways of blending a model driven and a data driven vision. A traditional approach where the data are integrated along a single modelling concept is not flexible enough in uncertainty studies to make robust decisions.

Rapid development of Machine Learning (ML) and its links with statistical inference enables developing new vision for “big data” integration and prediction modelling. Recently, ML algorithms, which are nonlinear and robust tools for the analysis, modelling and visualization of complex high dimensional data, have gained a great popularity in geoscience fundamental studies and applications.

The short course presents an introduction to basic ML models and discusses advanced topics of ML application in various geoscience modelling. Course participants will become familiar with a data driven approach to modelling vs model based. We will introduce you to the basic concepts and provide an overview of machine learning applications for spatial geoscience problems. The course will cover  regression and classification algorithms applications, how to tune them, validate and interpret the results. The course does not entail use of a particular software but we will share references to some codes available. Prior knowledge of machine learning is not required.

 

Date: Sunday 2 September 2018

Time: 9 am – 6 pm

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

Cost: 135 USD

Course Outline:

  • Introduction to data driven modelling of geo- and environmental data;
  • supervised and unsupervised learning;
  • basic machine learning algorithms;
  • complexity and predictability; feature selection;
  • uncertinty quantification in prediction modelling;
  • geological realism in predicting natural systems;
  • advanced topics and recent developments;
  • simulated and real data case studies.

Course Instructors:

Mikhail Kanevski is currently a full Professor of geomatics at the Faculty of Geosciences and Environment,University of Lausanne, Switzerland. He is teaching courses on Geographical Information Systems, Geostatistics and Environmental Data Mining. His scientific interests cover a wide range of basic and applied topics: geomatics (GIS and remote sensing), environmental modelling, geostatistics, machine learning and data mining, fractals in Earth Sciences, natural hazards and environmental risks, renewable energy assessment.

 

Vasily Demyanov is an Associate professor with Heriot-Watt University, Edinburgh. He lectures MSc and industry courses in geostatistical modelling and uncertainty quantification of subsurface reservoir predictions. Vasily’s research interests are in machine learning application to predict earth systems performance with respect to uncertainty. Vasily has organized several dedicated conference sessions and workshops on these topics. He is also an Associate Editor of Computers and Geosciences Journal.

 

Permanent link to this article: http://www.iamg2018.org/index.php/machine-learning-for-geoscience-modelling/