Introduction to Machine Learning (ML) for Geophysical Applications
More and more Machine Learning (ML) will play a role not only in society in general but also in the geosciences. ML resorts under the overall heading of Artificial Intelligence. In this domain often the word “Algorithms” is used to indicate that computer algorithms are used to obtain results. Also, “Big Data” is often mentioned, indicating that these algorithms need an enormous amount of input data to produce useful results.
Many scientists mention “Let the data speak for itself” when referring to machine learning, indicating that hidden or latent relationships between observations and classes of (desired) outcomes can be derived using these algorithms. A clear example is in the field of Quantitative Interpretation. For clastics we have a reasonable understanding in which cases known rock properties expressed in equations can be used to predict say pore fluids. But for carbonates it is often an enigma and we have to resort to statistical relationships. Then ML enters into the game. If we have many wells with known drilling results, the algorithms can derive non-linear relationships between seismic observations and the known well results (supervised learning). But sometimes it is already useful if an algorithm can define separate classes (say seismic facies), which then still need to be interpreted (unsupervised learning).
The aim of this 1-day course is to introduce how Machine Learning (ML) is used in geophysical applications. It will give an understanding of the “workflows” used in ML. The used algorithms can be studied separately using references. Power-point presentations will introduce various aspects of ML, but the emphasis is on computer-based exercises using open-source software. The course concerns a genuine geophysical issue: predicting lithology and pore fluids, including fluid saturations. The input data are Acoustic and Shear Impedances, Vp/ Vs ratios and AVA Intercept and Gradients. The exercises deal with preconditioning the datasets (balancing the input classes, standardization & normalization of data) and applying several methods to classify the data: Bayes, Logistic, Multilayer Perceptron, Support Vector, Nearest Neighbour, AdaBoost, Trees. This for supervised as well as unsupervised applications. Non-linear Regression is used to predict fluid saturations.
The objectives of this course are to:
- Have a good understanding on how and when ML can be applied effectively in the geosciences;
- Realize the workflows that can be used in ML;
- Solve the main issue of ML, namely choosing the appropriate algorithm and its parameters.
This course is meant for all those who are interested in understanding the impact Machine Learning will have on the Geosciences and then specifically the impact on seismic and non-seismic data acquisition, processing and interpretation. Hence, geologists, geophysicists and petroleum and reservoir engineers, involved in exploration and development of hydrocarbon fields, but also those working in shallow- surface geophysics.
About the Instructor
Jaap Mondt has a Bachelor’s degree in Geology (University of Leiden) and a Master’s degree in Geophysics (University of Utrecht), PhD in Utrecht on “Full wave theory and the structure of the lower mantle”. He then joined Shell Research to develop methods for Quantitative Interpretation. Subsequently worked in Shell Expro in London where he was actively involved in acquiring, processing and interpreting Offshore Well Seismic data. After his return to The Netherlands he headed a team for the development of 3D interpretation methods using multi-attribute statistical and pattern recognition methods. After a period of Quality Assurance of “Contractor” software for seismic processing, he became responsible for Geophysics in the Shell Learning Centre. During that time, Mondt was also part-time professor in Applied Geophysics at the University of Utrecht. From 2001 till 2005 worked on the development of Potential Field Methods (Gravity, Magnetics) for detecting oil and gas. Finally, became a champion on the use of EM methods and became involved in designing acquisition, processing and interpretation methods for Marine Controlled Source EM (CSEM). After retirement he founded Breakaway, providing courses on acquisition, processing and interpretation of geophysical data (seismic, gravity, magnetic and electromagnetic data).
In the last couple of years, he developed a keen interest in the use of Machine Learning for Geophysical Applications and developed a practical Machine Learning course for Geophysicists and Interpreters.