Novel constrained potential field data inversion techniques applied to exploration

Course Description

This short course introduces geoscientists to recent methods for potential field inversion developed to mitigate the limitations in the currently available commercial codes, starting with physical property inversion (i.e., inverting for density or magnetic susceptibility), followed by geometrical inversion (i.e., inverting for the geometry of rock units). This course provides an overview of unconstrained inversion techniques and of the application of advanced petrophysical constraints for physical property inversion using the Tomofast-x inversion engine. Geometrical inversion approaches such as level set and trans-dimensional inversion are also covered. With practical applications in mind, techniques are presented that allow running large scale inversion with, e.g., wavelet compression or efficient sub-sampling of the data to invert.

To complement the presented inversion approaches, the course also covers concept of null space - the set of model variations that leave the data misfit unchanged or nearly unchanged - and how this can be leveraged to generate alternative models. Methods to use null space navigation as a tool for model exploration, scenario generation and uncertainty assessment are presented and hands-on exercises guide participants through theoretical concepts and field case studies. 

Practical examples of recently developed codes rely on the Python language and run in the cloud using Google Colab and Jupyter Notebooks. The codes used for the practicals are open source, enabling participants to continue developing and applying these techniques independently after the course. 

A custom QGIS plugin will be provided to help users build input files for the inversion code. The codes are Open-Source so course participants will leave with the tools they need to further develop their skills. Instructions on how to install the full inversion code locally or on remote machines will be provided prior to the course. 

Demonstrations of other QGIS plugins of interest to perform processing of gravity/mag grids, field mapping and 3D geological modelling will be presented to demonstrate the scope for integrated geological and geophysical studies.

 This course might be particularly relevant for participants who have an interest the EAGE short-course "Uncertainty Quantification and Pattern Detection on Model Ensembles" as outputs from the taught methods can be used as input for this course.

Course Outline

•  Introduction

Overview of current state of inversion research globally, brief review of available codes, and specifics of what will be covered.

•  Introduction to unconstrained inversion and ADMM constraints using Tomofast-x 

General presentation of the Tomofast-x open-source potential fields inversion code, and introduction to the ADMM petrophysical bound constraints with field application.

• Hands on: Tomofast-x inversion & ADMM (part 1)
• Hands on: Tomofast-x unconstrained inversion & ADMM (part 2) 
• Introduction to null space analysis, the example of gravity and magnetics 

Exploring the concept of “null space”, how to perturb a model without changing (too much) its misfit and generate new solutions quickly.

• Introduction to geometrical inversions using level-sets, presentation of a field example. 
• Case study in the Pyrenees using gravity analysis of slab subduction

Field application of null space navigation to investigate several geological scenarios. 

• Hands on: Null space navigation (part 1) Synthetic models using gravity and magnetic data -- density and magnetic susceptibility perturbations.
• Hands on: Null space navigation (part 2) Field application using gravity data -- geometry perturbation and model sampling.
• Hand on:  catch up on either null-space navigation or Tomofast-x inversion. 
•  DEMO of other geology-geophysics QGIS plugins
• Introduction to trans-dimensional inversion. 
• DEMO trans-dimensional inversion and analysis of results
• DEMO Interactive examples using synthetic and field data using QGIS plugin 

 Group discussion and Q&A.

Participants’ Profile

The course is intended for geoscientists with experience in the modelling and interpretation of geophysical data who are interested in testing new tools and in learning about and gaining exposure to recent developments.

Prerequisites

Participants should have some knowledge of geophysical inversion and be familiar with the basics of Python. 

About the Instructor

Jérémie Giraud, PhD, is a geophysicist with a particular focus on the integration of geo-disciplines. After completing an MSc. Eng. in geophysics from the School and Observatory of Earth Sciences (2012, University of Strasbourg, France), he spent a few years at Schlumberger in Milan, Italy (2012-2015), where he gained experience in joint inversion and worked on developing and applying methodologies for integrated reservoir modelling.

After this, in 2018, Dr Giraud completed his PhD at the Centre for Exploration Targeting (CET) at the University of Western Australia (Perth). His research focused on integrating geological modelling and petrophysical data into geophysical inversion. Dr Giraud then worked on the development of methods for geometrical inversion at CET. In 2021, he was  awarded a Marie Skłodowska-Curie Fellowship, joining the RING Team (Research for Integrative Numerical Geology, National School of Geology) and collaborating with the Loop and MinEx CRC academic-industry consortia to further advance his work on geometrical inversions and develop methods for the exploration of alternative scenarios.

Since January 2025, Dr Giraud has returned to the Centre for Exploration Targeting in Perth, Australia, as the recipient of an Industry Fellowship with support from the Australian Research Council and industry partners. His current project, titled "Integrative geophysics Under cover: getting images of the Unknown Unknowns", focuses on developing techniques for the exploration of the space of geophysically acceptable models under geological and petrophysical constraints.