IAP-24-053

Computational innovation for the energy transition: Bayesian stratigraphic correlation of subsurface data

Understanding subsurface geology underpins the UK’s green energy transition by enabling technologies such as geothermal energy, carbon capture and storage, hydrogen containment, and the safe storage of nuclear waste. Our understanding of subterranean geological structure draws on geophysical, chemical and sedimentological data obtained from individual boreholes. Correlating geological formations between these boreholes traditionally uses visual methods. This subjective approach can struggle to integrate diverse datasets, and to quantify uncertainties. An objective and systematic approach to stratigraphic correlation that better predicts the structure and distribution of geological units will increase the feasibility of subsurface energy projects.

This project will expand the supervisory team’s novel correlation software StratoBayes¹, which uses a Bayesian algorithm to correlate geophysical and geochemical stratigraphic records. The successful student will extend the algorithm to broaden the range of stratigraphic data that can be used for geological interpretation, allowing the production of more reliable correlations based on large quantities of diverse data.

Starting from existing subsurface datasets held by the British Geological Survey (BGS; see for example ref. 2), the student will develop a multi-step modelling framework that can consecutively update stratigraphic models with new data. This involves upgrading the underlying Markov Chain Monte Carlo (MCMC) algorithm in StratoBayes to handle large subsurface datasets such as well log data, which are essential for exploring carbon capture and storage and or geothermal energy potential. The new multi-step algorithm will be refined using Lower Jurassic data and validated against well-established ammonite biozones.

Further methodological innovations will include integrating additional information such as bio- and lithostratigraphy, and automated lithology identification based on geophysical signals. Model development will be driven by continuous testing with real-world data sets, and the improved software be used test and refine stratigraphic models of key formations.

By generating high-resolution, probabilistic stratigraphic models, this research will provide crucial insights into the stratigraphic framework of the UK. These subsurface models will ultimately contribute to the precise deployment of green subsurface technology, advancing the UK’s transition towards a net-zero economy.

OBJECTIVE ONE: A multi-step Bayesian correlation framework

The first project phase will see the development of a multi-step modelling framework that builds correlations by sequentially integrating additional wells or sections.

1. Build modelling framework and integrate the existing StratoBayes algorithm to allow for sequential updating and expanding of stratigraphic models.

2. Validate the modelling framework against Lower Jurassic datasets from BGS. Adapt the modelling framework based on the test results.

3. Update the underlying MCMC sampling to accelerate model convergence in the new framework.

4. Apply the enhanced StratoBayes framework to subsurface datasets to test and evaluate existing stratigraphic correlations.

OBJECTIVE TWO: Methodological innovations in stratigraphic modelling

The second phase is focussed on methodological improvements to match the requirements of modelling diverse stratigraphic data sets.

1. Expand the algorithm to integrate bio- and lithostratigraphy via the Bayesian likelihood.

2. Add functionality to identify lithologies based on geophysical measurements.

3. Increase the capability to handle large datasets by exploring computationally more efficient algorithms than currently implemented.

OBJECTIVE THREE: Stratigraphic correlation of key geological units

The enhanced stratigraphic modelling software will be applied to challenging stratigraphic units using BGS data, including from wireline logs, biostratigraphy and cyclicity, in order to improve the understanding of subsurface structures critical for the UK’s energy transition.

1. Use the improved StratoBayes modelling framework to generate high-resolution, probabilistic stratigraphic correlations of the selected units.

2. Where available, compare the modelling results with existing stratigraphic frameworks, geological markers, and outcrop data.

3. Refine stratigraphic models in collaboration with BGS experts on regional geology.

The student will build the Bayesian modelling framework in a suitable programming language (e.g. R, Julia, C++) based on the existing StratoBayes software written in R. Full training in programming, Bayesian statistics and model development will be provided by the supervisory team and Durham’s Advanced Research Computing unit (ARC). Complex model runs of large datasets will be conducted on Durham’s Hamilton or Bede supercomputers.

Stratigraphic datasets will be provided by BGS and formatted by the student to enable their use in the modelling software. Expertise on stratigraphic principles and regional geology will be provided by the supervisory team and regional experts at BGS, and training will include visits to BGS National Geological Repository at Keyworth and a field workshop in southwest England.

Year 1

Introduction to StratoBayes; training in Bayesian statistical modelling, MCMC algorithms and computational implementation (Millard, Smith, Eichenseer, ARC). Implementing a multi-step modelling framework and testing and improving it with stratigraphic data sets in collaboration with BGS partners. Visit to the Keyworth core repository and coastal sections in southwest England, for training in stratigraphic principles (Burrel Garcia).

Year 2

Enhancement of the MCMC algorithm and application of the improved model to selected subsurface data sets from BGS. Evaluate existing stratigraphic correlations against the results obtained with StratoBayes. Expanding the Bayesian model to include bio- and lithostratigraphy.

Presentation of Year 1 Results at e.g. the Energy Geoscience Conference.

Year 3

Integrating a lithology identification component in the model and exploring computationally efficient alternatives to existing model components. Application of the model to key stratigraphic units, such as the Chalk Group and Kimmeridge Clay Formation. Refine resulting models with BGS experts.

Presentation of Year 2 Results at e.g. the International Congress on Stratigraphy.

Year 3.5

Preparation of thesis; publication of papers.

Presentation of Year 3 results at e.g. the EGU conference.

The student will benefit from fortnightly, interdisciplinary meetings of Millard and Smith’s collaborators, comprising Earth scientists, archaeologists and mathematicians. The student will be taught state-of-the-art Bayesian modelling, MCMC methods and programming (Millard, Smith, Eichenseer). Research visits to the British Geological Survey (BGS) will equip the student with practical knowledge of subsurface geology and stratigraphic modelling. Additional computing training will be obtained by Durham’s Advanced Research Computing team, including training on using supercomputers, preparing the student for a career within or outside Academia. Attending workshops on software development, stratigraphy and Bayesian modelling will provide additional training, and speaking skills and networking will be developed at scientific conferences.