Leeming, K., Martin, E. ORCID: https://orcid.org/0009-0005-9135-3382, Baker, S., Chazaridis, N., Dalton, D., Frayling, L., Kadochnikova, A. and Tsiakmakis, D. (2025) Data Study Group final report - British Geological Survey: identifying potential for Carbon Capture and Storage in Rock. Report. The Alan Turing Institute, London, UK. (doi:10.5281/zenodo.14973572)

Preview

PDF (Published Report) 50249 MARTIN_Data_Study_Group_Final_Report_British_Geological_Survey_(REPORT)_2025.pdf - Published Version Available under License Creative Commons Attribution. Download (12MB) | Preview

Abstract

The injection and sequestration of CO2 in deep geological repositories is a promising strategy for reducing atmospheric CO2 concentrations. Deep storage of hydrogen could act as a ‘battery’ for storing energy long term. Both technologies are supporting efforts to achieve the targets established by the UK Government’s Net Zero energy strategy [11]. Deep repository injection is a costly and complex undertaking; therefore, a careful, advanced assessment of the suitability of a candidate rock formation for injection and storage is essential and urgently needed. The suitability of a candidate rock formation depends on several properties of its sedimentology microstructure, and conventional analysis methods are time-intensive and costly on a per-sample basis. The development of
an automated and reliable characterisation process would be highly beneficial in determining the suitability of rock formations for injection and storage. The objective of this Data Study Group Challenge was to develop and demonstrate automated methods for the characterisation of microscopic rock samples using images of thin-section samples provided by the British Geological Survey. The dataset consisted of 119 thin-section rock samples, each imaged under eight different lighting conditions, 13 of which had additional labelled images. The labels indicated the mineral grains within the images. We divided the challenge into three related tasks: assessing porosity (using pixel-wise thresholding rules), identifying grain boundaries (using the watershed algorithm), and determining mineralogy (using the random forest and U-Net algorithms). We demonstrated that these approaches show promise for automating thin-section analyses and highlighted opportunities for leveraging the results of each task to improve the others. Porosity estimation can serve as an initial step for the watershed algorithm in grain detection, which, in turn, can define grains for applying the random forest feature-based mineral classification. The identified grain boundaries can be used to refine the mineral label images, which were available at a lower resolution than the other images. These refined label images could then be used to train the U-Net algorithm for mineral detection. We have identified several potential areas for further work, as each task could undergo iterations to enhance the algorithms. A key area for improvement across all tasks is the incorporation of a greater number of input image layers (i.e.,different lighting and polarisation conditions) to refine the outputs. Due to time and resource constraints, not all eight available lighting condition images were utilised for each task. Learning from all images simultaneously could enhance porosity estimation, grain boundary detection, and mineral classification. Additionally, this challenge facilitated discussions regarding the explainability of the algorithms used. The random forest feature-detection approach follows a more traditional human analytical process by identifying grains and their optical properties to perform classification. This approach may be preferable in some applications compared to the U-Net classification method, which utilises the optical input data in a less interpretable manner to make mineral classification decisions.

Item Type:	Monograph (Report)
Uncontrolled Keywords:	Data Study Group, The Alan Turing Institute, British Geological Survey, Carbon Capture and Storage, Net Zero, rock formations, automated and reliable characterisation process
Subjects:	Q Science > Q Science (General) T Technology > T Technology (General) T Technology > TP Chemical technology
Faculty / School / Research Centre / Research Group:	Faculty of Engineering & Science Faculty of Engineering & Science > School of Engineering (ENG)
Related URLs:	Publisher
Last Modified:	28 Apr 2025 11:39
URI:	http://gala.gre.ac.uk/id/eprint/50249

Actions (login required)

View Item

Downloads