Data Science

2023 | Taylor, R.L., Gerber, L. and Healy, D.

A climate emergency was acknowledged in 2021 (IPCC, 2021) as greenhouse gas concentrations in the atmosphere are continually rising and are a catalyst for global warming. A key factor of the acceleration of atmospheric greenhouse gases is fossil fuel combustion; national estimates indicate that 38% of atmospheric CO2 comes from road transportation (BEIS, 2019). To gain accurate local data and a better understanding of the sector’s impact, a network of Raspberry Shake seismometers has been deployed across Greater Manchester, as part of a UKRI NERC-funded project “Listen to Manchester” (Twitter: @listen2mcr). These seismometers possess high sensitivity to high frequency anthropogenic noise, making them suitable for capturing local seismic signatures.

The urgent need to comprehend and predict urban traffic patterns stems from the alarming levels of atmospheric CO2, which have reached their highest point in the last 650,000 years (Lüthi et al. 2008), and the visible air quality improvements seen during the 2020 global pandemic. Understanding factors influencing air quality and traffic volume is essential for sustainable urban planning and the development of effective transportation management strategies.

In recent years, the application of artificial intelligence (AI) techniques in geosciences has gained attention and has had positive impacts on geoscience research. In this study, we employ AI algorithms to detect seismic signals associated with anthropogenic noise. By extracting features from the power spectrum of the frequency domain, we predict urban traffic volumes along Manchester City Centre's Oxford Road corridor.

Our research aims to provide valuable insights into the local dynamics of CO2 emissions and the influence of road travel, particularly when combined with air quality data. These insights will be vital for smart city development and advancing our understanding of climate change. The integration of AI techniques with geoscientific data holds immense potential to facilitate evidence-based decision-making to mitigate its detrimental effects.

Keywords: Passive Seismic Monitoring, Machine Learning, Raspberry Shake.

2021 | Narock, T., Taylor, R.L., Goldstein, E., Boston, A., Irawan, D.E.

EarthArXiv is a preprint service for the Earth sciences — a web-based system that enables open access publishing of non peer-reviewed scholarly manuscripts before publication in a peer-reviewed journal. In this presentation, we provide analytics on the usage of EarthArXiv across a number of sub-disciplines of Earth science. Data indicate that the service in general is growing, but with submission rates varying amongst discipline. The trend of the preprint-to-postprint ratio for each discipline also provides insight into how the various Earth science communities are using the service. We investigate were preprints are published after submission to EarthArXiv and examine how many of these publication venues are listed in the Directory of Open Access Journals. Finally, we will discuss future opportunities we are exploring to make preprints more accepted and easier to use. 

Keywords: Metadata; Preprints; Citations; Data Science

| Taylor, R.L. and Chandler, M.

Fractures in rocks have a significant impact on their physical properties and are critical for various applications in the energy industry. They are particularly important in the study of carbon capture and storage (CCS) because they can provide pathways for CO2 to migrate from storage reservoirs. However, the identification and characterization of fractures in shale rocks is challenging due to the fine-grained, and otherwise featureless, nature of the material.

In this study, a deep learning approach for the automated detection of fractures in X-ray computed tomography (XRCT) images of shale rocks is proposed. Three convolutional neural networks (CNNs), including ResNet, VGG, and AlexNet, were trained on a dataset of manually labelled tomograms. The results revealed that the models had limited accuracy in fracture classification, with the CNNs correctly identifying fractures in only about 66% of cases, regardless of the architecture, and AUC of between 0.67 and 0.69, indicating the need for improvement. Consideration of recall/sensitivity as a key measure was highlighted due to the higher cost associated with false positives, particularly in real-world applications such as gas reservoir yield estimation. However, a maximum recall of 0.63 was achieved.

Keywords: Shale, Hydraulic Fracture, CO2 Sequestration, Deep Learning, Fracture Identification.

| Taylor, R.L., Seers, T., Hodgetts, D. and Rutter, E.H. 

...

Keywords:  Fracture, Photogrammetry, Computational Geoscience. 

| Taylor, R.L. 

...

Keywords:  Statistical Analysis, Gravimetry, Machine Learning. 

| Hodson-Tole, E., Taylor, R.L., Adeniran, I. and Degens, H. 

...

Keywords: Skeletal Muscle, Modelling. 

| Bagley, L., Taylor, R.L.,  Maudsley-Barton, S., Bryan, A. and Han, L. 

...

Keywords: Cardiopulmonary Exercise Testing, Machine Learning.