The Earth’s magnetosphere-ionosphere (M-I) system is a highly dynamic plasma environment which is driven by its interaction with the solar wind through a process called “reconnection”. Understanding the M-I system is important, as it provides the scientific basis of space weather (the harmful influence of our plasma environment on space- and ground-based technology).
Ionospheric observations of reconnection are uniquely capable of inferring the global extent, and hence global rate, of the reconnection process. They therefore hold the key to understanding how the global M-I system responds to solar wind driving. However, there are some key unknowns – in particular, fundamental dependencies such as how the upstream conditions control the spatial extent of the interaction process or the “size” of individual reconnection bursts are not known. Our recent work  opens up an exciting opportunity to probe these questions. However, to do so requires large-scale statistical studies, and the challenge is one of data volume. This is a challenge that can be addressed with data science techniques [e.g. 2]. In this project, we will develop automated algorithms to identify reconnection events based on ionospheric radar data, which will allow transformative statistical studies into the nature of M-I driving.
Previous studies of the ionospheric signatures of reconnection have either been individual case studies, or small-scale statistical studies based on events identified “by eye”. We will exploit a 20 year archive of data from a global network of coherent scatter ionospheric radars called SuperDARN (the Super Dual Auroral Radar Network). We will develop automated methods to identify such signatures, allowing much more robust statistical analysis. From a machine learning perspective, this task is challenging as it requires finding subtle spatial-temporal patterns involving many different scales in a noisy background. Over the last decade Deep Learning has made a step change in its ability to recognise such patterns. Technologies such as convolutional neural networks (CNNs), recurrent neural networks such as LSTMs (long short-term memory) and more recently dilated CNNs and attention mechanism have continued to push the state-of-the-art on this type of problem. We will develop and apply such techniques, train them on existing SuperDARN data, validate the resulting algorithms and then apply them to analyse the SuperDARN data statistically in order to determine the response of the ionosphere (and hence M-I system) to different modes of variability in the solar wind driving conditions.
The INSPIRE DTP programme provides comprehensive personal and professional development training alongside extensive opportunities for students to expand their multi-disciplinary outlook through interactions with a wide network of academic, research and industrial/policy partners. The student will be registered at the University of Southampton and hosted in the School of Physics & Astronomy. Specific training will include:
- Attendance at a national introductory summer school in solar-terrestrial physics at the start of the PhD
- Postgraduate lecture series and research seminar series provided in Physics & Astronomy
- Presentation and communication skills will be honed through participation in weekly group meetings, a group journal club, and presentation of findings at national and international workshops and conferences.
- Development of programming skills will be a key element of the training provided, as computer programming will be integral to the data analysis undertaken.
- As a member of the Vision, Learning and Control group the student will be trained in deep learning frameworks and high-performance computing in one of our within-group training sessions. They will also be invited to engage with our machine learning reading groups and other activities.
- The student will also be invited to join undergraduate and PGT ECS courses in machine learning and deep learning, and be offered the opportunity to work as a lab demonstrator.
Please see https://inspire-dtp.ac.uk/how-apply for details.
 Fear, R. C., Trenchi, L., Coxon, J. C., & Milan, S. E. (2017). How much flux does a flux transfer event transfer? Journal of Geophysical Research, 122, 12310–12327. https://doi.org/10.1002/2017JA024730
 Zhang, Yan, Hare, Jonathon and Prügel-Bennett, Adam (2018) Learning to count objects in natural images for visual question answering. International Conference on Learning Representations, Vancouver Convention Center, Vancouver, Canada. 30 Apr - 03 May 2018. pp. 1-17, https://doi.org/10.48550/arXiv.1802.05766