YES-27-D-13: Using machine learning to model and predict Earth’s magnetic field

Project overview

The geomagnetic field is a fundamental part of planet Earth, existing for at least 3.5 billion years. Yet far from being a constant, the field is very dynamic, exhibiting changes over periods from years to millions of years, including fluctuations and wandering magnetic poles which occasionally fully reverse in polarity. Our magnetic field also plays a significant but often under-appreciated role in modern life: it wraps around our planet and acts as an invisible force-field, providing long-term protection to life on the surface and modern technological infrastructure such as power-grids and satellites from harmful solar radiation. With the prospect of further extensive worldwide investment in satellite technology and space-based networks in the coming decades, being able to model and predict Earth’s magnetic field is scientifically and societally important.

Predicting the magnetic field is challenging, not the least because we do not yet have a complete representation that describes how the magnetic field changes. Although the basic physics of magnetic field generation in Earth’s liquid core is broadly understood, the complexities and extreme conditions within the Earth’s core make it computationally difficult to model. Of particular difficulty is that multiple processes: magnetic induction and diffusion, act together to produce magnetic field change; separating these is key to creating accurate models from data.

Aims and objectives

The novel aspect of this project is to apply recent advances in physics-informed machine learning to learn the different signatures of the various processes within the core, and then to apply this to the Earth. These methods will likely include physics-informed neural operators, which will be trained to partition maps of magnetic field change into contributions from the different physical processes. This will allow a new generation of models, which will not only help us better understand the dynamics of the core in the present day, but also lead to better short-term predictions of Earth’s magnetic field. One of the longer-term goals of the project is to assess the evidence for whether the geomagnetic field is likely to reverse and how fast it might do so.

Year 1:  Learn and experiment with physics-informed machine learning methods, likely including physics-informed neural operators, for signal separation using a simplified model of the geomagnetic field.

Year 2: Apply physics-informed machine learning methods to computational models of Earth’s core, where all processes are known explicitly. You will quantify how well the processes of induction and diffusion can be separated on this simulated but complex dataset.

Year 3: Apply this new methodology to the Earth, to produce not only new insights into the dynamics of the core but also a 5-year forecast of the geomagnetic field that will contribute to a magnetic field reference model developed by the global scientific community.

Applicant suitability

Applicants should demonstrate an interest in machine learning and understanding Earth’s mysterious magnetic field, although knowledge of either is not required and can be learned as the project evolves. The project would best suit applicants with a background in geophysics, physics, mathematics or related subject.

Training

You will be based within the Institute for Geophysics and Tectonics at the University of Leeds, and in addition to learning generic skills, you will receive project-specific training in geomagnetism and machine learning. By joining the deep-Earth research group at Leeds, one of the largest and most diverse internationally, and the cross-campus Scientific Machine Learning group, you will develop academic networks and peer-support. Throughout the project, you will work with scientists at the British Geological Survey (BGS), which includes the opportunity to spend up to three months working with the BGS in Edinburgh. You will also be encouraged to participate within a European-wide collaborative scientific team on the dynamics of Earth’s core.