Back to search

University of Leeds

Flows and Forecasts: Investigating the Geomagnetic Field using Machine Learning

Abstract

dc:description.abstract

The Earth’s magnetic field has wide ranging impacts on navigation, satellite operations and protection from space weather, making it important to understand its generation and forecast its evolution. The dominant surface field is generated by electrically conducting fluid motion in the outer core through the geodynamo. Over short timescales, these field changes can be approximated as advection-dominated, relating geomagnetic secular variation to core-surface flow. However, this inversion is underdetermined and requires physical constraints. Typical approaches to core flow inversions are to use global spherical harmonic representations, which can couple separated regions and smooth localised features. Regional inversions avoid this coupling, allowing a test of whether inferred structures are supported by local magnetic-field information. Geomagnetic behaviour may also be influenced by heterogeneous heat flux at the core–mantle boundary (CMB). Geodynamo simulations suggest that this forcing affects magnetic-field morphology although its structure and amplitude remain poorly constrained. This thesis describes the development of novel machine-learning approaches for investigating core-surface flow and CMB heat-flux heterogeneity. First, Physics-Informed Neural Networks (PINNs) are used to infer regional flows that fit local secular variation while satisfying a physical constraint, which are used to produce a mosaic of independent regional flows that recovers the planetary gyre. The PINN framework is then used to construct mesh-free regional and global flow models for geomagnetic forecasting. Hindcasts for 2020–2025 show that the regional approach produces lower root-mean-square misfits relative to CHAOS-7.18 than the other IGRF13 candidate models. The method is subsequently used to produce a candidate secular-variation forecast for IGRF-14 over 2025–2030. Finally, a convolutional autoencoder classifier is developed to infer CMB heat-flux heterogeneity from the radial magnetic field. The results show that this heterogeneity leaves classifiable information in the field, but further development is required before the method can reliably constrain the Earth’s CMB heat-flux structure.

Degree

thesis:*
Name dc:type.qualificationname
Ph.D
Level dc:type.qualificationlevel
doctoral
Grantor dc:publisher.institution
University of Leeds
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Shakespeare-Rees, Naomi Mara
Advisors dc:contributor.advisor
  • Livermore, Philip
  • Davies, Christopher
  • Rogers, Hannah
  • Beggan, Ciarán
  • Brown, William
  • Finlay, Christopher
ORCID person:identifier.orcid

Rights

Language dc:language.dcterms_iso639_2
eng

Identifiers

dc:identifier.*
Dc Identifier Grantnumber
NE/S007458/1
GA/22 S/011

Chain of custody

source
Harvested from
White Rose University Consortium
Base URL
etheses.whiterose.ac.uk/cgi/oai2
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
related terms
citation

Shakespeare-Rees, Naomi Mara. Flows and Forecasts: Investigating the Geomagnetic Field using Machine Learning. doctoral thesis, University of Leeds, 2026.