Modeling the Crustal Magnetic Field of Mars With Physics-Informed Neural Networks
OPEN ACCESS
Loading...
Author / Producer
Date
2025-11
Publication Type
Journal Article
ETH Bibliography
yes
OPEN ACCESS
Data
Rights / License
Abstract
Satellites orbiting Mars have measured crustal magnetic fields up to two orders of magnitude stronger than those on Earth. Although Mars currently lacks an active global magnetic field, this magnetization preserves a valuable record of the planet's early dynamo activity and crustal evolution. We present a high-resolution model of the crustal magnetic field of Mars, using all currently available magnetic field data collected by the MGS and MAVEN spacecraft (up to 02/2025), combined with a novel modeling approach. Notably, we incorporate recent low-altitude MAVEN data which contain short wavelength signals that were not available for previous models. We show that neural networks trained from spacecraft data can accurately predict the magnetic field at any location around Mars at orbital altitudes. These physics-informed networks use the equations of magnetostatics to enforce the conservative and solenoidal nature of the field, and are enhanced with bagging to mitigate the effect of noise in the data. Using this ensemble approach, we provide an estimate of uncertainties associated with our predictions. To demonstrate the performance of this method, we benchmark it against previous models using the same input and validation data subsets. Our model achieves an unprecedented resolution of spherical harmonics degree 139, corresponding to a spatial resolution of 153 km at the surface. Using our model to investigate small scale magnetic field signatures, we find that magnetic fields over ancient paleolakes are significantly stronger than other surface features or geological units, suggesting that serpentinization may have played a key role in magnetizing the crust.
Permanent link
Publication status
published
External links
Editor
Book title
Journal / series
Volume
130 (11)
Pages / Article No.
Publisher
American Geophysical Union
Event
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
Organisational unit
Notes
Funding
209123 - Using Magnetic Fields to Explore Terrestrial Bodies in our Solar System (SNF)