We have developed a probabilistic approach to geomagnetic field modelling, taking prior information in the form of training images, and using direct sequential simulation on the source surface.
Below is an example, based on an ensemble of images of the core surface field from a dynamo simulation (Aubert et al., 2013) and using Swarm magnetic field data as input. The figures below show the estimates posterior mean core surface field, the associated posterior standard deviation, example realizations of core surface fields compatible with the statistics of the training images and the data, the fit to the satellite data, the reproduction of the histogram of core surface field values and the prior and posterior power spectra of the core surface field compared to the CHAOS-7 model. The posterior mean contains power up to spherical harmonic degree 15 compatible with the dynamo models and the data, in comparison to conventional images of the core field which are truncated at spherical harmonic degree 13.
