A Hierarchical Bayesian Model for the Global Holocene Geomagnetic Field

Abstract We present a hierarchical Bayesian model of the global geomagnetic field for the last 8,000 years. The model is built solely with thermoremanent records, which include archeomagnetic and volcanic data. Building upon previous work, this model includes fewer approximations and treats model hyperparameters as probabilistic variables. A Student's ‐distribution is used to account for data uncertainties. Hamiltonian Monte‐Carlo sampling is applied to sample from the non‐linear posterior distribution. The posterior values for the hyperparameters correspond to a reduction in prior variances compared to earlier models. The recovered global statistics align with those derived from modern satellite and observatory data, indicating consistent properties of the geomagnetic field. The model provides a reconstruction of the Holocene geomagnetic field, offering insights into past field behavior, while acknowledging the limitations imposed by data scarcity. When compared to previous global reconstructions in terms of dipole moment and local predictions, the new model shows similar behavior for the Common Era, with slight differences and reduced variability for earlier times. The analysis of the recurrence of low‐intensity patches shows weak anomalies at the expected locations, but not a simple westward drifting. However, these become less well resolved further back in time.