A case for variational geomagnetic data assimilation: insights from a one-dimensional, nonlinear, and sparsely observed MHD system

Secular variations of the geomagnetic field have been measured with a continuously improving accuracy during the last few hundred years, culminating nowadays with satellite data. It is however well known that the dynamics of the magnetic field is linked to that of the velocity field in the core and any attempt to model secular variations will involve a coupled dynamical system for magnetic field and core velocity. Unfortunately, there is no direct observation of the velocity. Independently of the exact nature of the above-mentioned coupled system – some version being currently under construction – the question is debated in this paper whether good knowledge of the magnetic field can be translated into good knowledge of core dynamics. Furthermore, what will be the impact of the most recent and precise geomagnetic data on our knowledge of the geomagnetic field of the past and future? These questions are cast into the language of variational data assimilation, while the dynamical system considered in this paper consists in a set of two oversimplified one-dimensional equations for magnetic and velocity fields. This toy model retains important features inherited from the induction and Navier-Stokes equations: non-linear magnetic and momentum terms are present and its linear response to small disturbances contains Alfv'en waves. It is concluded that variational data assimilation is indeed appropriate in principle, even though the velocity field remains hidden at all times; it allows us to recover the entire evolution of both fields from partial and irregularly distributed information on the magnetic field. This work constitutes a first step on the way toward the reassimilation of historical geomagnetic data and geomagnetic forecast.

💡 Research Summary

The paper investigates whether variational data assimilation (VDA) can translate accurate magnetic‑field observations into reliable estimates of the Earth’s core dynamics, despite the complete lack of direct velocity measurements. To address this, the authors construct a highly simplified one‑dimensional magnetohydrodynamic (MHD) system that retains the essential physics of the induction equation and the Navier–Stokes equation. The model consists of two coupled partial differential equations: a magnetic‑field equation that includes a nonlinear advection term involving the velocity, and a momentum equation that contains the Lorentz force, viscous diffusion, and a nonlinear coupling to the magnetic field. Linearising the system reveals the presence of Alfvén waves, indicating that magnetic and velocity perturbations propagate together at a characteristic speed, thereby providing a natural dynamical link that can be exploited by data assimilation.

In the VDA framework, the state vector comprises both magnetic (B) and velocity (u) fields, while the observation operator maps only the magnetic component to the data space. The cost function combines a background term (penalising deviation from an a‑priori guess) and an observation term (penalising misfit to the measured magnetic field). The gradient of the cost function is obtained via an adjoint model derived from the toy MHD equations, and optimisation is performed with a limited‑memory BFGS algorithm. Crucially, the authors deliberately impose sparse and irregular magnetic observations, mimicking the real situation where satellite data are dense only in recent decades and historic ground‑based records are sparse and unevenly distributed.

Numerical experiments demonstrate three key findings. First, when the assimilation window spans several Alfvén times, the VDA system can reconstruct the full time‑evolution of both B and u, even in intervals where no magnetic data are available. This success hinges on the strong dynamical coupling: magnetic variations observed at the surface carry indirect information about the hidden velocity field, which the model propagates forward and backward in time. Second, the quality of the reconstruction degrades gracefully as observation noise increases or the number of observation points decreases; appropriate tuning of background error covariances and regularisation parameters mitigates these effects. Third, the inclusion of modern high‑resolution satellite magnetic data dramatically improves the ability to re‑analyse past geomagnetic records and to produce forward forecasts. By assimilating recent precise measurements together with older, coarser observations, the method yields a continuous, dynamically consistent history of the geomagnetic field and an informed prediction of its future evolution.

The authors acknowledge that the one‑dimensional toy model is a drastic simplification of the true three‑dimensional, spherical core dynamics, and that realistic error models for both observations and background states remain to be developed. Nevertheless, the study provides a proof‑of‑concept that variational data assimilation can, in principle, recover hidden velocity information from magnetic observations alone, thanks to the intrinsic MHD coupling. The paper concludes by outlining the next steps: extending the approach to full three‑dimensional geodynamo simulations, integrating real satellite and ground‑based magnetic datasets, and refining error covariance specifications. If successful, this line of research could revolutionise geomagnetic data re‑analysis, improve our understanding of core dynamics, and enable reliable geomagnetic forecasts for the coming decades.