Reconstructing the geomagnetic field over the observatory era with a Bayesian approach

More than one decade of global coverage geomagnetic data is now available from low-orbiting satellites. It has brought unprecedented constraints about the time changes of the Earth's magnetic field, enhancing the presence of interannual variations. Underlying physical processes must be investigated over longer series, such as that provided by ground observatories for the past century. However, all field models have been up to now based on regularization techniques that prevent the occurrence of rapid changes when less data are available. This process can result in biases when inverting sequentially core flow models from geomagnetic field models. Here we depart from the regularization process to reconstruct the field evolution over the observatory era. We deduce the prior covariance properties (variance and time correlation function as a function of harmonic degree) of the field model from the statistics of satellite field models. The posterior covariance matrix is then used to build an ensemble of field models that both fit geomagnetic data and satisfy the required covariance properties.