Variations in the geomagnetic field provide the only window into the dynamic iron core of planet Earth. We propose to develop a data assimilation framework that combines observations and numerical dynamo simulations in order to yield a realistic model of the Earth’s core dynamics. While data assimilation is routinely used in weather prediction, its application to geomagnetism is still in its infancy. The particular challenge is the fact that geomagnetic data are measured far away from the core dynamo and that relatively few data have to constrain a complex process. In our method of choice, the Ensemble Kalman Filter, the link between observations and core dynamics is based on the statistical correlations from several numerical dynamo simulations running in parallel. The main objectives are a) to provide an understanding of the essential ingredients in geodynamo simulations, b) to develop a model for the whole core dynamics from 1900 to today, and c) to use this model for forecasting the geomagnetic field evolution over several decades. Extensive numerical studies will serve to identify the best compromise between different assimilation techniques, dynamo model complexity, ensemble size, and computational costs. In collaboration with other SPP teams, this work will help to establish an up-to-date and integrated view of the Earth’s diverse magnetism from magnetospheric current systems to the dynamo process operating in our planet’s deep core.