The Earth's magnetic field has undergone temporal and spatial variations including polarity reversals. Paleomagnetic and historical magnetic field measurements suggest persistent distinct patterns of variation of the geomagnetic field taking place in different regions of the Earth. Global models of the Earth's magnetic field derived from geomagnetic satellite and observatory data, but also from historical and archeomagnetic data provide unique insights to the dynamical processes in the liquid outer core and the interactions between core and mantle. These processes evolve on different time scales. At least on the millennia time scale variation patterns may be explained by the thermal influence of Earth's mantle on the geodynamo. Whether this influence can already be discerned on the century time scale is still debated. We propose to continue our studies to model and to predict the Earth's magnetic field variation. In the prior project we have set up a scheme for predicting these changes. Yet, conclusive results are not obtained. The main reason for this are peculiar long-term trends related to the convective overturn in the outer core, but also of unknown origin. The presence of such long-term trends in the time series from which stochastic models (autoregressive models) of the temporal variability have been derived leads to a discrepancy between prediction and observation. Therefore, we will place great emphases to uncover the origin of such long-term trends and derive geomagnetic field models covering the past centuries. The new field models should provide a better description than previous models, as we plan to apply new techniques and greatly improved data sets. The extended field model will almost cover the convective overturn time scale of the liquid outer core and therefore should account for most of the observed non-stationarity (caused by the long-term trends) in the secular variation. This in turn will finally lead to refined results of the forecasting scheme. These studies should also include analyzes of numerical dynamo simulations where we want to quantify the influence of the lower mantle and the existence of stably stratified layer in the outer core on the long-term temporal variability of the Earth's magnetic field. Further, we seek to gain some understanding of the causes of the chaotic short-term secular variation. We believe that a consolidation of our knowledge of Earth's magnetic field variation can only be facilitated by a joint analysis of geomagnetic field observations covering the last few centuries and numerical simulation of the geodynamo.