Short-term polar motion forecasts from earth system modeling data

Polar motion predictions for up to 10 days into the future are obtained from predicted states of the atmosphere, ocean and continental hydrosphere in a hind-cast experiment covering 2003-2008. High-frequency mass variations within the geophysical fluids are the main cause of wide-band stochastic signals not considered in the presently used statistical prediction approach of IERS bulletin A for polar motion. Taking EAM functions based on forecasted model states, derived from ECMWF medium-range forecasts and corresponding LSDM and OMCT simulations, into account the prediction errors are reduced by 26%. The effective forecast length of the model combination is found to be 7 days, primarily limited by the accuracy of the forecasted atmospheric wind fields. Highest improvements are found for forecast days 4-5 with prediction skill scores of the polar motion excitation functions improved by a factor up to 5. Whereas bulletin A forecasts can explain the observed variance within the first 10 days only by up to 40%, half of the model forecasts reach relative explained variances between 40 and 80%.