This talk will present the predictability research carried out at ECMWF targeting the subseasonal and seasonal time scales. The aim of our work is to explore relevant directions to improve the skill of the ECMWF forecasting systems. This involves both exploring the predictability horizon of the earth system, as well as identifying those elements limiting the actual forecast skill. The aim is to guide future development of the ECMWF Seamless Earth-System forecasting system. 

The predictability research at these time scales encompasses a large number of challenging areas, such as detection and attribution of signals, measure of skill, separation of forecast errors from unpredictable events, relation between systematic model error with both forecast skill and physical processes. There is a strong need for data analysis approaches combining advanced statistics with dynamical constrains. The wide rage of problems can only be tackled with a strong engagement from the wider academic community.