|Publication Type||Technical memorandum|
|Year of Publication||2014|
|Secondary Title||ECMWF Technical Memorandum|
|Authors||Vitart, F, Balsamo, G, Buizza, R, Ferranti, L, Keeley, S, Magnusson, L, Molteni, F, Weisheimer, A|
ECMWF has produced 32-day forecasts routinely since March 2002 to fill the gap between medium-range and seasonal forecasts. The skill of the sub-seasonal forecasts at ECMWF has improved significantly over the past decade. This improvement can be linked to improved skill to predict the Madden Julian Oscillation (MJO) with an average gain of about 1 day of prediction skill per year, and improved tropical-extra-tropical teleconnections associated to the MJO. However, the MJO teleconnections in the Euro-Atlantic sector are still too weak compared to re-analysis. The skill of the ECMWF monthly forecasts to predict the North Atlantic Oscillation (NAO) and sudden stratospheric warmings (SSW), another source of sub-seasonal predictability, has also improved over the past 10 years, although the downward propagations associated to SSWs is much weaker in the model than in ERA Interim. Further improvements in the ECMWF sub-seasonal forecasts are expected in the coming years with, among other changes, the introduction of a dynamic sea-ice model, increased atmospheric and oceanic resolutions and the extension of the re-forecasts. Results from the MINERVA experiment suggest a modest improvement in sub-seasonal skill scores when increasing the horizontal resolution of the atmospheric model from T319 to T639 and T1279. In particular, the NAO forecast skill scores improve when increasing the horizontal resolution from T319 to T639. The extension of the re-forecasts will lead to a more accurate estimation of the forecast anomalies and probabilities. Preliminary results of monthly forecasts with an active sea-ice model show skill in predicting the evolution of sea ice in the sub-seasonal time-range. These past and future improvements in the monthly forecasts should make it possible now to extend the monthly forecasts to 45 or 60 days and produce skilful and reliable forecasts beyond day 30.