Abstract The impact of model biases on subseasonal forecast skill has been evaluated in the European Centre for Medium‐range Weather Forecasts (ECMWF) subseasonal forecasting system using a relaxation experiment where the nudging was performed towards bias‐corrected integrations of the same model. The relaxation was performed regionally or globally but only for the very large scale. As expected, this methodology significantly reduces model biases over the areas where the nudging was applied. Remarkably, the experiments also display significantly improved forecast skill in predicting the anomalies at a lead time of three to four weeks, particularly over the northern extratropics, where subseasonal forecast skill has been stagnant at this time range over the past 20 years in the ECMWF operational subseasonal forecasting system. An analysis of a North‐Pole relaxation experiment shows that the polar relaxation impacts biases outside the region of nudging, such as the representation of the subtropical jet stream in the ECMWF model, reducing midlatitude geopotential height bias and leading to a slight improvement in Madden–Julian Oscillation (MJO) teleconnections over the northern extratropics. These results set a lower threshold for predictability at the subseasonal time‐scales and indicate that there is a large potential for improving dynamical subseasonal forecasting skill in weeks 3 and 4 in extratropical regions by improved treatment of model error.
Vitart et al. (Tue,) studied this question.