The OSF pilot is based on subsampling a seasonal forecast ensemble with observational data. We use an ensemble initiliazed two months before the start of the season and choose the twenty-five members which most closely resembled the observations of precipitation in the second month of the model run, i.e. for the winter (DJF) prediction, we take an ensemble initialized in September and subsample it with October's observations. The verification was conducted using a hindcast ranging from 1993 to 2016, the maximum common period for all models.

The following seasonal forecasting models are included in the ensemble:

• European Centre for Medium-Range Weather Forecasts (ECMWF) SEAS5 system 51
• Météo-France system 8
• Centro Euro-Mediterraneo sui Cambiamenti Climatici (CMCC) SPS3.5
• UK Met Office GloSea6 system 603
• National Center for Environmental Prediction (NCEP) system 2
• Deutscher Wetterdienst (DWD) GCFS2.1
• Environment and Climate Change Canada (ECCC) CanSIPSv3.0 systems 4 and 5

This method can offer local increases in skill, but it does not improve the results in the whole Mediterranean region. Skill values depend on the season, as certain times of the year may offer windows of opportunity for seasonal forecasts.

 TEMPERATURE

PRECIPITATION