We examine the variation in boreal winter surface wind speed predictive skill between different Madden-Julian Oscillation (MJO) phases as well as the potential predictability using 20 years of ECMWF ensemble reforecasts. In examining the extended-range predictability we examine the sensitivity of predictive skill and potential predictability to averaging forecasts in time (1, 2, 4, 7, and 14 day windows are used) and spatial averaging (500 and 1000 km radial averages are used). Spatial averaging of the forecasted and observed wind speeds increases skill especially for window lengths less than 7 days. We suspect this is due to the increase in spatial correlation with longer windows. In the ECMWF ensemble reforecasts the day 4-8 and week 2 forecasts have correlations of at least 0.5 in the sub-tropics and lower portions of the middle-latitudes. For the week 2-4 forecasts skill is limited to the equatorial regions. Using the ECMWF forecasts we estimate that the potential predictability of the week 2 forecasts is a correlation of at least 0.7 in the sub-tropics and lower middle-latitudes. The MJO phase during the initial conditions has a slight impact on the regional variability of the extended-range prediction skill. In addition we analyze the 2020-21 through 2022-23 winters in the Navy ESPC ensemble and operational ensembles from the S2S database. We find evidence suggesting that certain aspects of the large-scale environment are predictable 2-3 weeks in advance in Navy ESPC in the case of the US West Coast wind event that occurred in early March 2023.

