Table 1. Overview of the three MPI-ESM ensemble generations used in this study, including information on the name within the MiKlip consortium, the ocean initialisation, the atmosphere initialisation and the number of ensemble members.
baseline1Anomalies from ORA-S4 reanalysisFull-fields from ERA-Interim/ERA4010prototype1Full-fields from ORA-S4 reanalysisAs above10 (from 15)prototype2Full-fields from GECCO2 reanalysisAs above10 (from 15)historical––10 (from 15)
Fig. 1
(a) Topography of Europe in metre, and grid points for CWT analysis (step 1 of SDD). The red point represents the central point at 10°E, 50°N (near Frankfurt, Germany), and the red crosses represent the surrounding 16 grid points used for the computation of the CWTs. The white box represents the region for figures (b) to (d). (b) Climatological mean of mean 10 m wind speed in metre per second for ERA-Interim (1979–2010) as obtained by SDD. (c) Climatological mean of annual Eout in 103 MWh for ERA-Interim (1979–2010) as obtained by SDD. (d) Explained variance between annual Eout time series for ERA-Interim (1979–2010) as obtained by SDD and as obtained by DD (DDera) per CCLM grid point. Grid points with significant correlation are dotted (t-test, 95% confidence level). Box 1 represents the subregion for the computation of the MSE skill scores as shown in Fig. 5 and Supplementary Figs. 2–5 (see also Section 4), and box 2 represents the subregion for the averages over Germany (7°E–14°E, 48°N–53°N) as shown in Figs. 2, 3, 4, Figs. 6, 7 and Supplementary Fig. 1.
Fig. 2
(a) Ranked probability skill scores (RPSSs) for large-scale MPI-ESM mean wind (blue) and SDD-simulated regional mean wind (red) for seven different lead times for the whole year, averaged over Germany (box 2 in Fig. 1d), for the baseline1 ensemble. (b) Reliability for large-scale MPI-ESM mean wind (blue) and SDD-simulated regional mean wind (red) for seven different lead times for the whole year, averaged over Germany (box 2 in Fig. 1d), for the baseline1 ensemble. (c)–(d) as (a)–(b), but for the prototype1 ensemble. (e)–(f) as (a)–(b), but for the prototype2 ensemble.
Fig. 3
RPSSs for SDD-simulated wind speed for seven different lead times for the whole year, averaged over Germany (box 2 in Fig. 1d), for the ensemble generations baseline1 (blue), prototype1 (red) and prototype2 (yellow), for different percentiles: (a) mean wind, (b) 75th percentile and (c) 90th percentile.
Fig. 4
Forecast skill scores for SDD-simulated Eout for seven different lead times for the whole year, averaged over Germany (box 2 in Fig. 1d), for the ensemble generations baseline1 (blue), prototype1 (red) and prototype2 (yellow). (a) RPSSs and (b) mean square error skill scores (MSESSs).
Fig. 5
MSESSs for SDD-simulated Eout for four exemplary lead times for the whole year for the ensemble generations baseline1 (left column), prototype1 (middle column) and prototype2 (right column). Reference forecast is the ensemble mean of the uninitialised historical runs.
Fig. 6
MSESSs (left column) and RPSSs (right column) for SDD-simulated Eout for seven different lead times for the four seasons, averaged over Germany (box 2 in Fig. 1d), for the ensemble generations baseline1 (blue), prototype1 (red) and prototype2 (yellow). (a) and (b) Winter (DJF), (c) and (d) spring (MAM), (e) and (f) summer (JJA), (g) and (h) autumn (SON). MSESS values under −1.0 are displayed in the corresponding bar. Note that yr1 for winter corresponds to months 12–14 after initialisation.
Fig. 7
(a) Time series of annual frequency-anomalies for CWT W+ in % (black line) and of annual Eout anomalies in 103 MWh (red line) for the ERA-Interim period 1979–2010. The correlation between both time series is given in the upper left corner. (b) MSESSs for CWT W+ for seven different lead times for the whole year for the MPI-ESM ensemble generations baseline1 (blue), prototype1 (red) and prototype2 (yellow). For details, see main text (Section 4.3).