OWEMES 2006, Civitavecchia, Italy Accuracy of Short-Term Predictions for 25 GW Offshore Wind Power in Germany Jens Tambke, L. v. Bremen, N. Saleck, U. Graewe, *J.-O. Wolff ForWind and *ICBM, Carl von Ossietzky University Oldenburg John A. T. Bye Physical Oceanography, The University of Melbourne, Australia Lorenzo Claveri FMI, Finnish Meteorological Institute, Helsinki, Finland Carsten Poppinga, Ulrich Focken, Matthias Lange energy&meteo systems, Oldenburg, Germany © Elsam A/S

Jens Tambke, University of Oldenburg / Slide 2 Overview  Numerical Weather Predictions: ECMWF & DWD  Wind Power Forecasts at FiNO1, North Sea  Combination of Inputs for Power Forecasts  25 GW Forecasts: Spatial Smoothing of Errors

Jens Tambke, University of Oldenburg / Slide 3 Analysed Observations in North and Baltic Sea Horns Reef FINO1

Jens Tambke, University of Oldenburg / Slide 4 Wind Speed Measurements at FiNO1  Location: German Bight / North Sea  45km northerly of Island of Borkum  Investigated Period: Jan-Dec 2004  Heights: 33m to 103m

Jens Tambke, University of Oldenburg / Slide 5  DWD:  Lokal-Modell LM by German Weather Service (DWD)  Horizontal Resolution: 7 x 7 km 2  Vertical Resolution: 35 levels  Temporal Resolution: 1 h  Horizon of Prediction: 48 h Evaluated Forecast Models at FiNO1  ECMWF:  European Center for Medium Range Weather Forecasts  Horizontal Resolution: 40 x 40 km 2  60 levels  3 h

Jens Tambke, University of Oldenburg / Slide 6 Typical ECMWF Wind Forecast at FiNO1, North Sea 48h Measured Wind Speed Predicted

Jens Tambke, University of Oldenburg / Slide 7 RMSE of Speed Forecasts at a single offshore site: FiNO1, 100m height, 12 months, 2004 DWDECMWF RMSE normalised to the mean wind speed (9.8 m/s): 15% to 30% 15% to 25%

Jens Tambke, University of Oldenburg / Slide 8 Accuracy of Speed Forecasts at FiNO1, 100m height, 12 months, 2004 Distribution of Speed Errors : not exactly Gaussian…

Jens Tambke, University of Oldenburg / Slide 9 Calculation of Power Forecasts  Here: typical Multi-Mega-Watt Power Curve  „Band Pass Filtering“ of the Time Series +/- 1m/s +/- 10% Power

Jens Tambke, University of Oldenburg / Slide 10 RMSE of Power Forecasts at FiNO1, 100m Height, 12 months, 2004 RMSE % of P(inst) DWD: 15% - 26% ECMWF: 14% - 22% Combination: 12% - 22%

Jens Tambke, University of Oldenburg / Slide 11 Correlation of Power Forecasts to Obs. at FiNO1, 100m height, 12 months, 2004 DWD: ECMWF: Combined:

Jens Tambke, University of Oldenburg / Slide 12 Accuracy of Power Forecasts at FiNO1, 100m height, 12 months, 2004 Distribution of Power Errors: Not at all Gaussian!

Jens Tambke, University of Oldenburg / Slide 13 Temporal Speed Gradients at FiNO1, 100m height, 12 months, /- 1m/s +/- 10% Power Interpolated ECMWF winds are too persistent Wind Speed Difference after 1 Hour

Jens Tambke, University of Oldenburg / Slide 14 Temporal Power Gradients at FiNO1, 100m height, 12 months, 2004 Interpolated ECMWF power is too persistent Power Difference after 1 Hour

Jens Tambke, University of Oldenburg / Slide 15 Power Forecasts for 25 GW Offshore Wind Farms, 12 months, 2004 Planned Wind Farms in the German Bight Source: 180km 0 0,2 0,4 0,6 0, diameter of region [km] Error Reduction Factor km Offshore Measurements for Forecast-Evaluation? DWD Wind Speed Analysis proved to be a good substitute! Spatial smoothing of errors same size as onshore? Red.-Factor = RMSE(single) RMSE(region)

Jens Tambke, University of Oldenburg / Slide 16 Irish SeaNorth SeaBaltic Sea 110m Wind Speed Analysis from Weather Service DWD March 20th, 2004

Jens Tambke, University of Oldenburg / Slide 17 Correlation of Power Forecasts at 25 single Sites in the German Bight DWD-AnalysisECMWF-Forecast agrees well with Obs.Decay too weak Decay weaker than Onshore!

Jens Tambke, University of Oldenburg / Slide 18 Distance of single sites Cross correlation of forecast errors at single sites Decay has same size as onshore 0 km180 km ECMWF-Forecasts evaluated against DWD-Analysis, 12 months, 2004 Correlation of Errors of Power Forecasts at 25 single Sites in the German Bight

Jens Tambke, University of Oldenburg / Slide 19 regional 25GW forecast 9% 13% Single Offshore Sites 12% => 3 GW RMSE % of P(inst) 18% 22% Offshore (180km) Error Reduction Factor (3h: 0.65) ~ 0.73 (48h: 0.82) RMSE of Forecasts for 25 GW Offshore Wind Power, 12 months, 2004

Jens Tambke, University of Oldenburg / Slide 20 Offshore 25GW 13% Single Offshore Sites 18% 22% 7% => 3.5 GW RMSE % of P(inst) 9% On+Offshore 50GW 5% 10% Error Reduction Factor: Offshore (180km) ~ 0.73 On+Offshore (800km) ~ 0.43 RMSE of Forecasts for 25 GW Offshore Wind Power, 12 months, 2004

Jens Tambke, University of Oldenburg / Slide 21 Conclusions  Combinations of Forecasts from DWD and ECMWF increase the Forecast Accuracy  Regional Smoothing of Errors: Same Size as Onshore  Offshore Forecasts (180km area): RMSE of 12%  On&Offshore Forecasts (800km region):RMSE of 7% Thank you for your Attention! This work was funded by the EU within the ANEMOS and POW’WOW Projects.

Jens Tambke, University of Oldenburg / Slide 22 Accuracy of Power Forecasts at FiNO1, 100m height, 12 months, 2004 Error Frequency DWD Abs(Error) < 15% at 36hours: 230 of 350 forecast runs ECMWF Abs(Error) < 15% at 36hours: 262 of 350 forecast runs

Jens Tambke, University of Oldenburg / Slide 23 Power Forecasts for 25 single Offshore Sites, 12 months, 2004 Increase with forecast time due to increasing importance of general upstream errors Cross correlation of forecast errors at single sites Look-ahead Time

Jens Tambke, University of Oldenburg / Slide 24 Correlation of Forecasts to Reference for 25 GW Offshore Wind Power, 12 months, 2004 On+Offshore 50GW: 0.98 – 0.91 Offshore 25 GW: 0.97 – 0.88 Single Offshore: 0.94 – 0.85

Jens Tambke, University of Oldenburg / Slide 25 Temporal Gradients of Power Forecasts in the German Bight Power Difference after 1 Hour

Jens Tambke, University of Oldenburg / Slide 26 Temporal Gradients of Power Forecasts in the German Bight Power Difference after 1 Hour