Exploring sample size issues for 6-10 day forecasts using ECMWF’s reforecast data set
Enhancing probabilistic Wind Power Forecasting using ECMWF’s 100m EPS winds
-
Upload
orlando-wolf -
Category
Documents
-
view
31 -
download
1
description
Transcript of Enhancing probabilistic Wind Power Forecasting using ECMWF’s 100m EPS winds
Enhancing probabilistic Wind PowerForecasting using ECMWF’s 100m EPS winds
Lueder von Bremen, Constantin Junk, Detlev Heinemann
ForWindCenter for Wind Energy ResearchUniversity of Oldenburg, Germany
EWEA 2012, Copenhagen, 17 April 2012
EWEA 2012, Copenhagen, 17 April 2012 - Lueder von Bremen
Meteorological Ensembles for Wind Power Forecasting
Spatial distribution of forecast uncertainty
Ensemble Prediction System with 100m winds for improved Wind Power Forecasting
Summary
Outline
EWEA 2012, Copenhagen, 17 April 2012 - Lueder von Bremen
Why weather forecasts can not be precise?
Lorenz Paradigm: Numerical Weather Forecasting is an initial state problemQuantify the uncertainty in the forecast to be aware of forecast errors
dx / dt = a (y - x) dy / dt = x (b - z) - y dz / dt = xy - c z (E. Lorenz, 1963)
x
x
?
?
EWEA 2012, Copenhagen, 17 April 2012 - Lueder von Bremen
Ensemble Prediction System (EPS) at ECMWF
Lorenz Paradigm: Numerical Weather Forecasting is an initial state problemQuantify the uncertainty in the forecast to be aware of forecast errors
EWEA 2012, Copenhagen, 17 April 2012 - Lueder von Bremen
Powergrams: Communicating Ensemble Forecasts
measurementensemble meandeterministic forecastsimulated wind power
Example: 1 March 2010 extreme event Storm Xynthia in Eastern Germany (50Hertz, ~10.7 GW)
+96h +48h
New: What is the distribution of uncertainty in the control zone
EWEA 2012, Copenhagen, 17 April 2012 - Lueder von Bremen
Example: Storm Xynthia (50 Hertz)1 March 2010, 0UTC
Ensemble Mean
Forecast Uncertainty(RMS Ensemble Spread)
+96h +48h
EWEA 2012, Copenhagen, 17 April 2012 - Lueder von Bremen
Forecast uncertainty(RMS Ensemble Spread)
+96h +48h
absolute forecast error
Example: Storm Xynthia (50 Hertz)1 March 2010, 0UTC
EWEA 2012, Copenhagen, 17 April 2012 - Lueder von Bremen
Evaluation of storm Xynthia (50 Hertz)1 March 2010, 0UTC
+96h forecast: crosses+48h forecast: bullets
The level of uncertainty can be used as indicator of a (likely) forecast error
uncertainty [MW]
EWEA 2012, Copenhagen, 17 April 2012 - Lueder von Bremen
Overcome the usage of 10m winds Isn‘t that old? New Ensemble System: 100m winds instead of 10m winds from 26 Jan 2010 at ECMWF ECMWF analysis speeds are weighted according to spatial wind power distribution „Regional Power Curve“ for 50Hertz (VET)
10m winds hubheight100m winds hubheight
Feb 2010-Apr 2011
EWEA 2012, Copenhagen, 17 April 2012 - Lueder von Bremen
Performance of the Ensemble System compared to a single forecast (here: Germany)
at D+2 ensemble mean starts to be better than single forecast New system is improved up to 30%
Single forecast
Ensemble Mean
new system with 100m windsold system
-30%
EWEA 2012, Copenhagen, 17 April 2012 - Lueder von Bremen
Can we trust the probabilistic forecast?
New system has much better reliability
Evaluation of reliability of the Ensemble Prediction System here: Event wind power>9GW in Germany +72h forecast horizon
new System with 100m windsold System:overconfident
EWEA 2012, Copenhagen, 17 April 2012 - Lueder von Bremen
Improvement of probabilistic forecast
Germany
50Hertz
Very strong (25%) improvement of probabilistic forecast skill
Continuous Rank Probability Skill Score: CRPSS=1-CRPSnew/CRPSold
EWEA 2012, Copenhagen, 17 April 2012 - Lueder von Bremen
Summary
Forecast uncertainty can be derived from meteorological Ensembles and is related to the occurence of fc errors
Geographical display of forecast uncertainty
at Day+2 Ensemble mean starts to be better than single fc
Probabilistic forecast score is improved by 10-25% using the new Ensemble Prediction system
Thank you for your attention
Contact: [email protected]
This work was supported by the Federal Ministry
for Science and Culture of Lower Saxony and the European
Commission in the SafeWind Project (Grant No. 213740).