Validation of boundary-layer winds from WRF Mesoscale Forecasts with applications to wind energy forecasting

Especially in Denmark, with a wind power share in the electricity supply of more than 20%, wind forecasts are important for both, wind farm operators and Transmission System Operators. Therefore, the Danish National Laboratory for Sustainable Energy (Risø DTU) carries out research with the aim of improving wind forecasts with regard to wind energy production and wind resource assessment.

A real-time weather forecasting system for Denmark based on the Advanced Research WRF mesoscale model is run in a operational mode to understand the model's deficiencies, its physical limitations and the forecast error behavior under certain weather situations. Short-term wind forecasts (0-48 hours) for the period May to October 2009 are verified against hourly winds from 10-meter conventional METeorological Aerodrome Report (METAR) measurements and wind measurements from the 160-meter meteorological mast located at the Risø National Test Station for Large Wind Turbines at Høvsøre, Denmark. These simulations use the Yonsei University PBL (YSU PBL) parameterization.

The verification of the 10-meter wind speed forecasts against METAR observations shows that winds are generally overestimated during the nighttime hours over land. The magnitiude of the mean errors is within 0.5 m/s for most land stations during daytime. When classifying the statistics according to the observed wind speeds, lower wind speeds are overestimated while underestimated at higher wind velocities. This also corresponds to previous results, where a poor man's ensemble was created by using different forecast runs for the same forecast times and compared with 60m wind measurements.

Wind turbines currently being installed have hub heights at or above 80 meters. Therefore, accurate prediction of the vertical structure of the wind profile is very important. We thus verified the WRF-simulated mean wind profiles at Høvsøre at heights of 10m, 40m, 60m, 80m, 100m, 116m and 160m, when the wind direction is between 30 and 125 degrees, where the terrain is flat and the flow homogeneous, not affected by the wakes of the wind turbines or land-sea discontinuities. The results reinforce the conclusions from the METAR validation: The forecasted mean wind speed is higher than the observations for the first levels up to 100m and it is lower upwards. The model-simulated wind profiles often do not deviate strongly from the logarithmic wind profile which indicates neutral conditions, while observations show stable conditions. Further analyses show that that the agreement between forecasted and observed wind profiles depends on seasonality, stability and extreme conditions.

To better understand these results, the performance of seven boundary layer parametrisations (Yonsei University, Mellor-Yamada-Janjic, MRF, ACM2, Quasi-Normal Scale Elimination, Mellor-Yamada Nakanishi and Niino Level 2.5, Mellor-Yamada Nakanishi and Niino Level 3 scheme) was evaluated. These simulations will aid in selecting the best parametrisations for our wind power forecasts and to find out which parametrisations can best capture the profiles of wind and temperature at hub height. We ran the WRF ARW for 5 consecutive days in the summer of 2009. Each run had a leadtime of 30 hours. The period was characterised by a high pressure system over Scandinavia, thus by dry conditions, and the wind was coming from the Northeast. No internal boundary layers and no impact of deep convection could therefore be encountered. Comparisons of the wind and temperature profiles for a grid point next to the above mentioned test site Høvsøre from that station indicate as well, that in very stable conditions WRF very often tends to “neutralize” the wind profile. In general the different schemes behave very similar. WRF has difficulties simulating the transition from stable/unstable boundary layers right in time and magnitude. It will be shown further, that the general wind situation is similarly captured by the different schemes, but wind events, like sudden changes in wind speed, were not caught in amplitude by any of the schemes.

The results of the validation of wind forecasts over Denmark show that errors in the forecasts are mainly related to two principal problems. First, because of its relatively low horizontal resolution, the 10-meter wind field simulated by the mesoscale model is not capable of resolving the influence of important topographical features in the vicinity of the measurement site. This problem can be partly overcomed by correcting or adjusting the WRF-simulated winds to the local site by the use of a microscale model. Second, the wind profiles and boundary layer structure simulated by WRF is often neutrally stratified even at night, when the observed boundary layer is stable. Below 30-100 meters WRF winds are overestimated, above this level underestimated.