Performance of Tuned Versus Untuned Wind Energy Forecasts

The Global Weather Corporation (GWC) produces forecasts for wind energy by combining hub-height windspeed forecasts from various weather models and tuning the forecast to observational data received in real-time. The tuning process involves a bias correction and calculation of a weighted average of the individual models. In this study, the performance of a fully tuned forecast (bias-correct weighted model average) is compared to that of an untuned forecast (un-weighted model average) for three sites in Europe for one full year. The locations differ significantly in the local terrain features: the first site is located in uniform, flat terrain similar to the Great Plains of the United States, the second in a coastal region, and the third in complex terrain near a mountainous region. For all sites, tuning improved the 2-hour-ahead MAE by at least 15% and by more than 40% in complex terrain. The day-ahead forecast MAE for the coastal site and the site in complex terrain improved by 17% and 23%, respectively. For the site in uniform terrain, the day-ahead forecast accuracy was similar to the untuned forecast, suggesting that tuning provides limited improvement in cases of uniform terrain where the untuned model bias is small. This result suggests the potential applicability of low-cost untuned wind power forecasts. Tuning to observations could still provide significant value in complex terrain where localized effects may be poorly captured by today’s generation of global models.