S6 Dynamical and statistical downscaling of ensemble forecasts for wind energy applications in Ireland

Sunday, 23 January 2011
Jennifer Courtney, University College, Dublin, Ireland; and C. Sweeney and P. Lynch

The objective of this research is to get the best possible wind forecasts for the wind energy industry in Ireland using dynamical and statistical post-processing methods. A cluster method allows eight forecasts to be selected from the 51 forecasts which are available in the ECMWF ensemble , by considering how the parameters that are important for wind energy change at forecast range up to 48 hours, over the same area as that on which the local area model (COSMO) will run. We use these representative ensemble forecasts to drive COSMO, which produces forecast data on two nested domains, of resolution ~7km and ~2.5km. Using Bayesian Model Averaging (BMA), these forecasts are further downscaled to specific wind farm locations around Ireland. BMA is a statistical process of assigning weights to each ensemble forecast, based on the ensemble's relative performance over a training period, to generate an optimal forecast. This results in a probabilistic wind forecast for each specified location. Forecast data produced by this method are then compared to observed wind speeds as well as traditional forecasts such as raw model data, rolling bias removal, and rolling trend removal, and skill scores are calculated.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner