Ensemble wind speed forecasting for a wind plant in British Columbia

This paper examines the performance of wind speed forecasts for an operating wind plant in British Columbia over a period of 6 months using an 11-member ensemble run by the University of British Columbia Geophysical Disaster Computational Fluid Dynamics Center. Three different hub-height interpolation methods, along with three different model output statistics (MOS) techniques, including linear regression, a 2-week running average bias, and Kalman filtering, are tested. The resulting ensembles are verified using bias, root mean squared error and mean absolute error plots. Specific weather events, which were poorly forecasted, are examined to determine why the ensemble was unable to capture the observed winds. Verification, using a bias calculation to identify systematic errors in specific ensemble members and a cross correlation to identify timing errors of such events, is preformed. Higher resolution models are used in hind-cast and different ensemble configurations are examined to determine if forecasts could be improved.