Operational Evaluation of a Wind-farm Forecasting System

Performance of a wind-forecasting system for a wind-farm in Ireland is reported. Forecasts were based on ensembles constructed from HARMONIE model runs every 6 hours, along with extra high-resolution HARMONIE runs every 12 hours. Statistical post-processing with Bayes Model Averaging (BMA) removed bias very effectively. The “raw” incremental skill provided by each extra ensemble member was negligible, but the net value, after BMA post-processing, was significantly larger. Thus, a small ensemble with BMA is more skillful than a larger ensemble with simple averaging only. A larger ensemble is still more skillful than a smaller one, if both use BMA.