Calibrated Hub-height Wind Speed Forecasts in Complex Terrain from a Multi-initial Condition, Multi-PBL Scheme WRF Ensemble

State-of-the-art wind forecast systems correctly capture wind speed variability as well as an accurate depiction of forecast uncertainty in terms of a probabilistic spread. In this study, a 48-member WRF ensemble is derived using two initial conditions (the Global Forecast System and North American Mesoscale model), eight planetary boundary layer (PBL) schemes, and three grid spacings (36-, 12-, and 4-km). A daily ensemble forecast was produced for four wind farm sites in the complex terrain of British Columbia for the year spanning 1 June 2014 through 31 May 2015.

Results from the raw ensemble varied at each location, but showed the need for bias correction and ensemble calibration to improve both forecast accuracy and reliability scores. Raw ensemble wind forecasts at hub-height were generally under-dispersive with some locations also showing systematic biases.

Bias-correction and calibration of the ensemble forecast system was performed using the Component-based Post-processing System (COMPS). Calibrated forecasts at each location were produced by first removing the bias of each member by using historical observations. Calibration of the spread of the ensemble was necessary to ensure the forecasts were reliable. Reliable forecasts have the property that the forecasted probabilities accurately match the observed frequency of occurrence. Reliable forecasts are important to energy planners to optimize generation, distribution, and market trading.