Kalman Filter bank post-processor methodology for the Weather Research and Forecasting Model wind speed grid model output correction

Abstract

The performance of the Weather Research and Forecasting (WRF) model, coupled with a bank of eight Kalman Filters (KFB) as a post-processor toolbox for the three hourly average WRF wind speed forecasts, is investigated and compared to the output of the WRF model alone. Two model set-ups, WRF and WRF+KFB, have been tested for the period January to December 2008 on nine locations corresponding to gradient wind towers of the Cuban Eolic program. Tests demonstrated that the KFB post-processing technique, using a third-order polynomial, combined with a four-point a priori moving window averager for covariance matrix computation, was the best configuration for improving the WRF grid model day-ahead wind speed forecast output. The WRF+KFB approach investigated has been shown to adapt to changing wind speed patterns and to offer improved wind speed forecasts for each location considered, whilst only requiring a limited data set for training purposes.

Keywords

• wind forecasting
• wrf
• post-processing
• kalman filter