High resolution forecasting for wind energy applications using Bayesian model averaging

Jennifer F Courtney,Peter Lynch,Conor Sweeney

doi:10.3402/tellusa.v65i0.19669

Jennifer F Courtney, Peter Lynch + Show 1 more

Open Access

https://doi.org/10.3402/tellusa.v65i0.19669

Copy DOI

Abstract

Two methods of post-processing the uncalibrated wind speed forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) ensemble prediction system (EPS) are presented here. Both methods involve statistically post-processing the EPS or a downscaled version of it with Bayesian model averaging (BMA). The first method applies BMA directly to the EPS data. The second method involves clustering the EPS to eight representative members (RMs) and downscaling the data through two limited area models at two resolutions. Four weighted ensemble mean forecasts are produced and used as input to the BMA method. Both methods are tested against 13 meteorological stations around Ireland with 1 yr of forecast/observation data. Results show calibration and accuracy improvements using both methods, with the best results stemming from Method 2, which has comparatively low mean absolute error and continuous ranked probability scores.

Highlights

Two methods of post-processing the uncalibrated wind speed forecasts from the European Centre for MediumRange Weather Forecasts (ECMWF) ensemble prediction system (EPS) are presented here
The focus of the current study is on medium-term forecasting of up to '48 hours, which relies on numerical weather prediction (NWP) and ensemble prediction systems (EPS)
These disimprovements are small in comparison to the improvements made to the calibration of the wind speed forecast at Knock Airport so overall it can be argued that post-processing the limited area models (LAMs) forecasts with Bayesian model averaging (BMA) does add value to the forecasts at this location

Summary

Introduction

Two methods of post-processing the uncalibrated wind speed forecasts from the European Centre for MediumRange Weather Forecasts (ECMWF) ensemble prediction system (EPS) are presented here Both methods involve statistically post-processing the EPS or a downscaled version of it with Bayesian model averaging (BMA). The value of an EPS is in its interpretation, and its forecast potential goes far beyond It was proposed by Gneiting et al (2007) that the aim of probabilistic forecasting is to maximise sharpness subject to calibration. An alternative definition of sharpness is the tendency of a probabilistic forecast to predict extreme values or deviations from the climatological mean and is an attribute of the marginal distribution of the forecasts Applying this definition would imply a wider interval to be considered a better forecast (Jolliffe and Stephenson, 2011). Throughout this article we will implement the former definition of sharpness as defined by Gneiting et al (2007)

Methods

Results

Conclusion