Mixture EMOS model for calibrating ensemble forecasts of wind speed.

Area of Science:

• Meteorology
• Statistical modeling
• Data science

Background:

• Ensemble Model Output Statistics (EMOS) is crucial for refining numerical weather prediction ensemble forecasts.
• Existing EMOS models often rely on single parametric distributions, potentially limiting their calibration accuracy.
• Calibrated probabilistic forecasts are essential for reliable decision-making in weather-dependent sectors.

Purpose of the Study:

• To develop and evaluate a novel EMOS model for wind speed forecasting.
• To enhance the calibration of probabilistic forecasts using a flexible mixture distribution approach.
• To compare the performance of the proposed model against established EMOS benchmarks.

Main Methods:

• A weighted mixture of truncated normal (TN) and log-normal (LN) distributions was proposed for EMOS.
• Model parameters and component weights were optimized using proper scoring rules over a rolling training period.
• The model was tested on diverse ensemble datasets, including ECMWF, ALADIN-Hungary, and UW ensembles.

Main Results:

• The proposed mixture EMOS model demonstrated improved probabilistic calibration and point forecast accuracy over raw ensemble and climatological forecasts.
• Significant performance gains were observed compared to TN and LN EMOS methods.
• The mixture model outperformed a TN-LN combination model, offering greater flexibility and avoiding covariate selection issues.

Conclusions:

• The novel mixture EMOS model provides a more flexible and accurate approach to wind speed forecasting.
• This method enhances the reliability of probabilistic weather predictions.
• The findings suggest broader applicability of mixture models in post-processing ensemble weather forecasts.