Evaluation of Cool-Season Extratropical Cyclones in a Multi-Model Ensemble for Eastern North America and the Western Atlantic Ocean
The operational models being evaluated include the 50-member European Centre for Medium-Range Weather Forecasts (ECMWF), the 20-member National Centers for Environmental Prediction (NCEP), and the 20-member Canadian Meteorological Centre (CMC). The ECMWF ERA-Interim Re-Analysis is used to verify cyclone properties of the TIGGE archive for the October to March cool season from 2007-2014. A check of the results will also be completed using the Global Forecast System (GFS) analysis for the verification. The long-term NCEP performance will be evaluated using the Global Ensemble Forecast System Reforecast, Version 2 (GEFS-R) and verified using the Climate Forecast System Reanalysis (CFSR) from 1985-2009.
The Hodges surface cyclone tracking scheme was used to construct cyclone tracks by tracking 6-hourly MSLP from the analyses and ensemble members. The cyclone tracks are analyzed and binned into different groups according to forecast period, cyclone intensity, and magnitude of the cyclone errors. Short-term forecasts (days 1-3), medium range forecasts (days 4-6), and long range forecasts (days 7-10) performance of these ensembles will be discussed as well as any benefits of combining these ensemble systems. For example, the 1-3 day forecasts of the GEFS-R ensemble mean underpredicts cyclones just off the East Coast by 0.5-1.0 cyclones per cool season and underpredicts 1.0-1.5 cyclones per cool season near portions of the Great Lakes, while showing a slight overprediction of 0.5-1.0 cyclones per cool season in the Northeast U.S. Additional metrics will be presented, which will be separated into different storm tracks (Miller A vs. Miller B), such as the bias (mean error), mean absolute error (MAE), and the root mean square error (RMSE) for both cyclone position and intensity. The probabilistic skill is assessed using the Brier Skill Score, among other metrics; to show how representative the ensemble spread is relative to the uncertainty in the ensemble forecast.