Friday, 28 July 2017: 3:30 PM
Constellation F (Hyatt Regency Baltimore)
Probabilistic seasonal forecasts from the North American Multi-Model Ensemble (NMME) are calibrated and consolidated using a regression methodology designed for application to ensemble prediction systems (Unger et al., 2009). Calibration is primarily used to improve the reliability of probabilities and weight individual models according to their skill. Models with little skill regress to near zero anomalies or may be removed entirely from the anomaly forecasts. The ensemble regression (EREG) method (Unger et al, 2009) uses the ensemble spread to represent conditional uncertainty of forecasts, while adjusting the model and multi-model PDF, improving the reliability of probabilistic forecasts, while maintaining the resolution, and minimizing the mean square error of the ensemble mean. Extreme seasonal forecasts are identified when the probability exceeding 1.28 (or 0.84) standard deviations above or below normal exceeds the climatological probability, approximately 10% (or 20%), and their skill is compared to three-category forecasts, i.e. probabilities of exceeding 0.43 standard deviations above or below normal. Each NMME model is calibrated individually using the EREG methodology and then combined. Weighting individual models of the multi-model ensemble by their overall skill and removing models with the lower skill is explored to maximize the value of the MME. The ensemble regression technique is compared to forecasts made by estimating the probability from the count of ensemble members exceeding the threshold for an extreme forecast. Forecasts are verified using the Brier skill score, as well as assessed for reliability of probabilities. Because skill is particularly low in some seasons and regions, calibrated (EREG) forecasts are on average only a marginal improvement on ensemble counts (CE) in some cases. While individual ensemble models often have negative skill when forecasting extremes, on average over North America, the combined NMME is found to have skill when forecasting extremes in all seasons.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner