Ensemble Forecasting – Reducing Systematic Error and Identifying Forecast Errors

Thursday, 18 December 2014: 8:30 AM
Hong Guan, SRG Inc., Colorado Springs, CO, United States, Yuejian Zhu, NOAA College Park, College Park, MD, United States and Bo Cui, IMSG, College Park, MD, United States
In order to identify the spatial and temporal distributions of systematic forecast errors, a state-of-art Global Ensemble Forecast System (GEFS) and 20-year reforecast (retrospective forecast or hindcast) have been used to investigate the statistical characteristics of forecast model and forecast uncertainties. GEFS is National Centers of Environmental Prediction (NCEP) daily operational forecast system since 1992. It was upgraded every other year. Latest GEFS (version 11.0.0) will be upgraded in early 2015 with new GFS (version 12.0.0) model, higher resolutions (T574L64 for 0-192 hours; T382L64 for 192-384 hours), 20 perturbed forecasts (plus one control forecast) and initiated every 6 hours. New GEFS is significantly improving forecast capability in terms of ensemble mean and probabilistic forecasts. There are three aspects in this study: 1). Ensemble forecasts could reduce systematic errors through an assimilation of model physics, i.e. stochastic perturbations and parameterizations, multi-model or multi-physics; 2). Enhanced ensemble forecasts could diagnostic model forecast errors in real time through forecast outliers from sequential ensemble forecasts (different lead-time) in terms of spatial and temporal variations to help forecasters to adjust their forecast guidance. Meanwhile, the information of forecast outliers could feedback to model developers to improve model parameterizations and reduce systematic errors; 3). Systematic forecast error could be partially removed through various statistical algorithms that will improve forecast reliability and accuracy for most forecast elements, such as precipitation, temperature, winds and et. al.