Model Postprocessing beyond the Realm of Weather and into Climate Prediction: Operational Experience at the Climate Prediction Center (Invited Presentation)

The skill of Numerical Weather Prediction (NWP) models decreases with lead time until, at some point, the prediction of individual weather systems is no longer useful. Useful skill can be extended further by predicting the range of weather conditions over multi-day periods. These predictions are usually expressed as departures from climatology to make their interpretation clear. Thus, at extended ranges, the forecasts begin to take on the characteristics of a climate forecast, in that individual weather systems are no longer forecast in favor of the general multi-day conditions compared to a climatological reference.

Recently, atmospheric models have improved to the point that they are capable of predicting and simulating the formation and evolution of weather system. When this capability is combined with oceanic prediction and run as ensembles, these models also become “climate forecast models” in the sense that they are capable of anticipating the influence of boundary conditions, such as anomalous ocean conditions (e.g. ENSO), on the statistics of weather.

This presentation traces the transition of model post-processing techniques beyond the realm of weather and into climate prediction. Post-processing techniques used for the 6-10 day outlooks in the 1970’s initially involved a heavy emphasis on specification equations relating predicted upper-level geopotential heights to surface temperature and precipitation. As the skill of the models improved, post-processing methods increasingly focused on the calibration of direct model forecasts of surface elements. Forecast ranges were extended to include the 8-14 days outlook, and, recently, outlooks for weeks 3 and 4. The evolution of the techniques used for model post-processing to support extended range weather forecasts, issued at the Climate Prediction Center (CPC) together with their improvement in skill will be reviewed. Particular attention will be given to the role of ensemble prediction in the improvement of forecast performance.

Climate Models were not used for the CPC 30-day and seasonal outlooks until the mid-1990’s. Initially the ensemble size of the atmospheric predictions were small and post-processing involved relative simple bias correction. As increasing computer power enabled larger ensemble sizes, and the models became increasingly accurate at reproducing climatic features, the post-processing methods grew in sophistication. Today’s CPC monthly and seasonal outlooks are largely based on a collection of ensemble members from climate model forecasts available from the North American Multi-Model Ensemble (NMME). The NMME is the result of a coordinated effort by several modeling centers to provide climate model forecasts on a timely basis in support of CPC operations. Retrospective forecasts back to the early 1980’s are provided for each of the models for calibration. Post-processing approaches applied to the NMME include regression based methods similar to Model Output Statistics (MOS), Bayesian processing, and blending information obtained from dynamical and statistical models. A summary of these approaches will also be reviewed in this presentation.