On the Use of Statistical Methods as Part of the Short- and Medium-range Forecast Process

For at least the last four decades, the underlying motivation for real-time statistical processing of operational model output has been to provide objective guidance to forecasters who use those same models in the preparation of daily forecasts. Statistical approaches such as MOS produce current estimates of, and correct model errors, and interpret model output variables in terms of the weather elements included in forecasts.

With the rapid changes to the amount and nature of predictive information available to forecasters, the question arises as to whether this is still the best way to assist forecasters in their forecasting process. Are model forecasts “good enough”? How can statistical methods be employed to provide useful guidance products for all short and medium forecast ranges of interest? Should we move on to new forms of “guidance”, which forms should they take, and what should be the role of statistical processing?

Starting with a brief history of statistical post-processing, with an international perspective and with a focus on the forecaster as the intended user of statistically-based products, this presentation will offer some possible answers to the questions posed above. Results from experiments will be used to illustrate the discussion points.