Comparing the Skill Displayed by Two Statistical Schemes that Interpret the ECMWF Ensemble Prediction System Control Model and the NCEP Global Forecast System (GFS) Model

Weather forecasters have access to a number of numerical weather prediction (NWP) models and a range of statistical sytems to interpret their output.

The primary purpose of the current paper is to compare the relative skill displayed by statistical systems used to generate predictions for Melbourne, Australia, when applied to the output of two NWP models.

The two NWP models subjected to this evaluation are the ECMWF Ensemble Prediction System Control Model and the NCEP Global Forecast System (GFS) Model.

The predictions evaluated are, for the official Melbourne observation site, estimates of the inter-diurnal change in minimum and maximum temperature and the amount and probability of precipitation, and for the Melbourne Airport observation site, estimates of the 9am and 3pm inter-diurnal change in wind direction and speed.

Also examined is the extent to which the forecasting skill displayed by the two sets of predictions might be enhanced by combining the predictions generated by the two models.

Preliminary results concerning the skill displayed by day-to-day predictions for Week One (Days 1 to 7), and based upon the evaluation of a relatively short period of generated predictions, indicate that, for some forecast parameters, those sourced from the output of the the NCEP Global Forecast System (GFS) Model are superior whilst for other forecast parameters, those sourced from the output of the ECMWF Ensemble Prediction System Control Model are superior.

Combining the predictions generated by the two models is shown to lead to predictions of greater skill than that displayed by either taken individually.

.

FIGURE Illustration of how combining the predictions generated by the two models may lead to predictions forecasts of greater skill than that displayed by either taken individually.

.