Another Approach to Forecasting Forecast Skill

Abstract

The skill of a medium-range numerical forecast can fluctuate widely from day to day. Providing an a priori estimate of the skill of the forecast is therefore important. Existing approaches include Monte Carlo Forecasting and Lagged Average Forecasting, both of which employ the spread between members of an ensemble of forecasts as a predictor. Instead of working with an ensemble, a new approach to predicting forecast skill is proposed that employs the persistence of the model forecast (within the latest integration) as the predictor. The correlation between this simple predictor and the forecast skill is found to be significant for the entire medium range, both over limited regions (e.g., the Pacific North America sector) and over the Northern Hemisphere. Both root-mean-square and pattern correlation skill scores are used to assess the performance of the forecast and the degree of persistence. The statistical significance of the results is estimated using a Monte Carlo technique. Discrimination of forecast skill is demonstrated using the recent Dynamical Extended Range Forecast experiments carried out by the National Meteorological Center, and is confirmed in tests with independent data.