engleski

Persistence-climatology forecasts for meteorological elements with irregular empirical distributions

For the evaluation of weather forecasts it is necessary to have a referent forecast that can serve as a benchmark showing how far our forecasts is from a perfect one. In the meteorological practice there is wide variety of such forecasts in use. Most often these are the forecasts based on climatology or persistence. However, in short range forecasting neither of these forecast are suitable. This is especially evident for meteorological elements with marked diurnal variation. Many authors (Gringorten, Sanders, Daan, Murphy, etc.) have suggested that forecasts based on some combination of climatology and persistence is optimal for the reference point in the verification procedure. These forecasts are based on the assumption that anomalies of meteorological elements have a tendency to persist with slow tendency to decay. This sort of forecast is known in statistical and meteorological literature as a forecast based on autoregression or as AR(1) model. This forecasts somtimes be called "naive" and "no-skill" forecasts although the term "intuitive" seems more appropriate. Some authors have suggested more sophisticated models as AR(2) or multiple regression as a standard of reference for various skill scores. The reason such forecasts have not found wide use in verification practice is, perhaps, because detailed climatological data and estimates of autocorrelation coefficients are needed for their use. Another reason is that large number of meteorological elements do not have normal distribution, and so direct application of linear regression is not possible. In this talk, we will demonstrate how described difficulty can be overcome with the example of autoregressive forecast formulation for visibility. The values of meteorological elements with irregular statistical distribution can be transformed to the equivalent normal deviates according to the empirical cumulative distribution. The transformed values then become a basis for the estimation of the correlation coefficient and the entry values in the regression equation. Estimated future value can be obtained by apllaying the inverse transformation. Our example is based on climatological data for visibility at airports Zagreb-Pleso and London-Heathrow. Final form of these forecasts is variable, making it possible to apply in verification of categorical and probabilistic forecasts.

verification of weather forecasts; visibility; persistence

nije evidentirano