linked in logo | IS-ENES | Contact

Exploring climate model data

Verification of seasonal predictions

Seasonal predictions are probabilistic in nature. From an ensemble the probability of above and below normal temperature, or above and below normal rainfall conditions for the coming period are derived. Verification then uses various metrics to compare the probabilities of these forecasts to the probabilities in actual observations. For that purpose so-called hindcasts are used, i.e. forecasts of past events. Hindcasts nowadays are typically 30years long, giving predictions starting every (3) month(s), using 10-50 ensemble members.

The verification metrics assess reliability, resolution (or discrimination) and sharpness of the forecasts. Each is needed but not sufficient. A full discussion of approaches and metrics is beyond the scope of these documentation pages. More can be found here. Various software packages exist to compute verification metrics, the EasyVerification and SpecsVerification R-packages are recommended.

 

The IS-ENES project has received funding from the European Union's Seventh Framework Programme for research, technological development and demonstration.

Disclaimer