Weather types prediction at medium-range from ensemble forecasts

Gabriel Jouan (1, 2) , Anne Cuzol (3) , Valérie Monbet (4, 2) , Goulven Monnier (2)

1 L@b SCALIAN
2 IRMAR - Institut de Recherche Mathématique de Rennes
3 LMB - Laboratoire de Mathématiques de Besançon (UMR 6623)
4 SIMSMART - SIMulation pARTiculaire de Modèles Stochastiques

Uncertainties assessment performed by numerical weather ensemble forecast system is active research in the statistical weather community. The uni-variate correction challenge of ensemble forecast known as univariate calibration is a well-known problem with linear models. The recent application of a non-linear approach from the machine learning domain offers a new way to perform calibration of the ensemble forecast. The multivariate component of weather forecast with medium forecasting range represents a high economic value for decision making but an actual difficult border to cross in calibration. One way to study the multivariate calibration and keep a high economic value is the weather types classification. For example, to plan maintenance activities outside or an important manifestation, good weather is preferred and can be studied in the ensemble forecast at medium-range. Good weather can be seen as a weather type created by the interaction of the wind speed and the precipitation. In this article, the multivariate calibration will be managed by a weather types classification problem using the non-parametric approach from the machine learning domain and ensemble forecast at medium-range.

Domaines: Statistiques [math.ST]

Gabriel Jouan, Anne Cuzol, Valérie Monbet, Goulven Monnier. Weather types prediction at medium-range from ensemble forecasts. 9th International workshop on Climate Informatics, Oct 2019, Paris, France. ⟨hal-02425230v1⟩