Towards Predictability Limit: Advancing the Deterministic Skill of Ensembles

• Published in: Perspectives on Atmospheric Sciences pp. 87 - 92
• Main Author: Kioutsioukis, I.
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2016; Springer International Publishing
• Series: Springer Atmospheric Sciences
• Subjects: Deterministic Model; EARTH SCIENCES; Ensemble Forecast; Ensemble Member; Environmental impact of natural disasters & phenomena; Error Dependence; Skill Difference
• ISBN: 9783319350943; 3319350943
• ISSN: 2194-5217; 2194-5225
• DOI: 10.1007/978-3-319-35095-0_13

Forecasts from deterministic models are subject to uncertainties in the input data as well as the model itself. Multi-model ensemble forecasts can improve the forecast skill under certain conditions. Generally, without particular consideration of the required restrictions, the majority of the studies demonstrate an increased forecast skill for the multi-model ensemble mean. We demonstrate through an intercomparison of two dissimilar air quality ensembles that unconditional raw forecast averaging, although generally successful, is far from optimum. The way to achieve an optimum ensemble is also presented. The skill gained from the proper ensemble averaging has at least the double magnitude with the skill improvement using the full ensemble. The combined skill earned from conditional ensemble averaging is comparable with the one obtained each decade as a result of the aggregated advancements in numerical prediction due to more and better assimilated observations, higher computing power and progress in our understanding of dynamics and physics.