Get access

The effect of domain size on the relationship between circulation type classifications and surface climate



The ability of circulation type classifications (CTCs) to resolve surface climatic and environmental variables is essential with respect to their application in synoptic climatological applications. This ‘synoptic skill’ depends on several factors including inherent properties of classification methods but as well varying boundary conditions. In this contribution the relevance of the size of the spatial domain for which CTCs are derived is investigated. To this end varying automated CTCs are applied to daily gridded sea level pressure data for 1950–2010 and in each case eight spatial domains of varying size centred around 44 locations spread over the greater north Atlantic European region. For the resulting more than 7000 CTCs the synoptic skill for daily temperature and precipitation taken from the E-OBS v4.0 data set has been estimated using varying evaluation metrics. Resulting values of evaluation metrics aggregated according to varying domain sizes reveal a distinct influence of the size of the domain on the synoptic skill of CTCs. In general highest skill appears to be achieved for domain sizes with a horizontal dimension of roughly 1300–1800 km (in west–east direction) thus covering most frequent size ranges of synoptic scale systems. However, optimal domain sizes tend to be smaller for precipitation (compared to temperature) in summer (compared to winter) and in more continental regions (compared to more oceanic regions). Distinct deviations from the overall finding of relatively small optimal domains emerge for temperatures above/below certain thresholds for which in certain locations and seasons continental scale domains yield highest synoptic skill. Finally the comparison of varying CTCs concerning the effect of the domain size for synoptic skill shows marked differences between methods and moreover clearly elucidates that differences in synoptic skill that can be attributed to varying domain sizes reach comparable magnitude than those related to differing methods.