On the identification of representative samples from large data sets, with application to synoptic climatology

Abstract

The analysis of large data sets in meteorological and air quality studies is often made though the examination of specific case studies, especially when time-consuming computational models are employed. This paper presents the development of a tool to identify specific case studies, termed as representative days, that would subsequently be modelled. The success of such tools should be judged on the discrimination between the specified cases: and the degree to which they capture and recreate historical characteristics of the original data set. The developed approach utilises a principal component algorithm with varimax rotation (r-PCA) and the subtractive clustering algorithm coupled with a cluster validity criterion. In this paper, the developed tool is applied to a data set from the North Sea, utilizing two years worth of data from the DNMI operational forecasting model. The results will be subsequently used in photochemical and radiative forcing modelling tools as part of the EC funded project AEOLOS, with the ultimate goal to estimate the global warming potential of non-radioactive tracing substances such as SF6 and PFCs, which are heavily used in the oil industry.