The computer-assisted classification of weather at Prague-Clementinum used the average linkage clustering technique. Since the results of the clustering exhibit the snowballing effect, the usual methods of determining the threshold aggregation level (i.e. the level at which the clustering procedure is to be terminated) appeared to be inapplicable. A new method based on Monte Carlo simulations of the means was developed. Its key idea is the termination of the clustering procedure at different aggregation levels in different parts of the data set. This ensures that the number of resultant clusters is reasonable, while minimizing the numbers of very small clusters and unclustered days.The weather categorization resulted in 44 clusters for 14 winters in the period 1965-1978. Thirty-one of the clusters had sizes of 5 or more days. The Monte Carlo scores, comparing the means and variances of the clusters with those of a large number of subsets chosen randomly, indicate that all the resulting clusters represent meaningful weather types.This study may provide a basis for nucleated clustering, which will enable us to deal with longer data series and to study the long-term trends of the properties of weather types.

No abstract

No abstract

No abstract