A Statistical Exmination of Sky Cover Changes in the Contiguous United States

Abstract

The Sette Weather Chart constructed in 1939 with data from 1900 to 1936 was used to describe the climatology of cloudless days and to predict cloudless days in the United States for 14 regions. When the data for days having 0.1 sky cover or less were collected and analyzed for the period 1950 to 1982 and compared to the Sette period 1900 to 1936, there was a definite statistical decrease in the annual monthly average of cloudless days using 45 cities common to both periods. Using the most recent data a fuzzy k-means clustering analysis identified 17 largely contiguous regions from a sample of 65 cities and constructed a degree of belonging for each city to these regions. The resulting clusters were examined for their uniqueness with discriminant analysis The error rate estimation of the discriminant was evaluated with the resubstitution method, a holdout Method and the Lachenbruch crossvalidation procedure. The results of the study not only show a decrease in cloudless days but also regional changes since 1936; additional cities can be classified into one of the 17 new regions.