Application of Non-hierarchically Clustered Circulation Components to Surface Weather Conditions: Lake Superior Basin Winter Temperatures

Abstract

An S-mode principal component analysis was applied to the correlation and covariance matrices of a 700 hPa height window centered over eastern North America. The first five components were retained and rotated. K-means clustering algorithm was applied to the four solutions to generate 22 circulation types for each. Regression models were built, relating monthly mean temperature anomalies to monthly circulation type frequencies. The results were compared to those obtained in an earlier study using a correlation-based approach to classify the same data set. It was found that i) the amount of within-type variability is comparable for the two and that ii) the difference between cold and warm days of a circulation type is for both classification approaches an anomaly located to the east of the lake. A detailed analysis of this difference indicated that some information is contained in the scores for the first five principal components (and could therefore be used to improve the regression results) but that some is contained in the higher order components and was thus lost. Differences in the PC scores between warm and cold days were used to divide the circulation types into warm and cold subtypes. This improved the regression results but the best results were still inferior to those that had been obtained with the correlation-based classification.