Abstract
Selected hourly surface observations from Madison, Wis. and Minneapolis-St. Paul, Minn. are used as basic data for a series of analyses to determine the feasibility of establishing weather classifications. Component analysis (factor analysis) is applied to a sample of January data for Madison to reduce the number of variables needed to suitably describe each day meteorologically and to create orthogonality among these new variables. With these results as the design matrix in regression analysis, a mathematical model for each day is constructed and each day is compared to all other days in order to classify similar days into distinctive weather types. Every day within each class is compared with the synoptic situation for that day to establish whether these types form a reasonable synoptic pattern. The temporal and spatial validity of these newly found weather types is tested by applying the foregoing results to an independent January sample for Madison and an independent January sample for Minneapolis-St. Paul. The basic analytic techniques are then applied to a Madison July sample. Specifically, the results indicate that the elements of a meteorological observation may be expressed by a smaller number of independent components that agree with our knowledge of dynamics; and these newly created components may be applied in a multivariate analysis to establish distinctive weather types. These weather types are synoptically reasonable and their distribution about the usual pattern of Highs and Lows strongly resembles cloud models and photographs from satellites.
*The research for this study was partially supported by National Science Foundation Grant GP-444.
**Current address: Joint Meteorological Satellite Programs Office, Headquarters, U.S. Air Force, Washington, D.C.
