Use of cluster analysis to define periods of similar meteorology and precipitation chemistry in eastern North America.

An aggregation approach is needed to overcome the prohibitive expense involved in exercising the Regional Acid Deposition Model to determine annual deposition loads in eastern North America. A necessary first step in this process is to identify those events which are representative of annual meteorology and to determine their frequencies and pollutant loadings. In this dissertation the statistical technique of cluster analysis is applied to transport vectors and precipitation chemistry data to identify and define representative periods for use in aggregation schemes. Transport vectors representing the travel of air masses in the twelve hours preceding their daily arrival at chemistry sampling sites were calculated for 1979, 1981 and 1983 from mixed-layer back trajectories, arranged in a three-day moving format, and subjected to cluster analysis. The mean transport vectors at twenty-two sites were plotted for each resulting cluster to examine their flow patterns. The associated precipitation and analyte deposition fields were also studied. The sensitivity of the clustering process to differing methods, different sites or number of sites and other factors was assessed. The resulting clusters represent groups whose mean air-mass transport fields are different from one another and correspond to weather situations seen on daily weather maps. The distributions of the transport variables are, in most cases, statistically significantly different from each other. The clusters were relatively insensitive to changes in number of sites or years used; however, different clustering methods yielded different results. The transport clusters were successful in defining wet, dry, polluted and non-polluted clusters in a median sense and in terms of significantly different distributions. The chemistry fields are a plausible result of the transport and persistence characteristics of each cluster. There was a large amount of overlap among the chemistry distributions and large variation within most of the clusters. Cluster analysis has been shown to he useful in the computer-assisted classification of spatial patterns of weather and pollution data and should be considered for use along with more widely used synoptic climatological tools such as principal component analysis and correlation analysis.PhDPhysics, Atmospheric ScienceUniversity of Michiganhttp://deepblue.lib.umich.edu/bitstream/2027.42/161915/1/8821570.pd