Temporal variations in ozone concentrations derived from Principal Component Analysis

The application of principal components and cluster analysis to vertical ozone concentration profiles in Tsukuba, Japan, has been explored. Average monthly profiles and profiles of the ratio between standard deviation and the absolute ozone concentration (SDPR) of 1 km data were calculated from the original ozone concentration data. Mean (first) and gradient (second) components explained more than 80% of the variation in both the 0–6 km tropospheric and 11–20 km troposphere–stratosphere (interspheric) layers. The principal components analysis not only reproduced the expected inverse relationship between mean ozone concentration and tropopause height (r2 = 0.41) and that in the tropospheric layer this is larger in spring and summer, but also yielded new information as follows. The larger gradient component score in summer for the interspheric layer points to the seasonal variation of the troposphere–stratosphere exchange. The minimum SDPR was at about 3 km in the tropospheric layer and the maximum was at about 17 km in the interspheric layer. The tropospheric SDPR mean component score was larger in summer, possibly reflecting the mixing of Pacific maritime air masses with urban air masses. The cluster analysis of the monthly ozone profiles for the 1970s and 2000s revealed different patterns for winter and summer. The month of May was part of the winter pattern in the 1970s but part of the summer pattern during the 2000s. This statistically detected change likely reflects the influence of global warming. Thus, these two statistical analysis techniques can be powerful tools for identifying features of ozone concentration profiles.