A classification scheme for air

Fig. 6. Box plot of the average PM10 concentrations on each cluster analysis.Figure optionsDownload full-size imageDownload as PowerPoint slide
Here it ANR 94 should be reminded that absolute levels of PM10 were removed from the data series by means of standardisation prior to the clustering. Therefore, the fact that the groups are better separated by PM10 levels than using the conventional categories is remarkable.
So far, each classification has given different groups of stations, based on some particular characteristics. The resulting groups overlap among the different classifications, because some stations share some characteristics but not others. A CHAID analysis of PM10 concentrations considering the two conventional classifications (by zone type and by station type) and the three clusterings of our work shows that a combination of these classifications provides refined results (Fig. 7). It shows that the classification using evolution patterns is the most influential one. Initially, 4 groups are formed; Evol-P Clusters 2 and 3 are merged into one category that groups half the stations.