Application of principal component analysis to atmospheric aerosol size distribution measurements

Measurements of atmospheric aerosol size distributions provide useful information, but the interpretation of these data sets is time consuming. Principal component analysis provides a systematic, robust, fast and easy to use method for simplifying any complex data sets, however, the method requires several modifications of the procedure before it can be applied to the aerosol size distribution data. A set of procedures and modifications are given here; these allow simplification of the aerosol number size distribution data via principal component analysis. Briefly, we begin by weighting the input data using the measurement uncertainties and then apply principal component analysis to the data. Aerosol component loadings obtained from the analysis are rotated by the orthogonal Varimax rotation to produce interpretable results. Finally, we normalize both the aerosol component loadings and scores; this gives physical meaning to the results.; The best application of the simplified size distribution results is to combine with different gas measurements to determine the sources and origins of the measured aerosol particles. We apply this method to five independent field study data sets that were measured at different locations in Southern Ontario and in British Columbia. The results are encouraging and provide useful insights for understanding the similarities and differences among different atmospheric processes observed at different measuring sites. Using the results obtained among different data sets measured from Southern Ontario, we identified several common source factors. These include the photochemical processes, local industrial emissions, regional pollution and the boundary layer dynamics. The photochemical process component represents the freshly nucleated particles that were observed at the measuring site. The local industrial emission component shows the source of the locally observed Aitken particulate matter and SO 2. The regional pollution component indicates a collection of the atmospheric constituents that have widespread sources and are present on a regional scale over Southern Ontario. The boundary layer dynamics component describes the variations of Ox and NOx due to the diurnal variation of the boundary layer depth.; Another usage for the simplified distribution results is to extract sub-sets of the original data for doing inter-comparison between different data sets. Using the simplified size distribution results, we successfully removed size distribution measurements that possess a significant number of nucleation mode particles. This allows direct comparison between the total number concentration measurements obtained by different condensation particle counters that have different minimum detectable particle sizes.; Finally, we use principal component analysis to combine the nano and long DMA size distributions. During the analysis, we confirmed no significant size shift is present between the two sets of size distributions but observed a systematic concentration difference between the two. We applied the same technique and attempted to combine the DMA and PCASP size distributions. The results were not as successful as the combination of the nano and long DMA size distributions. We believe the deviations are due to the very different measuring techniques used by the DMA and PCASP.; In summary, absolute principal component analysis is useful in simplifying atmospheric aerosol size distribution measurements and provides an easy method for analyzing aerosol number size distribution measurements.