Detrending moving-average cross-correlation based principal component analysis of air pollutant time series

This work investigates the principal component of air pollutants. The approach is based on detrending moving-average cross-correlation analysis(DMCA) and principal component analysis (PCA). We illustrate the advantages of this method by performing several comparative numerical analysis with traditional principal component analysis (PCA). The results indicate that the principal components obtained by DMCA-based PCA are more reliable in small and medium scale range, and the new method is relatively immune to additive trend and non-stationarity. To further show the utility of DMCA-based PCA in natural complex systems, six air pollutants data collected in Beijing from December 2013 to November 2016 are investigated seasonally. We found that the pollutants PM2.5, PM10 and CO are the most important factors affecting air quality of Beijing, and O3 is the secondary contaminants among four seasons. The contributors to the principal components in winter are the most stable for all time scales, and the second are that in autumn. With these physically explainable results, we have confidence that DMCA-based PCA is an useful method in addressing non-stationary signals.

Automated summary

This study investigates the main components of air pollutants by using detrending moving-average cross-correlation analysis (DMCA) and principal component analysis (PCA). The advantages of this method are illustrated through comparative numerical analysis with traditional PCA. The results show that DMCA-based PCA provides more reliable principal components in the small and medium scale range, and is relatively immune to additive trend and non-stationarity. In addition, the study examines the utility of DMCA-based PCA in natural complex systems using seasonal air pollutant data collected in Beijing. The findings reveal that PM2.5, PM10, and CO are the most significant factors affecting air quality in Beijing, with O3 as a secondary pollutant across the four seasons. The stability of the principal component contributions is highest in winter and second highest in autumn. These results, which can be explained physically, demonstrate the usefulness of DMCA-based PCA in addressing non-stationary signals.