A comparison of principal components using TPCA and nonstationary principal component analysis on daily air-pollutant concentration series

Abstract

Abstract We applied traditional principal component analysis (TPCA) and nonstationary principal component analysis (NSPCA) to determine principal components in the six daily air-pollutant concentration series (SO 2 , NO 2 , CO, O 3 , PM 2.5 and PM 10 ) in Nanjing from January 2013 to March 2016. The results show that using TPCA, two principal components can reflect the variance of these series: primary pollutants (SO 2 , NO 2 , CO, PM 2.5 and PM 10 ) and secondary pollutants (e.g., O 3 ). However, using NSPCA, three principal components can be determined to reflect the detrended variance of these series: 1) a mixture of primary and secondary pollutants, 2) primary pollutants and 3) secondary pollutants. Various approaches can obtain different principal components. This phenomenon is closely related to methods for calculating the cross-correlation between each of the air pollutants. NSPCA is a more applicable, reliable method for analyzing the principal components of a series in the presence of nonstationarity and for a long-range correlation than can TPCA. Moreover, using detrended cross-correlation analysis (DCCA), the cross-correlation between O 3 and NO2 is negative at a short timescale and positive at a long timescale. In hourly timescales, O 3 is negatively correlated with NO 2 due to a photochemical interaction, and in daily timescales, O 3 is positively correlated with NO 2 because of the decomposition of O 3 . In monthly timescales, the cross-correlation between O 3 with NO 2 has similar performance to those of O 3 with meteorological elements. DCCA is again shown to be more appropriate for disclosing the cross-correlation between series in the presence of nonstationarity than is Pearson’s method. DCCA can improve our understanding of their interactional mechanisms.