Correlation Structures of PM 2.5 Concentration Series in the Korean Peninsula

Abstract

In this paper, the authors investigate the idiosyncratic features of auto- and cross-correlation structures of PM2.5 (particulate matter of diameter less than 2.5 μ m ) mass concentrations using DFA (detrended fluctuation analysis) methodologies. Since air pollutant mass concentrations are greatly affected by geographical, topographical, and meteorological conditions, their correlation structures can have non-universal properties. To this end, the authors firstly examine the spatio-temporal statistics of PM2.5 daily average concentrations collected from 18 monitoring stations in Korea, and then select five sites from those stations with overall lower and higher concentration levels in order to make up two groups, namely, G1 and G2, respectively. Firstly, to compare characteristic behaviors of the auto-correlation structures of the two groups, we performed DFA and MFDFA (multifractal DFA) analyses on both and then confirmed that the G2 group shows a clear crossover behavior in DFA and MFDFA analyses, while G1 shows no crossover. This finding implies that there are possibly two different scale-dependent underlying dynamics in G2. Furthermore, in order to confirm that different underlying dynamics govern G1 and G2, the authors conducted DCCA (detrended cross-correlation analysis) analysis on the same and different groups. As a result, in the same group, coupling behavior became more prominent between two series as the scale increased, while, in the different group, decoupling behavior was observed. This result also implies that different dynamics govern G1 and G2. Lastly, we presented a stochastic model, namely, ARFIMA (auto-regressive fractionally integrated moving average) with periodic trends, to reproduce behaviors of correlation structures from real PM2.5 concentration time series. Although those models succeeded in reproducing crossover behaviors in the auto-correlation structure, they yielded no valid results in decoupling behavior among heterogeneous groups.