Multiscale time-lagged correlation networks for detecting air pollution interaction

In order to investigate the interaction of air pollution between neighboring cities, we develop a novel multiscale time-lagged networks framework, which contains three parts, namely, (1) a time-lagged cross-correlation coefficient between every pair of sites is calculated by using time-lagged height cross-correlation analysis (Time-lagged HXA), which allows us to quantify the potential time-lagged effect of pollutant concentrations time series between each pair of sites at different time scales. (2) A connection between each pair of sites can be defined as the cross-correlation coefficient with the idea of local temporal weights. Thus, a multiscale directed weighted network among the monitored sites can be constructed. (3) Five network attributes, namely, average clustering coefficient, network density, network structure entropy, node transmission rate, and attraction rate are employed to analyze the constructed networks. Besides, Markov process is utilized to assess the propagation patterns of pollutants at every monitored site and further to trace the potential pollution sources. By applying the proposed multiscale time-lagged networks to explore the interaction of SO2 and PM2.5 monitored in 43 neighboring sites in North China, some interesting findings help us better understand the propagation characteristics of air pollution between neighboring districts.