ITA and Beast methods were used to quantitatively analyze the nonlinear process of a PM2.5 concentration time series based on the PM2.5 concentration data of the three major urban agglomerations in China. The results showed that： ① the degree of the PM2.5 pollution in the three major urban agglomerations had decreased， and the high-concentration areas had noticeably shrunk. The degree of spatial polarization of PM2.5 concentration was reduced， and the spatial difference was narrowed. The PM2.5 concentration in most areas showed downward trends， but the degree of change was not the same. Compared with the YRD and PRD， the concentration of PM2.5 in the BTH was still at a relatively high level. ② The concentration of PM2.5 in the three major urban agglomerations had seasonal variation characteristics that were high in winter and spring and low in summer and autumn. There were obvious differences in PM2.5 concentration between winter and summer， and the convergence of PM2.5 concentration in summer was greater than that in winter. Areas with high PM2.5 concentration also had obvious downward trends， but the downward trends of PM2.5 concentration in the PRD were not obvious compared with those in the YRD and BTH. ③ The PM2.5 concentration time series of the three major urban agglomerations all had significant downward trends： Beijing-Tianjin-Hebei （BTH） > the Yangtze River Delta （YRD） > the Pearl River Delta （PRD）. The PM2.5 concentration had the largest downward trends in winter. The higher the PM2.5 pollution level， the greater the downward trends. ④ The trend component of the PM2.5 concentration time series in the BTH had two change points， and there was one change point in the seasonal component. The trend and seasonal components of the PM2.5 concentration time series in the YRD had no change point. There was no change point in the seasonal component but one change point in the trend component of the PM2.5 concentration time series in the PRD. These results can provide scientific references for regional air pollution control.
Keywords: Bayesian estimator of abrupt seasonal and trend change （Beast）; PM2.5; abrupt change; innovative trend analysis （ITA）; nonlinear variation; trend.