Evidence for regional heterogeneous atmospheric particulate matter distribution in China: implications for air pollution control

China has suffered from severe nationwide air quality degradation for decades. PM2.5, the atmospheric particulate matter with an aerodynamic equivalent diameter of less than 2.5 μm, is the most concerning atmospheric pollutant for heath. Pollution control policies are commonly applied nationwide, but atmospheric pollution may vary from one area to another, thus suggesting the need for different, adapted policies. However, there is little knowledge on pollution distribution in China. Therefore, here we used recurrent neural network and random forest models to analyze the wintertime regional PM2.5 patterns in four most polluted cities of China, which are Beijing, Shanghai, Guangzhou and Chengdu, from December 2014 to February 2019. We find that different megacities in China have completely different PM2.5 patterns, which remained unchanged during the past 6 years. CO plays a predominant role in shaping PM2.5 nationwide, and the importance of CO varies from region to region. Therefore, different regional PM2.5 control policies should be carried out for better regulation. Furthermore, we demonstrate that PM2.5 is not strongly linked with time series, inferring that PM2.5 concentrations at a given date are not linked with previous PM2.5 concentrations. This finding suggests that the chemical reaction equilibrium may get reversed and that the rate of chemical reactions of PM2.5 is faster than we normally think.