Ce Jia

Haze episodes recur in industrial regions worldwide and slow progress towards the health and urban air-quality targets of the Sustainable Development Goals. Regulators need to know where episodes start, how they spread and which plants to inspect first. We combined hourly PM_2.5 data for 107 counties in Shaanxi, China, with hourly stack monitoring of 346 industrial facilities for 2017–2022. From 102 haze episodes, we built a county network of the order in which pollution rises and ranked facilities by their association with counties that pass pollution on. Episodes usually started in Xianyang, Weinan and Tongchuan and reached Xi’an hours later, and Fuping, Weicheng and Jingyang were most critical for spread. The 20 largest emitters accounted for 42% of emissions but 11% of the propagation-weighted score. Nine facilities, stable across 150 re-estimations and scoring above 95% of placebo scores with time-shifted emissions, form a first-priority list (led by coal-fired power plants and heat suppliers) that applies mainly to the heating season. Network weighting changed the list and extended priority to downwind prefectures but barely improved prediction. The list can guide which plants to inspect first, and targeted controls at these plants are an option that requires plant-specific verification.