Abstract
While air quality has improved in many cities, short-term spikes in urban pollution continue to cause elevated health risks. To mitigate such risks, pollution alerts trigger short-term interventions (e.g. temporary industrial curtailments or shutdowns, on-road traffic restrictions, construction bans with dust control, and public health advisories) to rapidly cut emissions and exposure. However, the effectiveness of such alerts has remained uncertain. Here, we analyzed air quality and weather data from 57 cities across northern China between 2018 and 2022 and used a two-step machine learning chain to predict counterfactual concentrations under a no-alert (no-intervention) scenario. Our findings show that interventions enacted under alerts effectively reduced pollutant concentrations, with particulate matter (PM) decreasing by 20-40% and nitrogen dioxide (NO(2)) by 5-25% across different cities, reflecting the variability in alert effectiveness among locations. The reduction in PM(2.5) is estimated to have prevented nearly 54,000 ± 6,000 (∼11%) premature deaths during the study period. Over 80% of these avoided deaths occurred in regions characterized by heavy industries, high coal consumption, and dense population (e.g. Henan, Hebei, and Shandong), where alert-driven interventions had greater impacts. In contrast, service-oriented cities such as Beijing showed moderate but still measurable PM reductions (∼30 μg m(-3)) and correspondingly smaller health benefits. These results provide the first multi-year, multi-city evidence that pollution alerts-through the interventions they trigger-deliver significant and repeatable air quality and public health benefits, offering actionable support for short-term response protocols that complement long-term emission controls in cities worldwide.