Three-Phase Transition in China’s Future Air Quality under Ambitious Mitigation Pathways Revealed by Large-Ensemble AI Projections

Abstract Fine particulate matter (PM2.5) and ozone (O3) are major environmental health risks in China, yet their future evolution remains underexplored due to prohibitive computational costs. Here, we develop a hybrid deep learning framework (hyDL) that can reduce computational costs of chemical transport models by >99% while accurately capturing air quality sensitivities to meteorology and emissions (R > 0.9). This efficiency enables a large ensemble of daily air quality projections for China (∼50 km, 2015–2060) under six emission pathways, each coupled with outputs from 7 to 11 climate models. This ensemble reveals an exponential O3–PM2.5 relationship in megacities, suggesting China will likely evolve through three phases featuring distinct responses to emission controls under the ambitious mitigation pathway: an initial phase of effective PM2.5 reduction but persistent O3 challenges; a transitional phase of coimprovement; and a final phase of continued O3 mitigation with only marginal PM2.5 reductions. China is currently in the first phase but is likely to transition to the second under continuous emission controls. Incorporating demographic changes reveals a tug of war between declining pollution and population aging in shaping future health burdens, with only aggressive mitigation reducing mortality. Even under the cleanest pathway, mortality remains flat until ∼2035 before declining. These results highlight the need to incorporate demographic changes into long-term air-quality policy.