Explainable deep learning reveal the contribution of wildfire to ozone in China

As the climate warmings, the frequency and intensity of wildfires have escalated in recent decades.  While the adverse effects of wildfires on air quality are well-documented, their influence on atmospheric ozone in China remains unclear. Here, we apply deep learning and a trajectory-fire interception method (TFIM) to estimate wildfire contributions to ozone concentrations in Chinese cities from 2015 to 2023. Our findings indicate that wildfires influenced 15.1 ± 9.3% of all days during this period, with a wildfire-induced ozone concentration averaging 6.8 μg m-³. Over the nine-year study period, these concentrations exhibited a modest upward trend, increasing by 0.091 μg m⁻³ annually. Regions such as Southwest China, the Qinghai-Tibet Plateau, and Northwest China experienced the highest levels of wildfire-induced ozone. We further utilize SHapley Additive exPlanations algorithms to investigate driving factor behind wildfire-induced ozone. The burnt area, aging hour, and injection height of smoke have a large effect on wildfire-induced ozone concentrations. Finally, we evaluated the health impacts of wildfire-induced ozone, highlighting its significant implications for public health in affected regions.