Informing atmospheric pollution hotspots and exposure risks under climate change using machine learning: Evidence...

Between 2013 and 2022, air pollution exposure caused over one million deaths annually in China, and climate change may further increase such exposure. Here we apply time series and causal inference models to quantify the changes in CO, NO₂, O₃, SO₂, PM_2.5 and PM₁₀ concentrations and the relative impacts on all-cause deaths (ACD) in 2843 regions under climate change. We show extreme high temperature and drought increased by 2 and 3 days on average annually in China, whereas extreme precipitation and low temperature decreased by 2 days. In addition to the national upward trend in O₃ (average annual increase of 2.66 %), other air pollutants only in climate-sensitive regions (western, northeastern, and southeastern China) show regional increases. Climatic indicators such as surface pressure and lake bottom temperature play significant roles in regulating and forecasting air pollution. All air pollutants except SO₂ show positive causal associations with ACD, with each unit increase in pollutants associated with an average of 30.68 ± 24.87 deaths, and co-exposure (various combinations) and climatic indicators amplify health effects, but with increased uncertainty. Our results suggest that climate change and extreme climate events will further increase the risk of severe air pollution and associated deaths in China. SYNOPSIS: This study proposes a deep learning model based on pollutant concentration under the background of climate change (RSSFF) for predicting changes in pollutant concentration and its causal relationship with all-cause deaths.