Spatiotemporal patterns and drivers of public concern about air pollution in China: Leveraging online big data and interpretable machine learning

Abstract

Public concern about air pollution directly shape residents' risk adaptation behaviors, government policies, and environmental sustainability. However, long-term nationwide studies in China are limited, with even fewer examining the nonlinear mechanisms driving these dynamics. Using Baidu search data from 290 cities across China (2011−2022), we analyzed the spatiotemporal patterns of public concern about air pollution and its mismatch with actual observed pollution levels. We further employed an XGBoost-SHAP model to reveal nonlinear effects of various factors on public concern. The results show a rise-and-fall trend in public concern from 2011 to 2022, with a clear correlation between declining concern and reduced PM_2.5 levels after 2016. Concern was highest in coastal areas, the North China Plain, and northern coal-producing regions. In 2011, 980 million people had “high pollution-low concern,” dropping to around 200 million since 2016. Meanwhile, the population with “low pollution-high concern” steadily grew, highlighting China's progress in air pollution control and public environmental awareness. Actual air pollution levels are not the primary driver of public concern; education and income have the strongest influence. Public concern shows a roughly linear relationship with education, urban development, and media access. However, income and PM_2.5 levels display nonlinear effects: concern plateaus above a per capita income of 28,000 yuan and declines after 60,000 yuan. Similarly, concern stabilizes once PM_2.5 levels exceed 80 μg/m³. This study reveals the nonlinear effects and threshold dynamics driving public environmental concern, offering valuable insights to inform strategies for advancing public environmental awareness and strengthening environmental governance in China.