The impact of winter clean heating policies on air quality: a combined approach of machine learning and causal inference

The winter centralized heating strategy in China is one of the largest energy consumption systems globally. As coal and biomass are the primary sources of heating energy, the pollutants generated from their combustion have become a major source of air pollution. To improve air quality, China has successively implemented a series of winter clean heating policies. Previous studies have employed methods such as regression discontinuity, difference-in-differences, and synthetic control methods to evaluate the impacts of these clean heating policies, but limitations such as short-term impact, parallel trends assumption and overfitting should not be overlooked. In response, this study collected hourly air pollutant and meteorological data from 145 cities in China over the past decade and applied a weather normalization method based on random forest to decouple the impact of observed meteorological conditions on air pollutant concentrations. The decoupled, weather-normalized pollutant concentrations in each city were then aggregated weekly and applied to a synthetic difference-in-differences (SDID) model to assess the long-term impact of winter clean heating policies on air quality in China. The results showed that from 2014 to 2023, concentrations of AQI, SO₂, NO₂, PM_2.5, PM₁₀, and CO demonstrated a remarkable decline across all regions, with the largest reductions observed in the “2+26” cities. The SDID indicated that policy implementation had a significant impact on pollutant reduction in the “2+26” cities and northern cities. Through the SDID method, we conclude that policy implementation has a significant impact on pollutant emission reduction in "2+26" cities and northern cities. In addition, at the regional and city levels, the initial implementation of the policy yielded significant results, but over time the effects diminished and became difficult to predict. The results of this study can provide valuable support for clean heating policies, and provide important references for air quality improvement and climate change control.