Attribution of Black Carbon Variability in China (2000–2019) from a Perspective of Machine Learning

Substantial black carbon (BC) emissions in China have raised serious concerns owing to their significant influence on climate change and health. However, knowledge around the relative contributions of emissions and meteorological conditions to BC dynamics is limited but essential for air pollution management. Therefore, emission-driven (BCEMI) and meteorology-driven (BCMET) BC concentrations in China during 2000–2019 were quantified by a machine learning framework, focusing on five regions (NC: North China, YRD: Yangtze River Delta, PRD: Pearl River Delta, SCB: Sichuan Basin, and CC: Central China). Furthermore, driving mechanisms of key meteorological factors were investigated using Shapley Additive Explanation (SHAP). Results show a dominant role of emissions in shaping BC variability, with ratios of regional average BCEMI changes to total changes ranging from −140.50% to 76.40%. Especially, the most pronounced decrease occurred in NC during 2013–2019, with BCEMI dropping by 1.56 μg/m3. Even so, the impact of extremely adverse meteorological conditions on BC variations cannot be ignored. The highest annual mean BCMET in YRD (0.17 μg/m3) and PRD (0.30 μg/m3) was observed in 2004, while positive BCMET in NC, SCB, and CC peaked in 2013, with values of 0.26, 0.18, and 0.18 μg/m3, respectively. Regarding SHAP values of each feature, meteorological effects in NC, YRD, SCB and CC were dominated by boundary layer height and temperature, whereas those in PRD were mainly regulated by precipitation and wind. These findings provide a new perspective for attributing BC variability and offer valuable insights for optimizing regional BC control strategies and air quality models.