Explainable machine learning identifies the lower-tropospheric thermal gradient as the dominant seasonal driver of surface ozone in the Yangtze River Delta, China

Abstract Ground-level ozone pollution in the Yangtze River Delta (YRD) poses a persistent air quality challenge, yet the role of the vertical atmospheric structure remains poorly quantified. In this study, we employ an explainable machine learning framework to quantify the contribution of lower-tropospheric thermal dynamics on seasonal surface ozone variability from 2015 to 2022. We focus on the temperature difference between 850 hPa and 1000 hPa, denoted as T diff , which captures the vertical thermal gradient between the free troposphere and the near-surface layer. This gradient serves as a robust indicator of atmospheric stability: a strongly negative T diff reflects a steep lapse rate that enhances vertical mixing and pollutant dispersion, whereas a weakly negative or positive T diff indicates the presence of a thermal inversion that suppresses vertical exchange and promotes ozone accumulation near the surface. Our results reveal that T diff is the most statistically important predictor of surface ozone, outperforming both chemical precursors and surface meteorological variables. Although the influence of thermal stratification on air quality is well recognized, this study provides, the first systematic quantification of its dominant role relative to other drivers in the YRD using a data-driven, interpretable modeling approach. These findings identify the lower-tropospheric thermal structure as a key physical driver of surface ozone pollution and offer a practical metric for improving operational air quality forecasting.