Multi-scale characteristics and statistically associated factors of ozone pollution in Hangzhou based on machine learning

Abstract

Ozone pollution poses a significant air quality challenge in Hangzhou in recent years. This study investigated the multi-scale characteristics and statistically associated factors of ozone pollution in Hangzhou based on O₃ data from 12 monitoring stations in the city from 2018 to 2022, along with data on other pollutants and meteorology. The analysis utilized the PAM (Partitioning Around Medoids) clustering method, KZ (Kolmogorov–Zurbenko) filtering method, and XGBoost (eXtreme Gradient Boosting) combined with the SHAP (Shapley additive explanations) model. The results indicate that: (1) Clustering of the MDA8-O₃ (Maximum Daily 8-Hour Average O₃) data from monitoring stations using PAM reveals that Hangzhou can be divided into three sub-regions: west, central, and east, with varying degrees of ozone pollution from low in the west to high in the east. (2) Decomposition of the MDA8-O₃ concentration time series into long-term, seasonal, and short-term components highlights that the short-term components primarily drive the fluctuations in the original sequence. (3) At both temporal and spatial scales, disparities in the statistically associated factors of ozone pollution exist. Temporally, temperature and relative humidity dominate seasonal and short-term components, while long-term components are statistically associated with both temperature and long-term emissions. Spatially, temperature is the main factor in the west, but diminishes in the central and eastern regions, where other pollutants become more influential. Regional differences in emission sources near monitoring sites also affect statistically associated factors. The findings of this study can offer valuable insights for developing targeted strategies for ozone pollution control in Hangzhou.