Scholarlink

Against the backdrop of improving urban particulate matter (PM2.5) pollution in China, ground-level ozone (O3) has emerged as the second most critical air pollutant. Given China’s vast geographic expanse, O3 pollution exhibits strong regional characteristics. Exploring the key influencing factors in different regions can aid government agencies in developing targeted O3 prevention and control strategies. In this study, we develop a regional short-term O3 spatio-temporal prediction transformer model (RO3former) to accurately and efficiently forecast ground-level O3 concentrations in the Beijing-Tianjin-Hebei (BTH), Yangtze River Delta (YRD), Pearl River Delta (PRD), Fenwei Plain (FWP), and Chengdu-Chongqing (CC) regions. Additionally, we quantify the contributions of the driving factors through factor elimination experiments in these different regions. The results demonstrate the model’s strong performance, with correlation coefficients stabilized between 0.85 and 0.92 and normalized mean bias (NMB) ranging from −5.87% to 2.77% across the regions and seasons. The input factor ablation experiments reveal that the key forcing factors vary by region. Overall, the historical O3 concentration has the most significant effect on short-term O3 concentration prediction in China, followed by wind speed (u, v) and temperature. The spatial differences in drivers contribute to the heterogeneity of O3 concentrations and exposure risks across China. Therefore, future O3 management strategies should involve region-specific control policies tailored to the identified key influencing factors.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Learning for Interpreting Regional Surface Ozone Spatio-Temporal Prediction

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Canju Zheng,
  • Hengqing Shen,
  • Jianan Sun,
  • Guangliang Liu,
  • Haowei Cao,
  • Jie Zhang

摘要

Against the backdrop of improving urban particulate matter (PM2.5) pollution in China, ground-level ozone (O3) has emerged as the second most critical air pollutant. Given China’s vast geographic expanse, O3 pollution exhibits strong regional characteristics. Exploring the key influencing factors in different regions can aid government agencies in developing targeted O3 prevention and control strategies. In this study, we develop a regional short-term O3 spatio-temporal prediction transformer model (RO3former) to accurately and efficiently forecast ground-level O3 concentrations in the Beijing-Tianjin-Hebei (BTH), Yangtze River Delta (YRD), Pearl River Delta (PRD), Fenwei Plain (FWP), and Chengdu-Chongqing (CC) regions. Additionally, we quantify the contributions of the driving factors through factor elimination experiments in these different regions. The results demonstrate the model’s strong performance, with correlation coefficients stabilized between 0.85 and 0.92 and normalized mean bias (NMB) ranging from −5.87% to 2.77% across the regions and seasons. The input factor ablation experiments reveal that the key forcing factors vary by region. Overall, the historical O3 concentration has the most significant effect on short-term O3 concentration prediction in China, followed by wind speed (u, v) and temperature. The spatial differences in drivers contribute to the heterogeneity of O3 concentrations and exposure risks across China. Therefore, future O3 management strategies should involve region-specific control policies tailored to the identified key influencing factors.