Scholarlink

<p>Ground-level ozone pollution is a pressing urban environmental challenge shaped by complex formation mechanisms, spatial heterogeneity, and nonlinear responses to precursor emissions. Effective management requires understanding these dynamics and their drivers, including urban form and sectoral activities. This study develops a computational framework that integrates built-environment indicators, emission proxies, and satellite-derived observations within a spatial machine learning setting. The study area is partitioned into fine-grained hexagons using Uber’s H3 system, with five-dimensional (5D) built-environment indicators derived from multi-source data. Ozone sensitivity is inferred from the HCHO/NO2 ratio retrieved from Sentinel-5P, while spatial proxies for industrial, traffic, biogenic, and residential emissions are constructed from land-use, vegetation, population, and nighttime lights. Recognizing both local sources and spatial spillovers, we employ XGBoost with GeoShapley to predict ozone concentrations and interpret their nonlinear and spatially varying effects. Results reveal heterogeneous impacts of urban form and emissions, supporting differentiated mitigation strategies and advancing spatially interpretable machine learning in air quality research.</p>

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

Urban Characteristics Driving Spatial Heterogeneity of Ozone Pollution: Insights from Spatial Machine Learning and Geoshapley Analysis

  • 期刊论文

  • 发表日期：

    2026年5月29日

查看全文
  • Xianlong Wang,
  • Zhuoqian Yang,
  • Ke Han

摘要

Ground-level ozone pollution is a pressing urban environmental challenge shaped by complex formation mechanisms, spatial heterogeneity, and nonlinear responses to precursor emissions. Effective management requires understanding these dynamics and their drivers, including urban form and sectoral activities. This study develops a computational framework that integrates built-environment indicators, emission proxies, and satellite-derived observations within a spatial machine learning setting. The study area is partitioned into fine-grained hexagons using Uber’s H3 system, with five-dimensional (5D) built-environment indicators derived from multi-source data. Ozone sensitivity is inferred from the HCHO/NO2 ratio retrieved from Sentinel-5P, while spatial proxies for industrial, traffic, biogenic, and residential emissions are constructed from land-use, vegetation, population, and nighttime lights. Recognizing both local sources and spatial spillovers, we employ XGBoost with GeoShapley to predict ozone concentrations and interpret their nonlinear and spatially varying effects. Results reveal heterogeneous impacts of urban form and emissions, supporting differentiated mitigation strategies and advancing spatially interpretable machine learning in air quality research.