Scholarlink

<p>Long-term assessment of urban air quality is often hindered by the “Reverse Data Quality Pyramid,” where historical monitoring records lack the spatial resolution and multi-pollutant coverage of modern systems. This study addresses the critical data sparsity in Shanghai from 2014 to 2019, a period where district-level records were predominantly limited to single indicators. We present the first gap-free, high-resolution spatio-temporal air quality inventory for Shanghai and its 16 administrative districts spanning 2014-2025, covering seven key indicators (PM<sub>2.5</sub>,&#xa0; PM<sub>10</sub>,&#xa0; O<sub>3</sub>,&#xa0; NO<sub>2</sub>,&#xa0; SO<sub>2</sub>,&#xa0; CO, and AQI). The dataset is generated via a novel Hierarchical Spatio-Temporal Reconstruction Framework that employs Automated Multi-Layer Stacked Generalization (AMLSG). By integrating Deep Neural Networks with Gradient Boosting Decision Trees, the framework executes a dual-transfer learning strategy: temporally backcasting city-level baselines and spatially downscaling these to district-specific profiles using local PM<sub>2.5</sub> as a physical anchor. Validated against a masked ground-truth subset (2022–2025), the reconstruction demonstrates high fidelity, achieving a Normalized Root Mean Square Error (NRMSE) below 7% for all pollutants. This inventory bridges the gap between early monitoring efforts and current standards, providing a robust foundation for meso-scale environmental governance and epidemiological 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

Reconstructing a 16-District Spatio-Temporal Air Quality Inventory for Shanghai (2014–2025): Reverse Backcasting via Hierarchical Multi-Layer Stacked Generalization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月22日

查看全文
  • Wei Deng,
  • Hailin Yang

摘要

Long-term assessment of urban air quality is often hindered by the “Reverse Data Quality Pyramid,” where historical monitoring records lack the spatial resolution and multi-pollutant coverage of modern systems. This study addresses the critical data sparsity in Shanghai from 2014 to 2019, a period where district-level records were predominantly limited to single indicators. We present the first gap-free, high-resolution spatio-temporal air quality inventory for Shanghai and its 16 administrative districts spanning 2014-2025, covering seven key indicators (PM2.5,  PM10,  O3,  NO2,  SO2,  CO, and AQI). The dataset is generated via a novel Hierarchical Spatio-Temporal Reconstruction Framework that employs Automated Multi-Layer Stacked Generalization (AMLSG). By integrating Deep Neural Networks with Gradient Boosting Decision Trees, the framework executes a dual-transfer learning strategy: temporally backcasting city-level baselines and spatially downscaling these to district-specific profiles using local PM2.5 as a physical anchor. Validated against a masked ground-truth subset (2022–2025), the reconstruction demonstrates high fidelity, achieving a Normalized Root Mean Square Error (NRMSE) below 7% for all pollutants. This inventory bridges the gap between early monitoring efforts and current standards, providing a robust foundation for meso-scale environmental governance and epidemiological research.