21 JUL 2026 (TUE) 12:05 - 12:35

Daytime Hourly PM2.5 Concentration Mapping in Six Urban Regions Using GTNNWR

Mr. SUN Weixuan     

( Supervisor: Prof Bo Huang )

Abstract:

Urban PM₂.₅ exposure assessment is critical to the health and safety of residents. However, existing evaluation methods largely rely on daily mean concentrations and sparsely distributed ground based hourly observations, which fail to capture the spatially continuous variations of pollutants. This leads to a systematic underestimation of the actual inhalation doses of populations across different time periods and also makes it difficult for residents to understand the air pollution situation in their surrounding areas. To address this, we trained a Geographically and Temporally Weighted Neural Network Regression (GTNNWR) model using daytime Aerosol Optical Depth (AOD) data, multi source remote sensing auxiliary data, hourly meteorological data, and ground based PM₂.₅ observations across six urban regions: Bangkok, the Pearl River Delta, New York, Los Angeles, San Francisco, and Sacramento. Hourly PM₂.₅ concentration maps (at 1 km spatial resolution) were generated for the daytime period (6:00 – 18:00) on selected dates throughout the year 2021-2022. The model achieves continuous spatiotemporal coverage of PM₂.₅ concentrations at an hourly scale and effectively captures intra urban local spatiotemporal pollution heterogeneity. Based on the generated hourly pollution maps, multi faceted analyses were conducted from the perspectives of monthly means, diurnal means, and stubborn pollution zones, revealing the spatiotemporal differences and patterns of pollution across the six regions. Our results demonstrate that GTNNWR based hourly pollution maps can uncover the exposure risk time windows masked by daily mean values, providing a scientific basis for time targeted pollution interventions.