Dataset for \An AI-Based Super-Resolution Framework for Hourly-Hectometer Atmospheric NO\u2082 Reconstruction\

Urban air pollution exhibits pronounced spatiotemporal heterogeneity because of complex anthropogenic emissions and urban environments. Therefore, the emission sources identification, and the transport, transformation and health exposure assessments of atmospheric pollutants rely heavily on high-resolution monitoring. This work developed a deep learning model to achieve hourly-hectometer air pollutant mapping of nitrogen dioxide (NO2) in Hefei, China from kilometer-level hyperspectral satellite monitoring. Hectometer-resolution traffic flow and point of interest (POI) were integrated as representations of specific emission sources. Gaussian diffusion simulation was applied to capture the impact of atmospheric conditions and transport processes on ground-level NO\u2082 concentrations. The reconstructed hourly-hectometer NO2 distribution was validated with in-situ monitoring networks, horizontal ground-based remote sensing and mobile observations, with correlation coefficients of 0.81, 0.75 and 0.86, respectively. Through the 100 m-level NO\u2082 distribution, some minor emission sources were identified within urban area, such as river ports and congested road sections, which are hardly captured by kilometer-level hyperspectral satellite monitoring. The hectometer-scale distribution reveals an average underestimation of 21% in NO2 exposure in near-road communities when using the in-situ monitoring network data. Additionally, the number of days exceeding the daily NO\u2082 guideline can vary by up to 24% even between neighboring communities located just 530 m apart. The model offers a unique perspective on urban air quality assessment and pollution attribution.