Published May 27, 2026 | Version v1

Seamless hourly atmospheric NO2 TroVCD reconstruction with a spatiotemporal self-correlation informed deep fusion network

  • 1. ROR icon Wuhan University
  • 2. ROR icon Nanjing University of Information Science and Technology
  • 3. ROR icon Hong Kong Baptist University

Description

Accurate monitoring and characterization of nitrogen dioxide (NO2) distributions and trends are essential for understanding atmospheric processes and improving air-quality forecasts. The Geostationary Environment Monitoring Spectrometer (GEMS) has emerged as a new paradigm for atmospheric composition monitoring, providing unprecedented hourly observations of NO2 over East Asia. However, GEMS NO2 vertical column density (VCD) products inevitably suffer from data quality issues (data missing and measurement uncertainty), and previous explorations commonly targeted in only one aspect of quality improvement, which failed to provide seamless high-quality data. This study proposed a unified quality improvement framework based on spatiotemporal self-correlation informed deep fusion network (STSFN) to improve both the spatiotemporal coverage and the accuracy of GEMS hourly NO2 total VCD (TotVCD) and tropospheric VCD (TroVCD) products during 2023–2025. Validation results revealed a satisfactory performance in missing data reconstruction and accuracy enhancement. The final fused products achieved correlation coefficient (R) = 0.807, root-mean-square error (RMSE) = 0.184 DU, mean bias (MB) = 0.006 DU for NO2 TotVCD, and R = 0.644, RMSE = 0.187 DU, MB = -0.011 DU for NO2 TroVCD against the Pandora stations. The final fused datasets also showed high consistency with TROPOMI NO2 data, with R = 0.837 and 0.834 for TotVCD and TroVCD, respectively. These results indicate that our STSFN model can generate reliable and spatiotemporally complete NO2 VCD products. Furthermore, the final fused products enable detailed characterization of NO2 variability across multiple temporal scales, revealing clear regional patterns and spatiotemporal dynamics over East Asia. The seamless high-quality datasets also support the hourly tracking of pollution events, capturing both localized hotspots and extreme abnormal emissions, as well as regional pollution transport. Please note that this dataset specifically contains the hourly NO2 TroVCD products, while the corresponding NO2 TotVCD datasets are available at DOI: 10.5281/zenodo.20388990.

Files

2023.zip

Files (33.3 GB)

Name Size
md5:3ad88cb6582c51ce7f9f37fca898822a
11.1 GB Preview Download
md5:12b0e3ca651a2cedc7e3615ca417f528
11.1 GB Preview Download
md5:0bb09f97629a390f3d75793611de82f5
11.1 GB Preview Download