High spatiotemporal resolution estimation of AOD from Himawari-8 using an ensemble machine learning gap-filling...

Area of Science:

• Environmental Science
• Atmospheric Science
• Data Science

Background:

• Aerosol optical depth (AOD) data is often incomplete and unavailable in urban regions, hindering air quality research.
• Existing AOD products have limitations in spatial and temporal resolution for detailed urban air quality analysis.

Purpose of the Study:

• To develop a comprehensive, high spatiotemporal resolution aerosol optical depth (AOD) dataset for key urban areas in China.
• To overcome data gaps and improve the availability of AOD data for air quality applications.

Main Methods:

• An ensemble machine learning approach combining a random forest-based Space Interpolation Model (SIM) and a deep neural network-based Time Interpolation Model (TIM).
• The SIM filled spatial gaps in Himawari-8 AOD data at 0.05° resolution, and the TIM enhanced temporal resolution to 10 minutes.
• A full-coverage AOD dataset for Beijing-Tianjin-Hebei, Yangtze River Delta, and Pearl River Delta in 2020 was generated.

Main Results:

• The developed AOD dataset achieved a high spatiotemporal resolution of 0.05° and 10 minutes.
• Validation against AERONET and SONET in-situ observations showed satisfactory performance (R = 0.80), particularly in spring and summer.
• The model effectively reconstructed AOD data, addressing incompleteness and availability issues.

Conclusions:

• The ensemble machine learning model offers an effective method for reconstructing high-resolution AOD datasets.
• This advancement can significantly improve near-real-time air quality monitoring in urban environments.
• The generated AOD dataset holds potential for enhanced air quality research and policy development.