New Deep Learning Model to Estimate Ozone Concentrations Found Worrying Exposure Level over Eastern China

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Ozone (O3) concentrations are increasing in eastern China.
• O3 impacts human health, biodiversity, and climate.
• Accurate spatiotemporal O3 data is vital for exposure assessments.

Purpose of the Study:

• To develop a deep learning model for estimating daily maximum 8-hour average (MDA8) O3 across eastern China in 2020.
• To assess the spatiotemporal distribution of O3 and associated population exposure.
• To compare the model's performance against traditional methods.

Main Methods:

• Developed a deep learning model combining Long Short-Term Memory (LSTM) network with an attentional mechanism and residual connections.
• Utilized Tropospheric Monitoring Instrument (TROPOMI) total O3 column data, meteorological data, and other covariates as inputs.
• Validated model estimates against the China air quality monitoring network observations.

Main Results:

• The LSTM-based model demonstrated superior performance compared to random forest and deep neural network models.
• Achieved a cross-validation R2 of 0.94 and RMSE of 10.64 μg m−3.
• Estimated that 81% of eastern China's population experienced MDA8 O3 levels exceeding 100 μg m−3 for over 150 days in 2020.

Conclusions:

• The developed deep learning model provides accurate spatiotemporal O3 estimates.
• Significant population exposure to elevated O3 levels was identified in eastern China.
• Findings highlight the urgent need for strategies to mitigate O3 pollution and its health impacts.