Development of an integrated machine-learning and data assimilation framework for NO_x emission inversion.

Area of Science:

• Atmospheric Chemistry and Air Pollution
• Environmental Modeling
• Data Assimilation Techniques

Background:

• Nitrogen oxides (NOx) are major air pollutants with adverse health effects and contribute to secondary pollutant formation.
• Accurate NOx emission inventories are crucial for reliable simulations by three-dimensional chemical transport models (3D-CTMs).

Purpose of the Study:

• To develop and apply a comprehensive framework for correcting NOx emissions over China.
• To improve the accuracy of NOx simulations for scientific research and policy development.

Main Methods:

• Integration of 3D-CTM simulations, surface NO2 measurements, and three-dimensional variational data assimilation.
• Application of an ensemble back propagation neural network for emission adjustment.
• Correction of NOx emissions for China during the summers of 2015 and 2020.

Main Results:

• Significant reduction in simulation errors: root-mean-square error, normalized mean error, and normalized mean bias for NO2 decreased by approximately 40%, 40%, and 60%, respectively.
• Observed an average 5% decrease in NOx emissions in 2020 compared to 2015 across the domain.
• Noted substantial NOx emission reductions in Henan (24%) and Anhui (19%) provinces for 2020.

Conclusions:

• The proposed framework effectively corrects NOx emissions, leading to more accurate air quality simulations.
• The methodology is flexible for application to different regions and time periods.
• Provides reliable emission data crucial for air pollution research and policy formulation.