Mimicking atmospheric photochemical modelling with a deep neural network

Area of Science:

• Atmospheric chemistry and air pollution modeling.
• Application of artificial intelligence in environmental science.

Background:

• Accurate prediction of ambient ozone (O3) is vital for pollution control, especially with climate change.
• Current chemical transport models (CTMs) are computationally intensive, limiting their practical use in air quality management.

Purpose of the Study:

• To develop a novel, computationally efficient deep learning method (DeepCTM) to mimic CTM simulations for O3 prediction.
• To enhance the speed and effectiveness of O3 pollution control strategy design.

Main Methods:

• Developed DeepCTM, a deep learning model trained to replicate O3 concentrations simulated by CTMs.
• Input features included precursor emissions, meteorological factors, and initial conditions.
• Validated DeepCTM's scientific reasonableness against known atmospheric chemistry mechanisms.

Main Results:

• DeepCTM accurately reproduces CTM-simulated O3 concentrations.
• The model efficiently identifies key contributors to O3 formation and quantifies responses to emission and meteorological changes.
• Analysis in China showed O3 levels are sensitive to initial O3, emissions, and daytime meteorology, including solar radiation.

Conclusions:

• DeepCTM offers a scientifically sound and highly efficient tool for representing complex atmospheric systems.
• This method provides crucial, timely information for policymakers to design effective O3 pollution mitigation strategies.