Seasonally adaptive data-driven ozone prediction in megacity environments.

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Ground level ozone (O3) concentrations show persistent upward trends in urban megacities.
• O3 is formed via complex photochemical reactions sensitive to meteorological conditions, challenging traditional control strategies.
• Reductions in primary pollutants have not curbed O3 rise in urban environments.

Purpose of the Study:

• To develop and evaluate a comprehensive O3 prediction framework for megacity environments.
• To integrate meteorological variables, air pollutant concentrations, and traffic volume for improved O3 forecasting.
• To assess the performance of machine learning algorithms for O3 prediction and identify key predictive factors.

Main Methods:

• Utilized an eight-year hourly dataset from 37 administrative districts in Seoul, South Korea.
• Evaluated eight machine learning algorithms with hyperparameter tuning, selecting CatBoost as the top performer.
• Developed season-specific O3 prediction models to account for temporal variability.

Main Results:

• The CatBoost model achieved high accuracy (R2 = 0.93) in predicting O3 concentrations.
• Season-specific models demonstrated reduced prediction errors, with the winter model showing the highest accuracy (R2 = 0.96).
• Feature importance varied seasonally, with temperature and NO2 critical in warmer months, and wind speed, CO, and SO2 in winter.

Conclusions:

• A comprehensive, seasonally adaptive O3 prediction framework is effective for megacities.
• Static, year-round prediction approaches are limited due to seasonal variations in O3 formation.
• Integrating meteorological and anthropogenic factors is crucial for effective O3 control strategies.