Ozone concentration prediction using machine learning models integrated with 115 volatile organic compound...

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Machine Learning

Background:

• High-quality ozone concentration prediction is vital for effective environmental policy development.
• Volatile organic compounds (VOCs) are key precursors to ozone formation, and their monitoring data integration can enhance predictive models.
• Previous attempts using raw VOC data showed limited improvement in ozone prediction accuracy.

Purpose of the Study:

• To assess the impact of VOC monitoring data and dimensionality reduction on ozone prediction accuracy using machine learning models.
• To develop and evaluate a novel dimensionality reduction module (SARFE) for improving ozone forecasting.
• To compare the performance of different machine learning models integrated with the SARFE module.

Main Methods:

• Utilized monitoring data from 115 VOC species.
• Evaluated four machine learning models: LSTM, XGBoost, RF, and LightGBM.
• Developed and applied a novel module, SARFE (Statewide Air Pollution Research Center mechanisms + Recursive Feature Elimination), for dimensionality reduction.

Main Results:

• The SARFE module significantly enhanced the performance of all evaluated machine learning models.
• The SARFE-LSTM model demonstrated superior performance, achieving R² values from 0.78 to 0.92.
• SARFE-LSTM showed substantial improvements in predicting high ozone concentrations compared to traditional LSTM, with R², MAE, and RMSE improvements ranging from 36.54%-200.00%, 5.66%-18.62%, and 9.42%-22.45%, respectively.

Conclusions:

• Dimensionality reduction techniques, specifically the SARFE module, are crucial for effectively leveraging VOC monitoring data in machine learning models for ozone prediction.
• The SARFE-LSTM model offers a significant advancement in predicting ozone concentrations, particularly high levels, outperforming traditional LSTM.
• Historical ozone levels and isoprene were identified as key predictors, emphasizing the importance of precursor monitoring for accurate forecasting.