Improvements of response surface modeling with self-adaptive machine learning method for PM_2.5 and O₃ predictions

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Computational Modeling

Background:

• Accurate quantification of particulate matter (PM2.5) and ozone (O3) responses to precursor emissions is crucial for effective air quality control policies.
• Traditional polynomial function-based response surface models (pf-RSM) offer rapid predictions but suffer from computational overload and marginal effects, leading to prediction errors under strict emission controls.

Purpose of the Study:

• To introduce a novel Self-Adaptive Response Surface Model (SA-RSM) designed to enhance the computational efficiency and predictive accuracy of PM2.5 and O3 responses to precursor emissions.
• To address the limitations of pf-RSM, specifically computational burden and marginal effects, for improved air quality modeling.

Main Methods:

• Developed SA-RSM by integrating machine learning-based stepwise regression for robust model establishment and collinearity diagnosis to mitigate overfitting and marginal effects.
• Compared the performance of SA-RSM against the conventional pf-RSM using a pilot study case.

Main Results:

• SA-RSM reduced the required training dataset size by 70% for PM2.5 and 40% for O3, and decreased fitting time by 40% and 52% respectively, compared to pf-RSM.
• SA-RSM achieved a significant reduction in prediction errors: 49% for PM2.5 and 74% for O3.
• Isopleths generated by SA-RSM closely matched those from chemical transport models (CTM), indicating successful mitigation of marginal effects.

Conclusions:

• The SA-RSM demonstrates superior computational efficiency and prediction performance over pf-RSM for modeling air pollutant responses.
• SA-RSM effectively addresses the marginal effect issue, providing more reliable predictions, especially under stringent emission control scenarios.
• SA-RSM is proposed as a valuable scientific tool for policymakers to formulate effective PM2.5 and O3 emission control strategies.