Statistical and machine learning methods for evaluating trends in air quality under changing meteorological...

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Assessing anthropogenic emission impacts on air quality necessitates accounting for meteorological variability.
• Statistical methods, including multiple linear regression (MLR), are commonly employed to isolate emission-driven pollution trends.
• The efficacy of these statistical approaches in correcting for meteorological influences remains uncertain, hindering policy applications.

Purpose of the Study:

• To evaluate the performance of MLR and other quantitative methods in correcting for meteorological variability in air quality trend analysis.
• To assess the impact of anthropogenic emission changes in the US and China on PM2.5 and O3 concentrations.
• To develop and recommend improved statistical approaches for evaluating emission impacts on air quality.

Main Methods:

• Utilized simulations from the GEOS-Chem chemical transport model as a synthetic dataset.
• Compared the performance of multiple linear regression (MLR) with a random forest model.
• Designed a correction method using GEOS-Chem simulations with constant emission inputs.

Main Results:

• Widely used regression methods demonstrated poor performance in correcting for meteorological variability and identifying emission-related pollution trends.
• A random forest model, incorporating local and regional meteorological features, reduced estimation errors by 30%-42%.
• Quantified the inseparability of anthropogenic emissions and meteorological influences due to process-based interactions.

Conclusions:

• Standard statistical methods are insufficient for accurately evaluating the impact of emission changes on air quality.
• Random forest models offer a more robust approach to disentangling meteorological effects from emission trends.
• Recommendations are provided for enhancing statistical methods used in air quality policy evaluations.