Estimating PM_2.5 Concentrations in the Conterminous United States Using the Random Forest Approach

Area of Science:

• Environmental Science
• Atmospheric Science
• Data Science

Background:

• Parametric regression models are common for estimating PM2.5, but nonparametric machine learning and national-scale models are less frequent.
• Accurate estimation of fine particulate matter (PM2.5) is crucial for public health and environmental monitoring.

Purpose of the Study:

• To develop and validate a national-scale random forest model for estimating daily ground-level PM2.5 concentrations in the conterminous United States.
• To assess the contribution of aerosol optical depth (AOD), meteorological fields, land use variables, and spatial features to PM2.5 estimation accuracy.

Main Methods:

• Developed a random forest model integrating aerosol optical depth (AOD) data, meteorological fields, and land use variables.
• Incorporated convolutional layers for land use terms and nearby PM2.5 measurements to enhance prediction.
• Utilized cross-validation (CV) to evaluate model performance, calculating R-squared, Mean Prediction Error (MPE), and Root Mean Squared Prediction Error (RMSPE).

Main Results:

• Achieved a cross-validation R-squared value of 0.80, with MPE of 1.78 μg/m³ and RMSPE of 2.83 μg/m³.
• The model's prediction accuracy is comparable to previous national and regional studies using neural networks or regression models.
• Convolutional layers for land use and nearby PM2.5 measurements significantly improved CV R-squared by approximately 0.02 and 0.06, respectively.
• Variable importance analysis highlighted the convolutional layer for nearby PM2.5 measurements and AOD values as key predictors.

Conclusions:

• The developed random forest model provides accurate and reliable daily estimates of ground-level PM2.5 concentrations nationwide.
• The integration of satellite-derived AOD, meteorological data, and advanced machine learning techniques, including convolutional layers, is effective for PM2.5 modeling.
• This approach offers a valuable tool for air quality assessment and management at a national scale.