Multiresolution Analysis of HRRR Meteorological Parameters and GOES-R AOD for Hourly PM_2.5 Prediction

Area of Science:

• Environmental Science
• Atmospheric Science
• Data Science

Background:

• Accurate estimation of fine particulate matter (PM2.5) exposure is vital for assessing human health impacts.
• Satellite remote sensing data improve PM2.5 models but face challenges like data gaps and limited ground monitor availability.
• Nonlinear relationships between PM2.5 and meteorological factors complicate predictive modeling.

Purpose of the Study:

• To develop a high-resolution, reliable PM2.5 prediction model for California.
• To address data gaps in satellite-retrieved aerosol optical depth (AOD) and improve model performance.
• To enhance PM2.5 forecasting by integrating gap-filled AOD and preprocessed meteorological data.

Main Methods:

• Filled spatial gaps in Geostationary Operational Environmental Satellite-16 AOD using Goddard Earth Observing System Composition Forecasting AOD.
• Preprocessed meteorological data, including temperature from the High-Resolution Rapid Refresh model, using Daubechies wavelet.
• Ingested gap-filled AOD and processed meteorological data into a machine learning model for hourly PM2.5 prediction at 1 km resolution.

Main Results:

• The machine learning model achieved high reliability with an out-of-bag R² of 0.86 and RMSE of 9.27 μg/m³.
• Spatial cross-validation yielded an R² of 0.82 and RMSE of 9.82 μg/m³.
• The model demonstrated strong performance in predicting ambient PM2.5, particularly for California.

Conclusions:

• The developed model provides a reliable method for high-resolution PM2.5 estimation in California.
• This approach effectively overcomes limitations of traditional PM2.5 models, including data gaps and sparse ground monitoring.
• The model is especially valuable for regions prone to wildfires, offering crucial air quality insights.