Hour by Hour PM_2.5 Mapping Using Geostationary Satellites

Area of Science:

• Atmospheric Science
• Environmental Science
• Data Science

Background:

• Accurate estimation of ground-level fine particulate matter (PM2.5) is crucial for public health and environmental monitoring.
• Existing methods often lack the spatial and temporal resolution needed to capture dynamic PM2.5 events.

Purpose of the Study:

• To develop and evaluate a Deep Neural Network (DNN) model for estimating surface PM2.5 concentrations across the contiguous United States (CONUS).
• To assess the added value of integrating geostationary satellite data, specifically TEMPO (Tropospheric Emissions: Monitoring of Pollution), into PM2.5 estimation models.
• To compare the performance of DNN with other machine learning (ML) models like Random Forest and Light Gradient-Boosting Machine.

Main Methods:

• Utilized geostationary satellite-derived Aerosol Optical Depth (AOD) and radiance measurements.
• Incorporated meteorological parameters from the High-Resolution Rapid Refresh (HRRR) model.
• Employed a Deep Neural Network (DNN) model, comparing its performance against Random Forest and Light Gradient-Boosting Machine models using AirNow PM2.5 measurements.
• Enhanced the DNN model by integrating TEMPO Level 1b (L1b) data.

Main Results:

• The DNN model, even without TEMPO data, outperformed other ML models in estimating PM2.5 concentrations, showing significant improvements in Index of Agreement (IOA) and relative Root-Mean-Square Error (rRMSE).
• Hourly estimated PM2.5 concentrations closely matched observed data in temporal trends and spatial distribution over eastern CONUS.
• Integrating TEMPO L1b data into the DNN model further improved performance, indicated by an 8% higher R-squared and a 25% lower rRMSE.
• The most substantial improvements were observed during high smoke events, demonstrating TEMPO's capability to capture high PM2.5 concentrations.

Conclusions:

• Deep Neural Network models, especially when enhanced with geostationary satellite data like TEMPO, provide a robust framework for high-resolution PM2.5 estimation.
• The integration of TEMPO L1b spectrally resolved radiance data offers a novel approach to monitor PM2.5 dynamics during wildfire events.
• This study establishes a scalable framework for combining multi-satellite data with meteorological models to improve real-time air quality assessments.