Machine learning-based ozone and PM2.5 forecasting: Application to multiple AQS sites in the Pacific Northwest

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Pacific Northwest air quality is generally good but impacted by wildfires and wood burning, causing unhealthy events.
• Existing chemical transport models (CTMs) struggle to accurately forecast these specific air quality events in the region.
• Previous work demonstrated a machine learning (ML) system's reliability for ozone (O3) forecasting at a single site.

Purpose of the Study:

• To expand and evaluate an ML-based air quality forecasting system for both O3 and PM2.5 across the Pacific Northwest.
• To compare the performance of the ML system against the current operational CTM-based forecasts.
• To assess the ML system's effectiveness during wildfire and cold seasons for multiple pollutants.

Main Methods:

• Developed an ML forecasting system comprising ML1 (random forecast classifiers, multiple linear regression) and ML2 (two-phase random forest regression).
• Applied the ML system to predict O3 during wildfire seasons and PM2.5 during wildfire and cold seasons across available PNW monitoring sites (2017-2020).
• Evaluated ML forecasts against observations and compared them with CTM-based forecasts using metrics like R², Normalized Mean Bias (NMB), and Normalized Mean Error (NME).

Main Results:

• ML forecasts showed significant improvements over CTMs: O3 NMB reduced from 7.6% to 2.6%, NME from 18% to 12%.
• ML2 demonstrated superior PM2.5 forecasting, significantly reducing NMB and NME for both wildfire and cold seasons.
• The ML system captured more high-pollution events for both O3 and PM2.5 compared to CTMs.

Conclusions:

• The developed ML air quality forecasting system provides more reliable predictions than CTM-based forecasts for O3 and PM2.5 in the PNW.
• The ML system requires fewer computational resources and input datasets, making it a cost-effective solution.
• This low-cost, reliable ML system can effectively support regional and local air quality management efforts.