Forecasting short-term peak concentrations from a network of air quality instruments measuring PM_2.5 using boosted...

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Fine particulate matter (PM2.5) poses significant health risks.
• Accurate short-term forecasting of PM2.5 pollution peaks is challenging.
• Understanding precursors is crucial for effective air quality management.

Purpose of the Study:

• To develop and validate machine learning models for forecasting short-term PM2.5 peaks.
• To identify key meteorological and chemical precursors of PM2.5 pollution events.
• To assess the utility of these forecasts for public health interventions.

Main Methods:

• Utilized a gradient boosted machine with a binary classifier for peak prediction.
• Incorporated meteorological data from personal weather stations and measurement instruments.
• Developed separate models for short-term (hourly) and long-term (daily) averaged peaks.

Main Results:

• Achieved 80-90% accuracy in forecasting PM2.5 peaks within the next 60 minutes.
• Identified elevated nitrogen monoxide, nitrogen dioxide, lower temperatures, and wind gusts as significant precursors.
• Distinguished precursors for short-term (wind gusts, nitrogen oxides) and long-term (air temperature, atmospheric pressure) peaks.

Conclusions:

• Machine learning provides a reliable method for short-term PM2.5 peak forecasting.
• Local-scale variations in PM2.5 can be identified using networked data.
• The developed system can serve as an effective short-term warning system for exposure mitigation.