A statistical model for predicting PM_2.5 for the western United States

Area of Science:

• Environmental Science
• Atmospheric Science
• Data Science

Background:

• Wildfire smoke significantly impacts air quality and public health across the western United States annually.
• Accurate prediction of fine particulate matter (PM2.5) during wildfire events is crucial for mitigation and public advisement.

Purpose of the Study:

• To develop and evaluate a site-specific statistical model for predicting daily ground-level PM2.5 concentrations.
• To implement this model as an operational, continuously-updating forecast system for wildfire smoke events.

Main Methods:

• Utilized a multiple linear regression model incorporating previous day's PM2.5, satellite-derived fire variables (fire radiative power, National Fire Danger Rating System Energy Release Component), and smoke variables (aerosol optical depth, smoke plume perimeters).
• Adapted and modified a model developed for British Columbia to suit conditions in the United States.
• Developed a novel implementation method for an operational forecast system, integrating real-time data and updated remote sensing information.

Main Results:

• The statistical model explained an average of 78% of the variance in daily ground-level PM2.5 concentrations.
• The operational forecast system demonstrated improved performance with continuous updates and incorporation of early morning data.
• The model was successfully tested and utilized during the 2016 and 2017 wildfire seasons.

Conclusions:

• The developed statistical model provides reliable hourly predictions of PM2.5 levels during wildfire smoke events.
• The operational forecast system offers valuable tools for smoke incident specialists, public health officials, and air quality regulators.
• Predictions from this model are intended for public access, enhancing awareness and preparedness for wildfire smoke impacts.