Machine Learning-Based Integration of High-Resolution Wildfire Smoke Simulations and Observations for Regional Health...

Area of Science:

• Environmental Science
• Public Health
• Atmospheric Chemistry

Background:

• Large wildfires pose a growing threat to the western U.S., with the 2017 fire season causing extensive smoke events in the Pacific Northwest (PNW).
• Wildfire smoke contains fine particulate matter (PM2.5), which has significant public health implications.

Purpose of the Study:

• To evaluate the public health impacts of wildfire smoke in the PNW during the 2017 fire season.
• To improve estimates of PM2.5 exposure from wildfire smoke by integrating numerical modeling with observational data.

Main Methods:

• Utilized a coupled Weather Research and Forecasting (WRF) and Community Multiscale Air Quality (CMAQ) modeling system to simulate smoke transport.
• Integrated satellite-derived aerosol optical depth (AOD) and ground-level PM2.5 data using machine learning algorithms (ordinary multi-linear regression, generalized boosting, random forest).
• Applied a 10-fold cross-validation to assess the accuracy of PM2.5 estimation and used a short-term exposure-response function to estimate mortality.

Main Results:

• Data integration and bias correction, particularly with the random forest method, significantly improved surface PM2.5 estimations.
• An estimated 183 regional mortalities (95% CI: 0–432) were attributed to PM2.5 exposure during the 2017 smoke episode.
• Fire emissions accounted for 85% of PM2.5 pollution and 95% of the associated mortality in the PNW.

Conclusions:

• Wildfire smoke has profound public health impacts in the PNW, necessitating accurate exposure assessment.
• The study demonstrates the effectiveness of data fusion techniques in enhancing air quality modeling for wildfire events.
• A high-performance fire smoke forecasting and reanalysis system is crucial for mitigating public health risks in fire-prone regions.