Analysis of Properties of Biomass Burning Aerosols using Numerical Techniques
Loading...
Authors
Lu, Siying
Issue Date
2025
Type
Dissertation
Language
en_US
Keywords
Air Pollutant , Machine Learning , Particle Size Distribution , PM2.5 , Positive Matrix Factorization , Wildfire Emissions
Alternative Title
Abstract
The increasing frequency and intensity of wildfires in the Western U.S. have heightened concerns over the environmental and health impacts of wildfire-emitted aerosols. These aerosols contribute substantially to atmospheric particulate matter (PM), with their particle size distribution (PSD) playing a critical role in radiative forcing and respiratory health. However, distinguishing wildfire-related aerosols, especially under moderate or weak smoke conditions, remains challenging. Traditional source apportionment methods, such as Positive Matrix Factorization (PMF), have been widely used to analyze aerosol sources, yet their application to PSD data is limited. Additionally, machine learning (ML) techniques provide new opportunities for improving the estimation of PM, but their use in identification and classification of wildfire aerosols remains underexplored.To address these challenges, satellite imagery, hazard mapping systems, and wind back-trajectory analysis were used to identify wildfire smoke-influenced days in Reno, Nevada (NV). PSD data and air pollutant concentrations were analyzed to differentiate wildfire-related aerosols from other sources. PMF was used to quantify the contributions of wildfire emissions to air pollution, while aerosol sources were classified using k-means clustering and key factors influencing PSD variability under wildfire smoke impacting were examined. Finally, a Random Forest classification system (RFCS) was developed and trained using k-means clustering results to detect hourly wildfire smoke influence based on common air pollutant data.
The results indicate that wildfire smoke significantly increases aerosol particle size and contributes up to 65% of PM2.5 concentrations during wildfire seasons (July-September) and up to 47% on an annual basis in Reno, NV. PMF analysis effectively identified wildfire-related aerosols, while k-means clustering revealed that using PSD data and air pollutant concentrations can both successfully distinguish wildfire-influenced hours. Key factors influencing wildfire aerosol PSD variability were demonstrated. The RFCS successfully detected wildfire-impacted hours, even in cases of light smoke influence, demonstrating its potential for real-time air quality assessments. These findings advance our understanding of wildfire aerosol properties and highlight the effectiveness of numerical techniques for improving air pollution monitoring and mitigation strategies.