PMF Coupled With Random Forest Method Reveals Detailed PM2.5 Sources and Their Synergistic Effects on Air Pollution Control

Abstract

Understanding the sources of emerging PM2.5 pollution is crucial for developing effective air quality management strategies. This study combines positive matrix factorization (PMF) with random forest (RF) classification to reveal a detailed PM2.5 source identification for both day and night samples collected in Tianjin in 2022, during winter (including the Beijing Olympics period) and summer. When resolving the overlap of organic compounds between combustion and collinear sources, this model achieved accuracy, precision, recall, and F1 scores in a range from 85% to 91% on the independent test data set. Additionally, scenario simulations are applied to investigate the impacts from air pollution control strategies and large‐scale events on different emission sources. This methodology demonstrates the potential of combining receptor models, machine learning, and chemical analysis to identify overlapping air pollution sources, when samples are limited and conventional tracers are not available for PMF. In general, our results enhance the discrimination of the primary contributors to emerging air pollution from both traditional energy and sources, which can further support more flexible and season‐specific pollution control policies.