En
期刊文章
文章精选
当期目录
过刊浏览
高被引
高下载
特邀文章
专刊合集
待刊文章
Volume 117
您当前的位置：首页 > 期刊文章 > 过刊浏览 > Volumes 108-119 (2025) > Volume 117
Visibility drivers and light extinction apportionment in Urumqi: Integrating chemical reconstruction with machine learning
Wenyu Bai a b, Yingying Liu a b, Xueyan Zhao a *, Chunmei Geng a *, Wen Yang a
a State Key Laboratory of Environmental Criteria and Risk Assessment, Chinese Research Academy of Environmental Sciences, Beijing, 100012, China
b NEAC Key Laboratory of Ecology and Environment in Minority Areas, College of Life and Environmental Sciences, Minzu University of China, Beijing, 100081, China
10.1016/j.partic.2026.08.014
Volume 117, October 2026, Pages 366-379
Received 11 April 2026, Revised 8 August 2026, Accepted 11 August 2026, Available online 12 August 2026, Version of Record 20 August 2026.
E-mail: zhaoxy@craes.org.cn; gengcm@craes.org.cn
Highlights

• Reveals relative humidity as the dominant meteorological predictor of visibility (mean |SHAP| = 4.49).

• Quantifies winter light extinction dominated by secondary (NH4)2SO4 and NH4NO3 (combined > 90%).

• Provides the first machine learning based visibility source apportionment in arid western China.


Abstract

Identifying the key factors influencing atmospheric visibility is essential for improving urban air quality. This study analyzed high time-resolution observational data from Urumqi, an arid oasis city in northwestern China, from May 2024 to April 2025. PM2.5 sources were apportioned using Positive Matrix Factorization (PMF), and the revised IMPROVE algorithm and XGBoost-SHAP (Shapley Additive Explanations) model were applied to identify the chemical components and source categories responsible for visibility impairment, respectively. Monthly visibility peaked in June (33 km) and reached its lowest in December (5 km). The IMPROVE algorithm identified ammonium sulfate and ammonium nitrate as the dominant contributors to light extinction, with their combined contribution exceeding 90% in winter. The XGBoost-SHAP model identified relative humidity (RH) as the most critical predictor of visibility (mean |SHAP| = 4.49 km), followed by secondary sulfate, secondary organic aerosol (SOA), and vehicle emissions. Although ammonium nitrate contributed comparably to ammonium sulfate in the IMPROVE extinction budget, its predictive importance was lower in the machine learning model, reflecting the non-linear mediation of RH on sulfate-driven hygroscopic growth. These findings provide scientific evidence for targeted air pollution control in arid urban environments.

Graphical abstract
Keywords
Visibility; Sulfate; IMPROVE algorithm; XGBoost-SHAP model; Urumqi
文章导读