Revealing Drivers of Haze Pollution by Explainable Machine Learning

Abstract

Many places on earth still suffer from a high level of atmospheric fine particulate matter (PM2.5) pollution. Formation of a particulate pollution event or haze episode (HE) involves many factors, including meteorology, emissions, and chemistry. Understanding the direct causes of and key drivers behind the HE is thus essential. Traditionally, this is done via chemical transport models. However, substantial uncertainties are introduced into the model estimation when there are significant changes in the emissions inventory due to interventions (e.g., the COVID-19 lockdown). Here we applied a Random Forest model coupled with a Shapley additive explanation algorithm, a post hoc explanation technique, to investigate the roles of major meteorological factors, primary emissions, and chemistry in five severe HEs that occurred before or during the COVID-19 lockdown in China. We discovered that, in addition to the high level of primary emissions, PM2.5 in these haze episodes was largely driven by meteorological effects (with average contributions of 30–65 μg m–3 for the five HEs), followed by chemistry (∼15–30 μg m–3). Photochemistry was likely the major pathway of formation of nitrate, while air humidity was the predominant factor in forming sulfate. Our results highlight that the machine learning driven by data has the potential to be a complementary tool in predicting and interpreting air pollution.

Understanding Sources and Drivers of Size-Resolved Aerosol in the High Arctic Islands of Svalbard Using a Receptor Model Coupled with Machine Learning

“…To investigate how specific processes drive the different aerosol factors, the SHAP approach was applied to interpret the RF models. 15 SHAP is a game theoretic approach that is able to fairly distribute the anomaly of the total concentration among different parameters 14 where x i , j is the value of feature j for sample i . K is the total number of different parameters.…”

Dramatic changes in atmospheric pollution source contributions for a coastal megacity in northern China from 2011 to 2020

“…The 500 predicted concentrations were then averaged to calculate the weather-normalized concentration. The RFbased weather normalization technique has been extensively used to decouple meteorology from the observed concentrations and thus can detect interventions in emissions over time (Grange et al, 2018;Grange and Carslaw, 2019;Hou et al, 2022).…”

Ozone formation sensitivity study using machine learning coupled with the reactivity of volatile organic compound species

“…Ma et al, 2021). Thus, it is a promising alternative to account for the effects of meteorology on air pollutants and has been intensively used in atmospheric studies (H. Hou et al, 2022).…”

“…It should be noted that physical interpretability of the results is an important question when ML models are applied in atmospheric studies (Hou et al, 2022). However, explanations of ML results (e.g., RI) are somewhat vague because ML is a "black-box" model from the point view of chemical mechanism (Hou et al, 2022;Taoufik et al, 2022).…”