Application of the XGBoost Machine Learning Method in PM2.5 Prediction: A Case Study of Shanghai

• DOI: 10.4209/aaqr.2019.08.0408
• Publication Date: 2019-12-06T06:49:53Z
• Authors: Jinghui Ma; Zhongqi Yu; Yuanhao Qu; Jianming Xu; Yu Cao
• Keywords: 13. Climate action; 11. Sustainability; 01 natural sciences; 0105 earth and related environmental sciences

ABSTRACT Air quality forecasting is crucial to reducing air pollution in China, which has detrimental effects on human health. Atmospheric chemical-transport models can provide air pollutant forecasts with high temporal and spatial resolution and are widely used for routine air quality predictions (e.g., 1–3 days in advance). However, the model’s performance is limited by uncertainties in the emission inventory and biases in the initial and boundary conditions, as well as deficiencies in the current chemical and physical schemes. As a result, experimentation with several new methods, such as machine learning, is occurring in the field of air quality forecasting. This study combined hourly PM2.5 mass concentration forecasts from an operational air quality numerical prediction system (WRF-Chem) at the Shanghai Meteorological Service (SMS) with comprehensive near-surface measurements of air pollutants and meteorological conditions to develop a machine learning model that estimates the daily PM2.5 mass concentration in Shanghai, China. With correlation coefficients that are higher by 50–100% and a standard deviation that is lower by 14–24 µg m–3, the machine learning model provides significantly better daily forecasting of PM2.5 than the WRF-Chem model. Thus, this research offers a new technique for enhancing air quality forecasting in China.