Constructing and Evaluating Predictors for Data-Driven PM2.5 Forecasting Models

• Published in: International Journal of Environmental Research Vol. 19; no. 3
• Main Authors: He, Ran-Ran, Chen, Yu-Qiao, Tian, Lei, Shan, Lei, Sang, Xiao-Shuang
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.06.2025; Springer Nature B.V
• Subjects: Air pollution; Air quality; Annual cycles; Cities; Earth and Environmental Science; Environment; Environmental Engineering/Biotechnology; Environmental Management; Environmental research; Evaluation; Feature extraction; Forecasting; Geoecology/Natural Processes; Landscape/Regional and Urban Planning; Learning algorithms; Long short-term memory; Machine learning; Meteorological parameters; Natural Hazards; Outdoor air quality; Particulate matter; Permutations; Pollutants; Regression analysis; Research Paper; Time series; Warning systems; Air quality; Data-driven model; Feature engineering; PM; Machine learning
• ISSN: 1735-6865; 2008-2304
• DOI: 10.1007/s41742-025-00767-x

Data-driven methods for air quality forecasting are beneficial to early-warning of air pollution, especially in small and medium-sized cities. In this study, the Prophet model is used for extracting time-related features, such as trend, annual cycles, and weekly cycles, from the PM 2.5 series, and the gradient boosting regression trees (GBM) and long short-term memory networks (LSTM) are used to build forecasting models utilizing predictors including time-related features. All predictors are grouped into time-related features, antecedent air pollutant concentration, and meteorological parameters, and permutation feature importance (PFI) is used for measuring the importance of these feature groups. A series of forecasting experiments on a middle city in China are carried out, and the results are analyzed for a better understanding of the mechanisms of the forecasting skill. The result shows that time-related features extracted by Prophet can enhance the forecasting skill significantly, and this advantage is more apparent when the forecasting horizon is large. PM 2.5 time series has short memory and a rapid response to changes in meteorological conditions, and only meteorological conditions in the previous 2 days have useful signals for forecasting. The rapid response and short memory of PM 2.5 make LSTM show no advantage over GBM in this case study. We show that feature engineering is important for forecasting PM 2.5 , and the time-related features and antecedent pollutant concentrations can be seen as proxy variables for some variables that are difficult to obtain, such as pollutant emission and missing meteorological conditions. Highlights We summarize the information that can be used for building PM 2.5 forecasting models and provide a detailed explanation of the three classes of information. We show that feature engineering is important for forecasting PM 2.5 based on machine learning methods. Especially, the technology of time series analysis (the Prophet model used in this study) can be a useful tool for extracting features as inputs to machine learning models. Most times, the evaluation of features (predictors) is based on single features. In this study, we extend permutation feature importance to quantify the contribution of feature groups to the forecasting model. We think that the advantages of LSTM should be assessed carefully. As we have shown in the manuscript, the response of PM 2.5 to changes in meteorological conditions is rapid, which makes LSTM show less advantage over GBM as LSTM is designed to capture long-term dependencies.