Accurate PM_2.5 urban air pollution forecasting using multivariate ensemble learning Accounting for evolving target...

Area of Science:

• Environmental Science and Public Health
• Data Science and Machine Learning

Background:

• Air pollution, particularly fine particulate matter (PM2.5), poses significant global environmental and public health risks.
• Accurate PM2.5 forecasting is crucial for mitigation but remains challenging due to data non-stationarity and complex influencing factors.

Purpose of the Study:

• To develop an effective multivariate multi-step ensemble machine learning model for continuous 24-hour PM2.5 concentration prediction.
• To enhance PM2.5 forecasting accuracy in Ho Chi Minh City (HCMC) by considering diverse spatial and temporal factors.

Main Methods:

• Proposed a multivariate multi-step ensemble machine learning model incorporating meteorological conditions, PM2.5 rolling means, and temporal features.
• Established six real-time air quality monitoring sites across diverse areas (traffic, residential, industrial) in HCMC.
• Utilized statistical methods for comprehensive model performance evaluation.

Main Results:

• The proposed model demonstrated strong performance, significantly improving forecasting accuracy over existing HCMC PM2.5 models.
• Analysis identified the contribution of different feature groups to the model's enhanced predictive capabilities.
• Generated station-specific forecasting results, reflecting localized air quality variations.

Conclusions:

• The developed ensemble machine learning model offers a robust solution for accurate PM2.5 forecasting.
• Findings provide valuable insights for public health advisories and citizen travel planning in urban environments.
• The model's performance highlights the importance of multivariate and multi-site data in air quality prediction.