An ensemble multi-scale framework for long-term forecasting of air quality.

Area of Science:

• Environmental Science
• Data Science
• Atmospheric Science

Background:

• Accurate long-term air quality forecasting is vital for public health and environmental policy.
• Previous forecasting methods face challenges in capturing long-range oscillations and error propagation.

Purpose of the Study:

• To develop and validate an ensemble multi-scale framework for improved long-term air quality forecasting.
• To enhance the accuracy and robustness of daily PM2.5 predictions.

Main Methods:

• Utilized modified empirical mode decomposition (EMD) for stable component extraction.
• Implemented an ensemble mechanism for improved nonstationary oscillation resampling (NSOR) to mitigate error propagation.
• Applied the framework to forecast daily PM2.5 concentrations at 14 Hong Kong monitoring stations.

Main Results:

• The proposed framework effectively captures long-term variations in air pollution.
• Achieved an 8.4% reduction in root-mean-square error and a 9.2% reduction in mean absolute error for 100-day forecasts.
• Demonstrated improved robustness for 180-day and 365-day forecasting scenarios.

Conclusions:

• The ensemble multi-scale framework significantly enhances long-term air quality forecasting performance.
• The method is robust and applicable to general long-term time series forecasting.
• Improved forecasting aids in better air pollution control strategies and health impact evaluations.