An efficient long-term daily PM_2.5 concentration prediction method based on decomposition-ensemble and recursive...

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Accurate fine particulate matter (PM2.5) prediction is crucial for public health, travel, and environmental policy.
• Existing PM2.5 prediction methods often neglect the analysis of prediction error sequences.

Purpose of the Study:

• To develop an advanced PM2.5 concentration prediction method utilizing decomposition-ensemble and Recursive Error Correction (REC).
• To address the limitations of current methods by systematically analyzing and utilizing PM2.5 prediction error characteristics.

Main Methods:

• Proposed a novel PM2.5 prediction method combining decomposition-ensemble modeling with Recursive Error Correction (REC).
• Incorporated the Olive-Hawkins estimator with extreme learning machines to enhance robustness against outliers in error sequences.
• Applied the method to forecast daily average PM2.5 concentrations in Beijing for 1, 5, and 10 days.

Main Results:

• The prediction error sequence of PM2.5 concentrations exhibits approximate normality and predictability.
• The REC mechanism effectively suppresses error accumulation, improving prediction accuracy.
• The proposed model demonstrated superior long-term forecasting performance compared to existing methods, with accuracy improvements exceeding 40%.

Conclusions:

• The developed method significantly enhances the accuracy of long-term PM2.5 forecasting.
• The Recursive Error Correction mechanism, robust to outliers, is key to improved prediction.
• This approach offers valuable tools for environmental pollution prevention and control strategies.