A novel framework for high resolution air quality index prediction with interpretable artificial intelligence and uncertainties estimation.

Area of Science:

• Environmental Science
• Data Science
• Atmospheric Chemistry

Background:

• Accurate air quality index (AQI) prediction is crucial for environmental management.
• Existing models often overlook uncertainty estimation and output constraints.
• Forecasting AQI for multiple cities requires robust and adaptable methodologies.

Purpose of the Study:

• To propose a novel hybrid model, TMSSICX, for multi-city AQI forecasting.
• To incorporate uncertainty estimation and output constraints into AQI prediction.
• To enhance the accuracy and reliability of air quality predictions.

Main Methods:

• Time-varying filtered based empirical mode decomposition (TVFEMD) for sequence decomposition.
• Multi-scale fuzzy entropy (MFE) for component complexity analysis and clustering.
• Successive variational mode decomposition (SVMD) for high-frequency portion volatility reduction.
• SOABiLSTM and ICatboost for modeling decomposed components.
• XGBoost for ensembling sub-models and AKDE for interval estimation.

Main Results:

• The TMSSICX model outperformed 23 other models across all datasets.
• XGBoost ensembling reduced RMSE by 8.73% compared to SVM.
• SHAP analysis identified PM2.5 and PM10 as key drivers of long-term AQI trends.

Conclusions:

• The TMSSICX model offers a significant advancement in AQI prediction accuracy and reliability.
• The hybrid approach effectively handles data complexity and volatility.
• The findings provide valuable guidance for effective air quality management strategies.