A hybrid model for forecasting of particulate matter concentrations based on multiscale characterization and machine...

Area of Science:

• Environmental Science
• Data Science
• Atmospheric Chemistry

Background:

• Accurate prediction of particulate matter (PM) concentrations is crucial for public health and environmental monitoring.
• Traditional time series analysis faces challenges with the complex, nonlinear nature of air quality data.
• Advancements in machine learning (ML) and computational tools offer new possibilities for robust PM forecasting.

Purpose of the Study:

• To develop and validate a hybrid model for forecasting PM10 and PM2.5 concentrations.
• To evaluate the performance of Empirical Mode Decomposition (EMD) combined with various ML algorithms.
• To compare the efficacy of hybrid EMD-ML models against non-hybrid ML models for air quality prediction.

Main Methods:

• Utilized Empirical Mode Decomposition (EMD) to decompose PM time series into intrinsic mode functions (IMFs) for multiscale characterization.
• Developed hybrid models by applying individual ML algorithms (Random Forest, Support Vector Regressor, k-Nearest Neighbors, Feed Forward Neural Network, AdaBoost) to the decomposed IMFs.
• Validated models using air quality data from Makkah, Saudi Arabia, and Delhi, India, evaluating performance with RMSE, MAE, and MBE.

Main Results:

• The EMD-FFNN model achieved the lowest error rates for PM10 (RMSE=12.25, MAE=7.43) and PM2.5 (RMSE=4.81, MAE=3.02) in Makkah.
• For Delhi data, the EMD-kNN model showed the lowest PM10 error (RMSE=20.56, MAE=12.87), and EMD-AdaBoost yielded the lowest PM2.5 error (RMSE=15.29, MAE=9.45).
• Hybrid EMD-ML models demonstrated superior performance compared to non-hybrid ML models in forecasting PM concentrations.

Conclusions:

• The proposed hybrid EMD-ML models are effective for forecasting PM10 and PM2.5 mass concentrations.
• This approach can significantly enhance the development of rapid air quality warning systems.
• Multiscale characterization via EMD improves the accuracy of ML-based air quality prediction.