Air quality prediction using multi-source remote sensing data integration with hybrid deep learning framework.

Area of Science:

• Environmental Science
• Computer Science
• Remote Sensing

Background:

• Rising air pollution poses significant risks to public health and urban environments.
• Effective air quality monitoring and forecasting are essential for environmental management.
• Existing methods often struggle with the complexity and multi-faceted nature of air pollution.

Purpose of the Study:

• To develop a novel hybrid deep learning framework for accurate air quality prediction.
• To integrate multi-source remote sensing data with meteorological and ground observations.
• To enhance the reliability and performance of air quality forecasting models.

Main Methods:

• Introduction of the Multi-Modal Attention-based Spatio-Temporal Network (MAST-Net).
• Utilizing Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) for hybrid deep learning.
• Leveraging satellite data (Sentinel-5P, MODIS, Landsat-8), meteorological variables, and ground observations.
• Incorporating dynamic feature selection and uncertainty quantification.

Main Results:

• MAST-Net demonstrated superior performance compared to conventional approaches.
• Achieved Root Mean Square Error (RMSE) improvements of 23-31% for various air pollutants.
• Reached high correlation coefficients (0.91-0.94) for predicting PM2.5, PM10, NO₂, and O₃ concentrations.
• Validated performance across diverse geographic and seasonal conditions.

Conclusions:

• The proposed MAST-Net architecture offers a powerful tool for real-time air quality prediction.
• The hybrid deep learning approach effectively integrates multi-source data for enhanced forecasting.
• This framework holds significant promise for improving urban environmental management and public health protection.