Deep Citywide Multisource Data Fusion-Based Air Quality Estimation.

Area of Science:

• Environmental Science
• Data Science
• Artificial Intelligence

Background:

• Air pollution is a growing concern, yet comprehensive air quality data is often limited due to sparse monitoring stations.
• Current air quality estimation methods typically analyze data from limited regions and estimate each area independently.

Purpose of the Study:

• To propose a deep citywide multisource data fusion-based air quality estimation (FAIRY) method.
• To enable real-time, comprehensive air quality assessment across entire urban areas.

Main Methods:

• FAIRY constructs citywide data into images (meteorology, traffic, emissions, POI, air quality) and uses SegNet for feature extraction.
• A self-attention mechanism fuses multiresolution features, while residual connections refine low-resolution features with high-resolution ones.
• Tobler's first law of geography is incorporated to leverage spatial correlations between adjacent regions.

Main Results:

• FAIRY demonstrated state-of-the-art performance on the Hangzhou city dataset.
• The proposed method achieved a 15.7% improvement in Mean Absolute Error (MAE) compared to the best baseline.
• The model effectively integrates diverse data sources for accurate air quality prediction.

Conclusions:

• FAIRY offers a robust and accurate solution for citywide air quality estimation by leveraging deep learning and multisource data fusion.
• The method addresses the limitations of sparse monitoring networks and improves the spatial and temporal resolution of air quality information.
• This approach has significant implications for environmental monitoring and public health initiatives.