Deep learning framework for forecasting en route airspace emissions considering temporal-spatial correlation.

Area of Science:

• Environmental Science
• Computer Science
• Aerospace Engineering

Background:

• Accurate forecasting of air transport emissions is vital for environmental impact assessment and mitigation strategies.
• Forecasting air traffic emissions requires considering both temporal patterns in historical data and spatial influences.
• Existing methods may not fully capture the complex dynamics of emissions in evolving air transport systems.

Purpose of the Study:

• To develop a novel deep learning framework for accurate forecasting of en route airspace emissions.
• To integrate spatial and temporal features effectively for improved emission prediction.
• To provide an early warning indicator for monitoring air traffic emissions.

Main Methods:

• A three-channel deep learning network combining a graph convolutional network (GCN), a gated recurrent unit (GRU), and an attention mechanism.
• The GCN extracts spatial dynamics, the GRU captures temporal dependencies, and the attention mechanism models global temporal trends.
• Evaluation using real-world air traffic datasets in complex airspace networks.

Main Results:

• The proposed deep learning framework significantly outperforms existing state-of-the-art benchmarks.
• The model demonstrates superior performance across various evaluation metrics and forecasting horizons.
• The framework successfully extracts spatial, temporal, and global temporal dynamics for accurate emission forecasting.

Conclusions:

• The developed deep learning model offers a robust and accurate alternative for forecasting air traffic emissions.
• The framework effectively utilizes publicly available traffic flow data for emission prediction.
• An extension index is proposed as a valuable early warning tool for stakeholders monitoring air traffic emissions.