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

The air transport system is currently in a rapid development stage, accurate forecasting emissions is critical for identifying and mitigating its environmental impact. Accurate forecasting depends not only on temporal features from historical air traffic data but also on the influence of spatial factors. This paper proposes a deep learning-based forecasting framework for en route airspace emissions. It combines three-channel networks: a graph convolutional network, a gated recurrent unit, and the attention mechanism, in order to extract the spatial, temporal, and global temporal dynamics trends, respectively. The model is evaluated with real-world datasets, and the experimental results outperform existing state-of-the-art benchmarks on different evaluation metrics and forecasting horizons in complex airspace networks. Our method provides an alternative for forecasting air traffic emissions using publicly available traffic flow data. Furthermore, we propose an extension index that can be taken as an early warning indicator for stakeholders to monitor air traffic emissions. [Display omitted] •Importance of considering temporal and spatial characteristics in air traffic forecasting is explained.•Nonlinear time characteristics and spatial heterogeneity in air traffic data are revealed.•Feasibility of using traffic flow to indirectly forecast air traffic emissions is demonstrated.•Practical application prospects of the air traffic emission forecasting model are illustrated.