Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Handling Weather Uncertainty in Air Traffic Prediction through an Inverse Approach

G. Lancia, D. Falanga, S. Alam, G. Lulli

TL;DR

This paper tackles the problem of forecasting weather-induced flight reroutes with long lead times by proposing a 3-D Gaussian Mixture Model within a Mixture Density Network framework. The approach uses a two-branch architecture to fuse high-resolution convective weather maps and wind fields with traffic data, enabling probabilistic forecasts of altitude, latitude, and longitude up to 60 minutes ahead. Key contributions include a mathematically grounded 3-D MDN, a robust two-branch architecture with a SPD covariance constraint, wavelet and power-transform pre-processing, and an explainability framework based on Vanilla Gradient saliency maps. The model achieves high accuracy (MAPE $<0.02$, $R^2$ near 0.99 across leads) and offers substantial training-time efficiency (≈4 h vs ≈15 h) with promising prospects for real-time ATM deployment and extension to additional airspaces and data sources.

Abstract

Adverse weather conditions, particularly convective phenomena, pose significant challenges to Air Traffic Management, often requiring real-time rerouting decisions that impact efficiency and safety. This study introduces a 3-D Gaussian Mixture Model to predict long lead-time flight trajectory changes, incorporating comprehensive weather and traffic data. Utilizing high-resolution meteorological datasets, including convective weather maps and wind data, alongside traffic records, the model demonstrates robust performance in forecasting reroutes up to 60 minutes. The novel 3-D Gaussian Mixture Model framework employs a probabilistic approach to capture uncertainty while providing accurate forecasts of altitude, latitude, and longitude. Extensive evaluation revealed a Mean Absolute Percentage Error below 0.02 across varying lead times, highlighting the model's accuracy and scalability. By integrating explainability techniques such as the Vanilla Gradient algorithm, the study provides insights into feature contributions, showing that they contribute to improving Air Traffic Management strategies to mitigate weather-induced disruptions.

Handling Weather Uncertainty in Air Traffic Prediction through an Inverse Approach

TL;DR

This paper tackles the problem of forecasting weather-induced flight reroutes with long lead times by proposing a 3-D Gaussian Mixture Model within a Mixture Density Network framework. The approach uses a two-branch architecture to fuse high-resolution convective weather maps and wind fields with traffic data, enabling probabilistic forecasts of altitude, latitude, and longitude up to 60 minutes ahead. Key contributions include a mathematically grounded 3-D MDN, a robust two-branch architecture with a SPD covariance constraint, wavelet and power-transform pre-processing, and an explainability framework based on Vanilla Gradient saliency maps. The model achieves high accuracy (MAPE , near 0.99 across leads) and offers substantial training-time efficiency (≈4 h vs ≈15 h) with promising prospects for real-time ATM deployment and extension to additional airspaces and data sources.

Abstract

Adverse weather conditions, particularly convective phenomena, pose significant challenges to Air Traffic Management, often requiring real-time rerouting decisions that impact efficiency and safety. This study introduces a 3-D Gaussian Mixture Model to predict long lead-time flight trajectory changes, incorporating comprehensive weather and traffic data. Utilizing high-resolution meteorological datasets, including convective weather maps and wind data, alongside traffic records, the model demonstrates robust performance in forecasting reroutes up to 60 minutes. The novel 3-D Gaussian Mixture Model framework employs a probabilistic approach to capture uncertainty while providing accurate forecasts of altitude, latitude, and longitude. Extensive evaluation revealed a Mean Absolute Percentage Error below 0.02 across varying lead times, highlighting the model's accuracy and scalability. By integrating explainability techniques such as the Vanilla Gradient algorithm, the study provides insights into feature contributions, showing that they contribute to improving Air Traffic Management strategies to mitigate weather-induced disruptions.

Paper Structure

This paper contains 17 sections, 20 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Data
  3. Data acquisition
  4. Dataset Creation
  5. Methodology
  6. Model's Mathematical Foundations
  7. Model's Architecture
  8. Model's Pre-processing
  9. Model's Selection
  10. Results
  11. Explaining the predictions
  12. Conclusion
  13. Mathematical Details
  14. Consistency of predictions
  15. Wavelet Pre-processing
  16. ...and 2 more sections

Figures (4)

  • Figure 1: A schematic representation of the
  • Figure 2: Double y-axis plot with the validation metrics for the at various lead-times. Red dots denote the average $R^{2}$ coefficient, while the blue ones denote the average . Error bars were evaluated as the standard error mean. On the x-axis the values of the lead time, while along the vertical axis the values attained by the metrics.
  • Figure 3: -based saliency maps expressing the average percentage of importance explained by each feature. Each saliency map refers to one specific prediction attribute (i.e., latitude, longitude, and altitude). The importance is here expressed as the magnitude of the gradients. On the x-axis, the features are reported. On the y-axis, the estimated importance levels. Uncertainty, estimated using the , is negligible and thus not visually discernible.
  • Figure 4: Average fraction of energy explained by reconstructing weather images from deepest wavelet level. 95% Confidence intervals were reported, but they are too small compared to the data points on the graph. The dotted line denotes the 90% of energy explained.