Testing and validation of forecasts for weather-induced operating restrictions in air traffic management based on Machine Learning models

Met4Airports is a research and development project funded by the German Federal Ministry for Digital and Transport (BMDV), aiming at the prediction of relevant planning and control parameters of air traffic management (ATM) by means of artificial intelligence (AI). It focusses on the effects of selected weather phenomena such as thunderstorms, fog, and snowfall, as they pose a significant disturbance for air traffic, causing capacity constraints for airports and en-route and approach sectors. The predicted quantities at the airport are runway and airport capacities, delays of individual flights as well as average delays for varying timespans up to 24 hours lead-time. Furthermore, capacities of the flight sectors close to the airport are predicted. These impact predictions can be used to support and optimize decision-making processes in ATM and enhance the situational awareness of decision makers.

Throughout the development process, various machine learning (ML) models are examined, relying on both meteorological forecast products of Deutscher Wetterdienst and air traffic data of airport operators (Flughafen München GmbH and Fraport) and air traffic control (Deutsche Flugsicherung). A detailed insight into this more technical part of the project as well as results of feature and hyperparameter studies from the machine learning process are given by Koser et al. [1].

The other major part of the project contains the validation and testing of the new impact predictions based on two different approaches using historical data from 2021/22. For the first approach, five days with thunderstorm events at or in the surrounding of the Munich airport are selected. The situations are displayed in a dashboard including all relevant weather and flight information of the respective day together with the new impact forecasts. Air traffic controller examine the selected situations and assess the benefit of the impact forecasts. The second approach compares the impact forecasts with the already available information from the ATM process. Both calculations of long-term statistics as well as single day studies are considered.

While the final evaluation of the air traffic controller of the first approach will be available during the upcoming summer, first analyses of both approaches already indicates that the ML models provide viable impact predictions on the selected thunderstorm days. Advantages over the already available information from the A-CDM system are visible for both statistical and single-day analyses. However, it can also be seen, that in some cases the impact predictions do not provide any profit, which might be due to the deficiencies in the input data like wrong weather predictions or air traffic disturbing processes which are not captured by the system.

[1] Koser et al. Development and optimization of Machine Learning methods to predict weather-induced operating restrictions in air traffic management, OSA1.9 Machine Learning in Weather and Climate, submitted for EMS 2023