Design and Evaluation of a Multi-Model Weather Impact Translation System with Forecast Confidence

Air traffic management decisions are facilitated by translating convective weather mosaics into operational impact metrics such as airspace blockage, permeability, or sustainable traffic flow rates. As one example, the Route Availability Planning Tool (RAPT), developed in the 2000s and deployed in four terminal areas in the U.S., translates the Corridor Integrated Weather System (CIWS) 30 min forecast into a discrete green/yellow/red blockage impact metric along terminal departure routes. Modern machine learning techniques now enable significant enhancement of weather translation by synthesizing multiple deterministic, ensemble, and probabilistic forecast products and providing continuous permeability and flow rate impact scales that include quantitative measures of forecast confidence. A range of research issues must be resolved, however, to apply these techniques in an effective manner, including defining appropriate weather impact metrics, accommodating tradeoffs between controller workload and varying traffic flow rates, and understanding how to convey and use forecast confidence information in operational decisions.

The Traffic Flow Impact (TFI) capability was developed and extended between 2014 – 2019 to serve as a testbed with which to explore these issues. TFI uses a supervised learning approach to translate four convective weather forecast products into airspace permeability and flow rate metrics in more than 50 regions in the U.S. with a 12hr forecast horizon. Forecast confidence intervals are also displayed based on the learned skill of the component models. TFI has been deployed since 2014 as part of the Consolidated Storm Prediction for Aviation (CoSPA) research system and can be accessed at many FAA facilities and airlines.

This presentation will outline key issues related to machine learning applied to multi-model weather impact forecasting and the display of confidence information for air traffic management. Lessons from recent field evaluations and their implications for this type of forecast system will be discussed.

DISTRIBUTION STATEMENT A. Approved for public release. Distribution is unlimited. This material is based upon work supported by the Federal Aviation Administration under Air Force Contract No. FA8702-15-D-0001. Any opinions, findings, conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Federal Aviation Administration.