Shadow Evaluation of Real-Time Machine Learning Services in the Houston Airspace

NASA is conducting a series of field evaluations between 2022 through 2030 to develop and demonstrate, in an operational environment, new technologies supporting efficient airspace operations. As part of the evaluations, NASA deployed the Machine Learning Airport Surface Model to enable predeparture Trajectory Option Set rerouting in the North Texas airspace. After a successful field evaluation in North Texas, the Houston airspace was selected as a new location to validate scalability of the approach and benefits. This paper provides shadow mode validation results of the Machine Learning Airport Surface Model running in the Houston airspace. Shadow mode consists of the system running passively in real-time while generating predictions for departures and arrivals, but without users acting on system recommendations. The shadow evaluation is an important step towards validation to ensure behavior of the system and machine learning algorithms running in real-time matches results generated in offline training.