Start descending to avoid conflicts, cut level-offs, and save fuel.

What is it about?

When the Top of Descent (TOD) is chosen without accounting for downstream traffic interactions, descents are often interrupted by level-offs to avoid conflicts. To support green aviation practices such as Continuous Descent Operations (CDO), this paper proposes a three-step learning model to predict TOD locations that enable interaction-free descent trajectories. The first step identifies critical areas—zones with interacting flights that disrupt descents—by analyzing non-CDO flights. The second step employs Confident Learning on both non-CDO and CDO data to predict CDO-capable flights, addressing label noise to uncover non-CDO flights that could perform a successful CDO. The third step trains a random forest-based model to analyze how interactions with other flights influence TOD decisions, ensuring successful CDOs despite potential conflicts. Experimental results using Air Traffic Management System (ATMS) data from November 2019 in the Singapore Flight Information Region (FIR) demonstrate that predicted TODs eliminate level-offs, increasing average vertical separation by 1418.59 feet. Fuel analysis reveals that interaction-free descents from the predicted TOD save up to 343 kg of fuel for an A330 aircraft, with an average saving of 54.56 kg per flight across major aircraft types. These results highlight the model's potential for optimizing TOD decisions and promoting sustainable aviation.

Featured Image

Photo by Neo Lee on Unsplash

Photo by Neo Lee on Unsplash

Why is it important?

This work is timely because fuel & emission-reduction targets are pushing wider optimization in Continuous Descent Operations, yet TOD decisions still often ignore downstream traffic and trigger level-offs. It is unique in jointly predicting whether a flight can do Continuous Descent Operations and where to start descent, using interaction-aware features and label-noise handling (Confident Learning) rather than optimizing descents in isolation. The result is a practical, data-driven pathway to reduce level-offs and fuel burn in real operations, which can directly inform next-generation decision support tools and increase the paper’s relevance to both researchers and practitioners.