Cellular Automata Reveal Hidden Inefficiencies in Tokyo Airport Arrivals

Every day, thousands of aircraft descend on the world’s busiest hub airports, and every unnecessary turn, hold, and slowdown burns fuel, emits carbon, and adds to the already intense workload of air traffic controllers. A new study published in Frontiers in Aerospace Engineering has borrowed a mathematical tool from highway traffic modeling—the cellular automaton—to dissect exactly where and when these inefficiencies occur, using Tokyo International Airport, better known as Haneda, as its proving ground.

The research, led by Adriana Andreeva-Mori of the Aviation Safety Innovation Hub at the Japan Aerospace Exploration Agency (JAXA), addresses a persistent problem in aviation performance measurement. Most existing efficiency metrics compare an aircraft’s actual flight path to some idealized benchmark—often the great-circle distance between two airports—and then tally up the excess time or distance flown. But these traditional metrics have a blind spot: they cannot distinguish between the different ways controllers adjust a flight. An aircraft slowed down to absorb delay looks the same in a distance-based metric as one vectored far off course, even though the two interventions have very different implications for fuel burn, emissions, and the predictability of the traffic flow.

Andreeva-Mori’s solution draws inspiration from the Nagel–Schreckenberg model, a cellular automaton developed in 1992 to simulate freeway traffic. In that model, a road is divided into discrete cells, and each vehicle advances a certain number of cells per time step depending on the space available ahead. Traffic jams emerge naturally from simple rules governing acceleration and braking. Applying the same logic to the sky required a conceptual leap: instead of roads, the study uses published instrument flight routes as the one-dimensional ‘lanes’ along which aircraft are mapped, and instead of cars, the cells are occupied by arriving aircraft.

The data underpinning the analysis comes from CARATS Open Data, a release of radar flight track information by the Japan Civil Aviation Bureau covering roughly 4,000 flights per day, with position updates every ten seconds. The study focused on arrivals to Haneda’s Runway 34L under north-wind conditions, which enter the Tokyo Approach Control Area through two main waypoints, SPENS and SELNO. Haneda had recently become the eighteenth airport worldwide to adopt the Point Merge System, an arrival sequencing method, in July 2019, making it an ideal candidate for this kind of scrutiny.

The first step was to determine the right cell size. The key assumption is elegant: flights that experience no interference from other traffic—so-called non-vectored or unimpeded flights—should advance exactly one cell per ten-second time step, regardless of whether they are in the enroute or transitional phase of descent. By analyzing thousands of historical trajectories and selecting those that stayed within one nautical mile of the published route for at least 95 percent of their data points, the researcher calculated average cell sizes for each phase and each arrival stream. Out of the SPENS traffic, 726 flights qualified as non-vectored in the enroute phase and 3,008 in the transitional phase; for SELNO, the corresponding figures were 58 and 238.

Once the mapping was established, every flight’s actual trajectory could be projected onto its nominal route, and its velocity—expressed as cells moved per time step—could be computed at every point along the way. A velocity of 1 means the aircraft is proceeding exactly as planned. A velocity of 0 means it has been decelerated or vectored, effectively standing still in the mapped space. A velocity of 2 or higher indicates shortcuts or acceleration, while negative velocities reveal holding patterns, where an aircraft momentarily moves backward along the route.

The velocity analysis yielded strikingly clear patterns. On a sample day in September 2019, flights arriving via SPENS with average velocities below a threshold of 0.89 cells per time step were almost invariably the ones with significant path stretching in their original radar tracks, while flights moving at near-nominal velocity hugged their published routes. Space-time diagrams of entire traffic flows revealed that controllers were making separation adjustments in two distinct zones: before aircraft entered the Tokyo Approach Control Area, and shortly after entry. On one September afternoon, two SPENS flights that were only five cells apart at 16:40 had been stretched to eleven cells apart by 16:52—all before crossing the sector boundary, indicating that the spacing work had been done well upstream of the approach control area.

Perhaps the most revealing finding concerns the relationship between velocity and the number of empty cells ahead of each aircraft. Statistical analysis across two sample days showed that decelerations occurred when the gap to the preceding aircraft was smaller than nominal—medians of nine cells for SPENS and thirteen for SELNO on one day—while accelerations happened when more space was available, with medians reaching 13.5 and 20 cells respectively. This mirrors the fundamental behavior of the Nagel–Schreckenberg model on highways, where drivers brake when the car in front is too close and speed up when the road ahead is clear. The fact that the same statistical structure emerges in the terminal airspace around one of the world’s busiest airports suggests that cellular automata are not just a convenient analogy but a genuinely predictive framework for air traffic.

Across 52 sample days, the proportion of cells flown at nominal velocity ranged from 78 to 88 percent, while decelerations accounted for between 6 and 16 percent of all mapped cells. Only three days showed more than 1.4 percent of cells with negative velocities—the signature of holding patterns—confirming that such extreme interventions are rare but do occur. The day with the fewest decelerations happened to be the day with the most accelerations, meaning many flights were able to fly shorter routes than nominal, a small but meaningful efficiency gain.

The study culminates in a proposed combined metric that weights each velocity category according to the priorities of the evaluator. If predictability is the goal—critical for trajectory-based operations where aircraft must fly precisely timed four-dimensional paths—the highest weight goes to nominal velocity. If time efficiency is paramount, shortcuts and accelerations are rewarded instead. Holding and vectoring can be penalized separately. Because the metric requires only publicly available track data and published routes, with no need for confidential flight plans or proprietary fuel models, it can be applied to any hub airport worldwide. The author notes that the approach currently addresses only lateral inefficiencies, and that vertical and speed-profile inefficiencies—which European research has shown contribute similarly to overall fuel waste—remain targets for future work. Still, the framework provides a computationally lightweight platform for understanding where air traffic systems deviate from seamless operation, and for measuring progress as the industry pursues net-zero emissions by 2050.

Subject of Research: Cellular automata-based evaluation of air traffic inefficiency and predictability at Tokyo International Airport

Article Title: Air traffic inefficiencies and predictability evaluation using route mapping—the Tokyo International Airport case

Article References: Andreeva-Mori, A. (2024). Air traffic inefficiencies and predictability evaluation using route mapping—the Tokyo International Airport case. Frontiers in Aerospace Engineering, 3, Article 1347229. https://doi.org/10.3389/fpace.2024.1347229

Image Credits: AI Generated

DOI: 10.3389/fpace.2024.1347229

Keywords: air traffic management, cellular automata, Nagel-Schreckenberg model, Tokyo International Airport, Haneda, flight efficiency, predictability, route mapping, air traffic controllers, fuel burn, emissions, Point Merge System