676 feet in Sixty Seconds

On 4 August 2026 an Air India Airbus A320, registration VT-EXO, operating flight AI 2379 from Phuket to Delhi, encountered a sudden vertical excursion in cruise over Odisha. Seventeen people were reported injured. The Aircraft Accident Investigation Bureau has taken up the investigation, with the French BEA participating as accredited representative and Airbus providing technical specialists.

The investigation will reach its own conclusions from the flight recorders. In the meantime, the aircraft’s own broadcast surveillance data is publicly available, and it is capable of establishing certain facts with precision. This note sets out what that data shows, states the method, and is explicit about the limits.

SUMMARY OF FINDINGS

• The excursion began at 04:02:49 UTC and was contained within approximately 62 seconds.
• Pressure altitude rose from 36,001 ft to a peak of 36,401 ft, then fell to a trough of 35,725 ft — a peak-to-trough span of 676 ft.
• The onset consisted of two distinct pulses separated by a four-second plateau, not a single displacement.
• True airspeed fell from 464 kt to 446 kt and ground speed from 481 kt to 457 kt, indicating a sharp wind-field boundary.
• Roll angle remained within ±5.4°. This was a vertical event, not a lateral upset.

The excursion reconstructed from broadcast surveillance data, 04:02:00–04:05:20 UTC, at approximately eight times real time. Barometric and geometric altitude traces are shown together; the two onset pulses are marked. Peak load factor is not determinable from this data.

Method and sources

All data used here is broadcast by the aircraft itself under ADS-B and Mode S, and captured by public receiver networks. Three independent datasets were examined: barometric altitude at native sample resolution; geometric altitude referenced to the EGM96 geoid; and Mode S Enhanced Surveillance registers carrying true airspeed, indicated airspeed, Mach, roll angle and track rate.

Sampling is uneven. Through the onset it falls to roughly half a second; before and after it runs at 20 seconds or more. Altitude is quantised in 25-foot steps. This matters, and it constrains every figure below. Earlier accounts of this occurrence drawing on 30-second sampled data reported an excursion of about 250 ft; that figure is an artefact of sampling that missed the peak.

The excursion

The geometric (GNSS) trace, which is independent of static pressure, gives a peak-to-trough span of approximately 748 ft over the same seconds. The two references agree to within their respective quantisation, which is itself a useful cross-check: the barometric altimetry was behaving normally throughout.

Two pulses, not one

At half-second resolution the onset resolves into a structure that coarser data renders as a single smooth curve.

Pulse 1 spans seven quantisation steps and is well clear of the noise floor. Isolating it further: the aircraft is level at 04:02:49.99 and has risen 174 ft by 04:02:52.47 — a rise of 2.47 seconds from a standing start.

On load factor: what cannot be claimed

The obvious question is what acceleration the airframe and its occupants experienced. This data cannot answer it, and it is important to say so plainly rather than to fill the gap.

Injuries in turbulence arise from gust loading lasting one to three seconds. The displacement produced by such a pulse is small: a 1.5-second event at 0.69 g moves the aircraft less than one 25-foot quantisation step. A one-second event of 1.55 g leaves no trace at all. Sampling that resolves the trajectory does not resolve the load.

Two bounding calculations can be offered, and both are labelled as estimates. Constraining a fit to begin from level flight, Pulse 1 implies a load factor increment in the region of +0.9 g if the acceleration were constant across 3.5 seconds, rising toward +1.8 g if concentrated in the 2.5-second sub-interval the data indicates. Separately, applying the standard gust-response relation to the implied updraft yields an increment of +0.7 to +0.9 g. These are convergent but assumption-dependent: aircraft mass, lift-curve slope and gust gradient are all estimated rather than known.

No load factor should be asserted from surveillance data. The A320 records normal acceleration on the flight data recorder at 8 Hz. That parameter, and only that parameter, settles the question — including whether any structural inspection threshold was met.

Airspeed and wind field

Ground speed fell from 481 kt to 457 kt across the encounter while true airspeed fell from 464 kt to 446 kt. Indicated airspeed dropped from 259 kt to 247 kt before recovering to 253–255 kt by 04:05:30. Mach and indicated airspeed were absent from the record between 04:03:20 and 04:04:10, so the true minimum is not known.

A ground speed change of that magnitude at constant track and constant altitude cannot originate with the aircraft; it reflects a change in the air mass. A second, smaller ground speed loss of 18 kt appears at 04:10 on the same track, which suggests a sheared region rather than an isolated cell. The aircraft had completed a step climb from FL340 to FL360 at 03:49, fourteen minutes before the encounter.

Position

The encounter occurred near 21.335°N 84.905°E, over Odisha. The nearest aerodromes were Rourkela at 55 nm, Jharsuguda at 59 nm, and Bhubaneswar at 83 nm. Of these, Bhubaneswar was the nearest field adequate for an A320. The aircraft continued to Delhi, landing at 05:37 UTC, approximately 94 minutes after the encounter.

Limitations

Altitude is quantised at 25 feet; adjacent readings differing by one step are indistinguishable from measurement dither and have not been differenced. Sampling collapses to six seconds or more after 04:03:04, so the recovery phase — including the unloading that would most plausibly account for injuries — is not resolved. Mode S register values may be held rather than refreshed, so apparently constant readings are not proof of constancy. No temperature data was broadcast. This note draws no conclusion as to cause.

Sources: ADS-B Exchange Mode S and position data; GNSS track (EGM96); FlightAware and Flightradar24 track logs for cross-validation. Aerodrome coordinates from published sources.

Analysis date: 13 August 2026. Occurrence date: 4 August 2026.

Safety Matters Foundation is an independent aviation safety organisation. This note is published in the interest of open safety discussion and is not a finding of cause. The investigation is the responsibility of the AAIB.