New Glenn NG-4 — Blast & Debris Forensic Analysis

On 28 May 2026, Blue Origin's New Glenn vehicle — a 98-metre-tall rocket loaded with 1,815 tonnes of liquid oxygen and liquid methane — exploded during a static fire test at Launch Complex 36. Local news station WRAL captured the event on a fixed camera, providing a side-on view from several kilometres away.

This analysis uses that footage, supplemented by pre- and post-event Planet Labs satellite imagery and Google Maps measurements, to reconstruct the blast wave, estimate the explosion energy, and determine where on the vehicle the failure originated.

No proprietary data, Blue Origin internal documents, or government reports were used. Every number derives from publicly available sources and first-principles physics.

Both phases are subsonic throughout the measurement window — the blast front was already below the speed of sound at t = 4 s, meaning the acoustic transition occurred before the camera captured anything useful. The Sedov-Taylor strong-shock phase was over well before 4 seconds.

The slight kink in R(t) at ~335 m reflects a change in acoustic decay rate as the pressure pulse spreads into an ever-larger volume of air. It is not a Mach 1 crossing — both phases are firmly subsonic.

The overall R ∝ t¹·⁰ relationship (rather than R ∝ t⁰·⁴ of Sedov-Taylor) confirms we are observing a decaying far-field blast, not the prompt shock of a detonation.

The lightning protection masts (182.88 m / 600 ft tall) serve as a vertical ruler in the image. At a camera distance of several kilometres, perspective distortion over the vehicle height is less than 1°, so the horizontal pixel scale applies equally vertically.

The mast tops appear at y = 660 px. Projecting down using the known mast height gives mast-base ground level at y = 819 px. The explosion origin at y = 794.8 px sits 28 m above mast-base ground.

The launch platform itself is elevated above that reference level — estimated at approximately ~10 m by comparing mast and platform shadow lengths in Google Maps satellite imagery at the same sun angle. This estimate carries its own uncertainty of similar magnitude, and we treat it as a round-number correction. Subtracting it, the explosion origin sits approximately ~18 m above the launch platform surface — the true base of New Glenn — with a ±17 m uncertainty.

At ~18 m above the vehicle base, the origin straddles the boundary between the BE-4 engine section (0–15 m) and the base of the lower booster / propellant tank section (15–60 m). The wide uncertainty band means no firm distinction between the two is possible, but the central estimate points to the engine/booster interface.

New Glenn carried 395 t of liquid methane (LCH₄) and 1,420 t of liquid oxygen at a mixture ratio of 3.6:1. The lower heating value of methane gives a total chemical energy of 19.75 TJ — equivalent to 4,720 t of TNT if every molecule reacted perfectly.

In practice, an explosion converts only a fraction of its chemical energy into blast wave. This fraction — the blast efficiency — separates a slow fire from a city-levelling detonation. Our measurement is consistent with a blast efficiency of roughly 4–6%, giving an effective yield of ~210 t TNT — characteristic of a moderately energetic vapour cloud explosion.

It is worth noting that this measured efficiency is considerably lower than values used in regulatory licensing. US DoD and FAA standards for LOX/hydrocarbon propellants at launch pads use 20% TNT equivalence as the baseline for quantity-distance calculations. For LOX/LNG (methane) specifically, NASA's interim blast model notes the potential for methane-oxygen mixtures to behave as a sensitive high explosive under certain mixing conditions, adding further uncertainty that justifies the conservative regulatory approach. The gap between the measured ~210 t and the licensing basis ~944 t (20% of 4,720 t) is not a discrepancy — it is the difference between what happened in this particular event and the worst-credible-case that must be bounded for public safety.

The explosion produced a fireball that grew to approximately ~495 m diameter before rising and dispersing, burning for a duration timed by eye off the live video at ~35–40 s. That duration is indicative only — neither set of screengrabs spans the full burn, so it is not a photogrammetric measurement (for comparison, the Roberts correlation predicts ~22 s). The diameter, by contrast, was measured from the video frame at maximum extent: the left edge extends off-screen, so the right-side half-width from the origin was doubled assuming symmetry.

These observed values bracket the predictions of standard fireball correlations — Roberts (1982), Moorhouse & Pritchard, and TNO Yellow Book — all developed for hydrocarbon releases of roughly 1–10 tonnes. Extrapolating nearly 400× beyond their calibration range, the agreement within 15–20% is notable.

The slightly larger measured diameter likely reflects the LOX co-oxidiser providing additional internal oxygen, making combustion more complete — and the fireball marginally bigger and longer-burning — than a pure hydrocarbon release would produce.

The survival of steel pressure tanks at 170 m while a tented structure at 375 m was damaged resolves through material thresholds: steel pressure vessels survive tens of thousands of kPa, while tented fabric fails at as little as 0.1–0.3 kPa. Both outcomes are consistent with the same blast yield.

More revealing is the tree scorching at 450 m. The blast overpressure there is 0.14 kPa — below any structural threshold. But the thermal flux from a ~500 m fireball burning for 35 seconds is approximately 56 kW/m² at that distance — well above the ~12 kW/m² threshold for vegetation ignition.

Beyond ~300–400 m, thermal radiation was the dominant hazard. This is the defining signature of a large VCE / deflagration. A detonation shows exactly the inverse relationship.

Alongside the blast and thermal hazards, the footage shows numerous incandescent fragments thrown clear of the fireball. One bright fragment was tracked across 8 consecutive frames, one second apart, to reconstruct its trajectory and estimate where it came down.

The scale is anchored to the 73 m separation of the two lightning masts (measured from satellite imagery via two independent workflows). This horizontal baseline is chosen over the towers' height — it needs no estimate of ground level or of the tower tops, which bloom into the fireball. To pin the scale and check the camera geometry, three landmarks at known coordinates (the masts, a water tower, and the Cape Canaveral lighthouse) were measured in a clearer frame: the observed pixel-separation ratio (lighthouse-to-tower vs water-tower-to-tower) matched the prediction from their surveyed positions to 2.5%, confirming the camera location (~9.1 km SSW) and that the view is within ~6° of broadside to the mast line — so the parallax (foreshortening) correction is negligible (~0.99) and the on-screen mast gap is the full 73 m. Fragment positions were then measured against this validated scale; the integration facility base sets the altitude datum.

The eight points trace a clean ballistic arc — apogee ~360 m, reached roughly 10 s after the event (video t ≈ 24 s; event onset taken as t ≈ 14 s). The per-second vertical drop accelerates consistent with gravity, confirming ballistic motion. Projecting the descending limb forward to the ground gives an impact of ~650–770 m: the spread combines the drag-vs-no-drag fall with the residual (few-pixel) scale uncertainty.

Two independent checks support the reconstruction. First, the landmark geometry: surveyed coordinates predict the on-screen spacing of three reference points to 2.5%. Second, the back-extrapolation: tracing the fitted arc backwards returns it to the launch vehicle near the mast midpoint. The measured arc and the geodetically-anchored scale converge on an origin at the pad, where the rocket stood.

Placing the result against US DoD explosive-safety separation distances (DESR 6055.09) is instructive. The estimated impact at ~650–770 m is roughly 1.7–2.0× the default storage hazardous fragment distance of 381 m. It falls within the emergency-withdrawal distances and well within the maximum-fragment-distance figures, so it sits comfortably inside the site's wider explosive perimeter.

The point that has always quietly nagged at me: the standard hazardous fragment distance is treated as effectively bounded above a certain net explosive weight. Yet here is a single fragment, from a vehicle far in excess of that quantity, landing at up to roughly twice that bound. Containment still held — it stays well within the site's wider explosive perimeter, so this is not a safety finding. But it is a tangible reminder that the bounded-distance assumption and the physics of a very large energetic event do not sit together quite as tightly as the tidy number implies.