Amazon Prime Flight Down in Texas

You may have seen this.  A Prime Air (Amazon) flight went down in TX yesterday, depending on your local value of yesterday.  Briefly, it looks to me like the thing suffered a catastrophe at altitude.  Good comms were in evidence leading up to the loss, but no communications were heard from the plane as it began a descent that can be described as free-fall modified by the aerodynamics of a broken airplane.

The plane is a 25-year-old 767 which spent decades in the care of several non-US airlines, including two years in storage before being picked up by Atlas, who held the lease (wet/dry unknown) while operated for (or by) prime Air.

Good job DocLor for the link to Flight-Aware, which has the majority of this puzzle laid out.

I followed up on the aircraft’s report “descending via the LINKK”.  See the linked approach plates, absolutely necessary to follow what comes next:

When the pilot reports “descending via the link”, the readout gives his altitude as 17.9K ft, speed 320Kts, (indicates he is a (H)eavy), and descending in excess of 1,000 fpm, which is normal.  Descent rates, especially in IFR, vary between 1,000 fpm at a minimum (well either we’re descending or not, right?) and 2,500 fpm.

I don’t actually know where any of these named navigation waypoints are — I’m just fitting the evidence to the approach cards.  So I could be off about the location og GILLL.  There’s some talk on the forums about the right turn being for weather avoidance.  But I still think that this is the programmed right turn at GILLL.

It looks to me that the aircraft followed route JEPEG / GIRLY / BBQUE / LINKK / GILLL, and at last turned for GARRR.  Per the approach plates, this suggests an (ATC) intended set-up to approach for a landing at RWY 26L or 26R.  [Yup, they were  directed in the audio to follow thr transitions for 26L]  Critically, the aircraft would be limited to between 10K and 8K feet as it crossed GILLL, and should have been doing 240kts.

Now, the approach plates are advisory.  ATC instructions are STRONGLY advisory, and the Captain at all times retains discretion over what his aircraft does.  So it may be that prior conversations or known schedule issues affected the speeds flown — facts not yet in evidence.  That said:

The aircraft is on a course consistent with the 269 required inbound to LINKK, slowing from over 300kts to somewhere between 290 and 300kts at LINKK.  His airspeed* crossing LINKK should have been down to 250kts per the approach plates.

The aircraft is required to cross LINKK between 15kft and 12kft altitude at 250kts, and to turn right to course 301, continue descent or maintain to cross GILLL between 10kft and 8kft at speed 240kts.

We see him cross GILLL at time mark 1:06 in this video.  He turns consistent with the expected transition from a course of 301 (LINKK to GILLL) to a new heading of 325 (GILLL to GARRR).  His speed was not 240, but 290kts.  he came across right at the top limit of altitude for GILLL, which is 10kft.

This to me is where the story gets interesting.  He is now high and hot, as they say (not in the airfield sense, but flight profile sense), and he needs to lose both altitude and airspeed.

This is assuming that he does not wish for some reason to keep the additional speed for scheduling reasons, but at this point he is close enough to the airport that excess speed no longer helps even in a time crunch.  The faster you go close to the airport, the more time you’re going to spend trying to bleed off the excess energy.  It’s challenging to lose airspeed and altitude at the same time, as it’s naturally a trade-off.

But I believe that this story begins with getting too high and too fast, and then trying to come down from that.

Just before the pilot reported “descending via the LINKK”, there is a controlled descent from 13:26 to 13:30 in which the aircraft never lost LESS THAN 2500 fpm, but maintained speed, and in the last minute (13:29 to 13:30), actually dumped 30 kts while shedding very nearly 4kft.  There’s only one way to shed that much energy, and that’s aerodynamics.

Just after he crosses GILLL and turns right, in the minute from 13:27:00 to 13:28:00, the aircraft dumps another 50 kts while losing 2000 feet.  he has gotten his speed down to 250kts (or so), which is the max speed for the initial deployment of flaps on the 767.  If the previous hard-braking descent stressed components of the airframe, this is where they finally let go.**

More to follow.

This weather radar picture (swiped from tells a different story than the wx in the video above:

The time in this one is supposed to be 13:20, and the system is moving west to east.  That makes this line of weather significantly ahead of the one seen in the video, (last seen still far west of the bay by 13:40).  This line doesn’t show up on the video at all.

It could be a time sync issue; latency in availability of web products somehow.  Or maybe there was nothing wrong with the plane except for being configured to lose altitude and airspeed and then running out of both in a wind shear.  At their altitude, the first thing they would experience hitting a line like this would be a massive tailwind, and a massive updraft.  If it hit them hard enough, it could have caused an aerodynamic stall on an aircraft already doing its best not to fly very well.  This would be consistent with the unexpected left turn — typically one wing stalls before the toehr and drops, causing an uncommanded turn.  This is also consistent with the brief climb indicated on Flight Aware.  They may have gotten the wings level without being able to restore level flight, which would be consistent with the rest of the story.

The dog not barking but biting very hard would be an unseen line of weather.  If the wx radar is right at the airport or a little northwest of it, the smarty systems might have decided it was an echo of the obvious line just east of the city and blanked it out.  Hold this thought.

There’s a comment on a forum from a guy who says that he was in the air, in the area at the time, heard the calls, saw the wx.  Said it was mostly “false ground returns”.  Except that maybe it wasn’t after all.

I can’t get to mp3 files from where I am now, but check this out:

I agree with the “Pull Up!” observation noted at that link.


*actually, the approach plates may be spec’d in kts over ground, as opposed to KIAS (Kts Indicated AirSpeed), as they deal with distances between fixed physical features.  I won’t be finicky for now.

** In the mudflats video somewhere, there’s a shot of the “smirk” showing that the rudder was with the aircraft when it hit.  So it’s unlikely that they ripped it off the way that one Airbus went in in 2001 or so.


24 thoughts on “Amazon Prime Flight Down in Texas”

  1. May they rest in peace.

    It is strange to listen to Air Traffic Control because you hear voices that will no longer be heard in minutes. Things were fine until they weren’t.

  2. drlorentz:
    ATC audio

    Doc, I am unable to listen, but a guy on a forum said that around the six minute mark there’s some cross-talk with the A/C requesting to go west around the wx, and Houston telling them to go east of it instead, due to a bunch of departures taking the west route.  This was elided in the VASAviation video.


  3. It made a right turn just before crossing the coastline and then a right-left at the end. Was that first right turn part of the intended approach?

    If not, then something may have fallen off over the gulf and we won’t know. Otherwise, the area between the coast and crash site is mostly farmland, so any large piece like an engine or a control surface should be found easily.

  4. Haakon Dahl:
    I agree with the “Pull Up!” observation noted at that link.

    I did not hear that. It does not seem to make sense to hear that immediately after an “OK” unless it was to avoid a controlled flight into terrain or it was to avoid a crash with another aircraft. But, the rapid descent interval must have come after the “OK” so controlled flight into terrain is unlikely. And we saw no other aircraft.

    Would you get a “pull up” in another circumstance such as a slight deviation from approach?

  5. By coincidence, this week’s edition of Aviation Week, which came to hand today, has an article about Amazon Prime Air.  This is, at present, not a competitor to package delivery companies for direct-to-customer shipment.  It is used exclusively for internal shipment of stock from central locations to fulfillment centres (warehouses and sorting facilities), of which Amazon presently operates more than a hundred in North America.  This, in turn, reduces Amazon’s shipping costs between the fulfillment centres and the ultimate customer, which is crucial to the profitability of Amazon Prime, where customers get two day free shipping for an annual subscription fee.

    An analyst is quoted as saying, “Amazon is building an irregular network of ad hoc flying, not scheduled routes.  Amazon’s schedule compared to FedEx’s Memphis operation showed very little overlap in capacity, frequency, or markets served.  Overlap means competition, and that is not what we have going on here.”

    Prime Air presently operates forty Boeing 767 aircraft leased from two lessors, and plans to add another ten and possibly more in coming years.  The aircraft are “wet leased”, in which the lessor provides the aircraft, crew, maintenance, and insurance, while the lessee (Amazon) decides when and where they fly.  The fleet currently log about 6 to 7 hours of flight per day, which is less than half the utilisation of 767s operated by FedEx and UPS.

    Amazon reports it paid US$ 27.7 billion in what it calls “shipping costs” in 2018.  Even a small percentage reduction by stocking closer to the customer can make a substantial contribution to the bottom line.

  6. ctlaw:
    It made a right turn just before crossing the coastline and then a right-left at the end. Was that first right turn part of the intended approach?

    If not, then something may have fallen off over the gulf and we won’t know. Otherwise, the area between the coast and crash site is mostly farmland, so any large piece like an engine or a control surface should be found easily.

    If I’m right about the placement of the 5LNCs (5-Letter Name Codes), the first right turn, out int he Gulf, was crossing LINKK, and the second was GILLL.  The subsequent left turn I am attributing to a stall, as they should still have had miles to go to reach GARRR, the next scheduled left turn.

  7. An NTSB “Investigative Update” has just been posted.  This changes the complexion of the event.

    About 12:38, the controller informed the pilots that they would be past the area of weather in about 18 miles, that they could expect a turn to the north for a base leg to the approach to runway 26L, and that weather was clear west of the precipitation area. The pilots responded, “sounds good” and “ok.” At this time, radar and ADS-B returns indicated the airplane levelled briefly at 6,200 ft and then began a slight climb to 6,300 ft.

    Also, about this time, the FDR data indicated that some small vertical accelerations consistent with the airplane entering turbulence. Shortly after, when the airplane’s indicated airspeed was steady about 230 knots, the engines increased to maximum thrust, and the airplane pitch increased to about 4° nose up. The airplane then pitched nose down over the next 18 seconds to about 49° in response to nose-down elevator deflection. The stall warning (stick shaker) did not activate.

    FDR, radar, and ADS-B data indicated that the airplane entered a rapid descent on a heading of 270°, reaching an airspeed of about 430 knots. A security camera video (figure 4) captured the airplane in a steep, generally wings-level attitude until impact with the swamp. FDR data indicated that the airplane gradually pitched up to about 20 degrees nose down during the descent.

    The update further notes that in addition to the pilot and copilot there was a “nonrevenue jumpseat pilot” on the flight.

    It appears that the engines were set to maximum thrust and the descent commanded.  There is no evidence of weather causing an upset or loss of control.

  8. Beginning to look creepy. At any rate, as they had stabilized altitude and even climbed a bit, that does away with my suspicion about an airplane configured for rapid descent getting ahead of the crew, with or without weather.


  9. There was one small change between the initial report read in comment #17 and the version currently posted on the NTSB Web site.  The original version attributed the nose-down to “control column input”, while the present version says “response to nose-down elevator deflection”.  The difference is subtle but significant: control column input implies the yoke was pushed forward, while elevator deflection could have been commanded by the autopilot or caused by a mechanical malfunction.  I don’t know if the flight data recorder on the 767 measures both control column movement and elevator deflection independently.


Leave a Reply