This was a very close call. Initial reports indicated it was like an ordinary go-around. No. "The incident pilots advanced the thrust levers when the airplane was about 85 feet above ground level. Flight data recorder data indicate the airplane was over the taxiway at this time. About 2.5 seconds after advancing the thrust levers, the minimum altitude recorded on the FDR was 59 feet above ground level."
They overflew an Airbus A340 on the taxiway, tail height 55 feet.
Indeed. I'm curious what options the pilot of UA1 (the 787 sitting at the end of the taxiway) had. There literally was no time to move the plane anywhere else.
Both pilots have said they thought the lit runway was 28L, and thus the taxiway, without visible approach lights, was 28R. If the pilots thought the approach lights for the desired runway were obscured by ground fog, which is possible at SFO, they might have made that assumption.
Wait for the NTSB report. This one will be closely analyzed for years.
Something is deeply wrong somewhere if ground fog can make an airplane land on a taxiway. Either airport design, or pilot training, etc... But this should not be an understandable mistake.
"...the minimum altitude recorded on the FDR was 59 feet above ground level." [emphasis added]
AGL (elevation above ground level), not MSL (elevation above mean sea level). The runway/taxiway elevation is already taken into account. They were 59 feet above the taxiway.
They were very close. See [1]. They were over the centerline of taxiway C, and 85 feet AGL crossing taxiway W. Minimum altitude was 2.5 secs later. There were three planes ahead, an A-340, a B-787 (also 55 feet high at the tail), and some UAL jet of about the same size.
A few feet lower and there would have been a collision.
My point, by the way, was that tails are relatively speaking, breakaway structures. So maybe not much would have happened.
I've had three accidents where my vehicle was hit from the side by another, moving at ~100 km/h. In two of them, the other vehicle hit crumple zone stuff, and I hardly felt the impact. In the third, there was frame-frame contact, and I experienced extreme angular acceleration. It almost ripped my head off.
Investigators could not hear what the Air Canada captain and co-pilot said to each other during the aborted landing because their conversation was recorded over when the plane made other flights, starting with a San Francisco-to-Montreal trip the next morning. Recorders are required to capture only the last two hours of a plane’s flying time
I wonder what the procedure is in a near Miss like this. It looks like not enough of the 'process' to capture evidence kicked in soon enough. Judging by the photo and distances this looks like it was an imminent disaster, which hadn't been clear to me previously.
Retrofits to safety equipment are common when new issues are discovered. That said, an enforcement action or a procedure change is probably necessary in this case.
For sure, and it might get changed now, but it wouldn't surprise me if planes are still flying with age old cockpit voice recorders, simply because the requirements on that front haven't changed. (Bearing in mind this is just an educated guess, I have no idea about whether or not they have changed)
30 was just a number I pulled out of the air because I have flown aircraft that were about that old. Of course, like you say, equipment could have been retrofitted so it's hard to know.
Before I couldn't imagine how anyone could make this mistake, but now I see it. The pilots were familiar with SFO and knew that there were two runways, but didn't realize the left hand runway lights were turned off, so they had an off-by-one error when choosing the runway.
Yeah, that's the key new info for me too, something I missed during the previous discussion - I think it wasn't mentioned at all then. The key quotes from the OP article:
"Runway 28L was closed to accommodate construction; its approach and runway lights were turned off, and a 20.5-ft-wide lighted flashing X (runway closure marker) was placed at the threshold. Runway and approach lighting for runway 28R were on and set to default settings, which included a 2,400-foot approach lighting system [...]
Both pilots said, in post-incident interviews, they believed the lighted runway on their left was 28L and that they were lined up for 28R. They also stated that they did not recall seeing aircraft on taxiway C but that something did not look right to them."
recently harrison ford screwed up and landed on the taxiway at KSNA in southern california. he didn't even do a go around, just landed on the taxiway - and he gets to keep his certificate apparently http://www.ocregister.com/2017/04/04/harrison-ford-will-keep...
> Runway 28L was closed to accommodate construction; its approach and runway lights were turned off, and a 20.5-ft-wide lighted flashing X (runway closure marker) was placed at the threshold.
i'm not a pilot and have no idea what it looks like when landing a plane at night, so i really shouldn't be commenting here, but: the report says the one operational runway had 2400 ft of approach lighting that was functional. The disabled runway had no approach lighting, and presumably the taxiway didn't have any approach lighting because nobody should be approaching a taxiway. So yes, they were just "off by one", but they were off by one on a dataset of one.
Yeah, but once you decide that the leftmost runway you see (with the dark void on its left) is the left runway, then obviously the parallel strip of lights to the right of it is the right runway. Sure, the lights look odd, and your gut's giving off this feeling, but you see, that's the left runway, and I don't see the right runway anywhere else, so this must be the...
They probably got stuck in several cycles of that before the contrary evidence was overwhelming enough to break them out of their off-by-one mental model.
Can you see the X at the beginning of the runway while sitting in the pilot seat? I was told the pilots' eyes are expected to be at mid-windshield height, but pilots prefer their seats in slightly reclined position and no position is compulsory.
It's meant to be seen when approaching the disabled runway, not as a generic landmark to help the pilots orient themselves. So it's not prominently visible when approaching the taxiway.
>> Can you see the X at the beginning of the runway while sitting in the pilot seat?
On approach you should be looking at the threshold and the numbers. That's right around where the X was, so it would be clearly visible to anyone trying to land on the closed runway. They were not trying to land on that one, and since they were over the taxiway the X was actually two "lanes" away from where they were. I don't think the X is terribly relevant to what happened here.
I a am a pilot. At night, after a long flight to a place you don't live at, you are desperately looking to orient yourself. They looked at the charts, saw 2 runways, and found them. It can be extremely disoritening, especially at a place like sfo where there are two parallel sets of runways.
What we need is better computer guidance. The tech is available, our legislature is not. Look at the incredible "taking out freedoms" pushback against adsb to see what I'm talking about.
Only ADS-B Out (which refers to an aircraft broadcasting its position and other information) will be required in the near future. ADS-B In (which refers to an aircraft receiving the broadcasts and messages) remains optional.
Also, due to the common airport design of runways and parallel taxiways, it isn't uncommon to fly either near or over waiting planes on final approach, so this may not have been cause for alarm even if they were equipped with ADS-B In.
Well, they were off by one in adjusting the length of their internal model of the dataset (two runways) with the currently active model (one) and thus suffered a buffer overflow.
Anyway, score one (more) for the robots, who would have picked up the correct ILS localizer.
28L was dark apart from the X, and the taxiway was lit (albeit with taxi Lane instead of runway lights). Expecting two lit lanes and approaching the one on the right, they certainly saw something "off" but this wasn't a case of pilots choosing to avoid the only lit lane.
This literally could have been one of the worst accidents in American aviation history had the pilots of the landing airplane failed to react in time. There were 4 fully loaded and fueled passenger jets directly ahead of them.
The final NTSB report will probably calculate this, but I'm guessing if they had waited another second or two before shoving those throttles forwards they would have hit a plane (or three).
Whats even more scary, is I get the impression that the pilots only shoved the throttles forwards because the tower told them to abort.
No, they started pulling up just before hearing from the tower; click on the link at the bottom of the referenced page: "The full investigative update and other information about the incident are available online at: https://go.usa.gov/xRPFk " for more detail, including:
o At 2356:04 PDT, ACA759 reappeared on the local controller’s ASDE-X/ASSC display as it passed over the first airplane positioned on taxiway C.
o About 2.5 seconds after advancing the thrust levers, the minimum altitude recorded on the FDR was 59 ft agl.
o At 2356:10 PDT, the local controller directed ACA759 to go around. The airplane had already begun to climb at this point (see figure 4).
This means that if they hadn't acted until they heard from the tower, it would have almost certainly been a disaster.
Yeah, the 3 video frames in that report are matched to transmissions the timestamps. Unless the angle is deceiving, you can sorta visually extrapolate from the descent of the first two at :01 and :04, that 6 seconds later, when the go-around was give at :10, they would have been on the ground had they not pulled up.
If you listen to the recording the pilot suspected something was not right, that's why he asked for confirmation that the runway is clear. Most likely he already anticipated a high probability of a go-around.
They teach you in driving school to never overtake in a two-way road unless you are 100% sure you can make it. I'd assume pilots would he held to at least equal standards...
Driving school does everything it can to dissuade you from ever overtaking on a 2-lane because driving in a big, slow line is what's acceptable for the lowest common denominator.
Driving school treats people like idiots by default.
Professional training assumes people have the capacity to evaluate situations and make their own decisions on the appropriate course of action.
Comparing training intended for the general public and training for professionals is very much a stretch IMO.
The NTSB does not do what-ifs in the sense of calculating expected body count had an accident not been averted. They do, however, often investigate the performance envelope involved in an incident/accident and the timeline for possible corrective actions. So in an investigation it's quite common for the NTSB to look into how initiating corrective action at different points in time would have affected the outcome. e.g.: Sentences like "the pilots could have [taken corrective action X] until Y seconds before collision is a pretty standard sentence to see in an investigation report.
This kind of information is important for determining the available safety margin and the proper changes to recommend in order to prevent similar incidents in the future.
Yeah, that kind of conclusion makes more sense as opposed to "If the events wouldn't have played out as they did, X would have happened". No one really cares, except to know that it would have been bad, which is why there is an investigation to begin with.
Agreed. It's fairly uninteresting (from an accident-prevention perspective, not a news-selling one) to muse about cutting it close, as soon as you get into seconds-territory, you're way too close. The fuck-up was lining up for the wrong approach in the first place, landing on the taxi-way is insanely dangerous, even if it's empty.
That's what I think as well, it should be sufficient to know that terrible things would happen and like you say, just getting in to a situation like that is bad enough.
This incident as well as the previous ones shows that pilots need better support in these conditions. Lots of airports are set up with multiple parallel runway configs. It wouldn't be all that difficult to add computer driven confirmations of the runway an aircraft is approaching on. Given the frequency of these incidents as well as the cost of the consequences it seems like a reasonable addition. Such an addition won't end the problem, but will add another safety catch.
The problem with these things is how do they interact: what do you do in the cockpit when several alarms are going off at the same time? Just adding "one more alert" seems easy but... This is actually a deep problem with automation, how do the parts interact. When writing software this is a constant battle, between modularity and connection.
It's actually even worse than that: In more than one aviation accident multiple alarms and alerts actually served to confuse the pilots as to what's going on, causing them to not perform the action that would have saved the flight.
Information overload in the cockpit is a very real thing, particularly in an already task-saturated time like the landing approach. It's entirely unclear that adding yet another alert chime requiring the pilot's attention would help anything.
Yeah I definitely agree with this concern. Rather than an alarm it could be part of the landing checklist. The person not flying would say, "Check approach on <whatever runway cleared>", the person flying would check the display and respond "Confirm approach on <whatever>".
Me too. This is something that already exists today [0]. It can detect when the aircraft is not lined up with a runway and issues a "Taxiway Landing" aural and visual alert.
So the guy keying up and saying "Where is this guy going?" and "He's on the taxiway" might actually have made a difference in the outcome, huh:
> The flight crew of the first airplane in queue on taxiway C (UAL1) transmitted statements regarding ACA759, one of which mentioned the alignment of ACA759 with the taxiway while ACA759 was on short final (see figures 2 and 3).
> The flight crew of the second airplane in queue on taxiway C switched on their airplane’s landing lights as the incident airplane approached.
It may be doubly fortunate that that multiple pilots and controllers did not attempt to say this at the same time. At Tenerife, both the Pan-Am crew and the tower suspected that the KLM aircraft had started its take-off roll, and both tried to warn its crew at the same time, but the resulting radio interference between the concurrent transmissions made the messages unintelligible.
The author states that there is a feasible technical solution that, at the time of the article (2002) had not been mandated. Does anyone know if the situation has changed?
28L is longer than 28R. Similar to how 28R (or at least its lighting) is longer than taxiway C. The general geometric appearance of the lighting was probably similar enough to what they were used to that it looked mostly correct. As for the X, it was 20 feet wide and about 1000 feet away from where they were aiming, so it would not be relevant to the incident.
>> Maybe I'm looking at this wrong, but 28L is most definitely shorter than 28R.
You are looking at it wrong. 280 degrees from north would be approaching from the upper right heading down and to the left on the map. From the pilot point of view 28L would be to the east and extends further out into the bay.
That video doesn't match the sequence of events in the report (Which makes sense, it was published a few weeks ago). According to the report the remarks from UAL1 come as the airplane is overflying the taxiway.
I've seen people mentioning this could be a big disaster, but I'm curious exactly what the effects could be. Specifically, ignoring the wheels for the moment, would the hull of the landing plane slide off on contact at all, or would the planes just get squashed? I have no intuition about friction / stiffness of materials at that scale. (I'm also assuming the planes were at least a little bit offset to the side)
With the wheels down, they would just tear through the hull of the lower plane, right?
We're discussing a matter of 10s of feet between the near-collision we saw, clipping the tail of one plane, or hitting the plane behind at landing speed, so I imagine the set of "what could have happened" is rather large.
That said, I suspect the following three crashes give _some_ indication of what might have happened:
USAir 1493 - 737 collides with a turbo-prop on landing; the smaller plane is essentially crushed by the 737
Asiana 214 - Landing gear and tail clip seawall on approach
Tenerife disaster - 747 strikes the side of a second 747 a few seconds after leaving the ground on takeoff; engines, lower fuselage and main landing gear struck the plane. 583 fatalities.
I highly recommend watching the related episode of "Mayday" Air Crash Investigation about Tenerife.
Hulls are so fragile that they've often shred up in flight (De Havaillan, DC-9). Landing is around 200-250kmh only, but planes have caught fire just sliding on runways or terrain. The quantity of jet fuel immediately means catastrophe.
This. Sure. But pure speculation. The forward momentum of that airplane could have clipped a tail and taken it off while pulling up and max thrust. He was wayyyy below minimum decision height (called out and set before landing, usually a few hundred feet).
OTOH, he could have very well collided with the other Airbus on the taxiway rapidly halting his forward momentum and exploding into a fireball subsequently taking out everything for half a mile ahead.
What's alarming her is the distance of the Air Canada jet to the ground, less the tail of the Airbus = this guy was FEET AWAY from clipping that jet.
Actually the first plane on the ground was staggered with respect to the others and the final two ones had quite a bit of separation so another, not too unlikely outcome, is that it would have hit UAL1 and then veered of the taxiway, wrecking the ACA flight but nothing else (Save for possible some last-row passengers on UAL1).
The hull of a jetliner is about the thickness of sheet metal and disintegrates. The engines are denser and plow through things. Here's NASA footage of a controlled crash:
Large objects tend not to be stiff against collision. It would look like a high speed road crash - remember that the landing plane is travelling at over 150km/h.
I would love to see the video that frame is taken from to watch the arc of the plane as it initiated go-around and missed a collision by just a few feet.
The NTSB has obtained a security camera video from SFO of the incident approach that will be released along with the other factual information when the public docket for this incident is opened in the next several months.
I don't think GPS is accurate enough for that sort of thing, you'd need ground radar, working with onboard computers that understand which runway you're cleared for, etc. (I'm not a pilot, I'm speculating.)
If you read the Guardian article from planetjones' comment:
The investigators said that as the Air Canada jet approached the taxiway just before midnight after a flight from Toronto, it was so far off course that it did not appear on a radar system used to prevent runway collisions.
The next paragraph states they are working on improving that.
It's called ILS, Instrument Landing System, and it's been around for literally 80 years.
> The first landing of a scheduled U.S. passenger airliner using ILS was on January 26, 1938, when a Pennsylvania Central Airlines Boeing 247D flew from Washington, D.C., to Pittsburgh, Pennsylvania, and landed in a snowstorm using only the Instrument Landing System
Presumably they must have done a visual approach or something, because their ILS would have been outright incapable of directing them down the taxiway, as it's for obvious reasons not equipped for ILS approaches.
ILS is separate from the radar system mentioned by parent. ILS uses directional radio signals from each end of the runway to provide a glideslope and lateral guidance, not radar.
The signals are transmitted from the ground and used for guidance by the aircraft. No part of ILS could alert ATC to anything since the ground stations are only transmitters.
Trouble there is that, at least to the best of my knowledge, ILS glide slope and localizer are generally only surfaced in the cockpit as a pair of bugs vertically and horizontally peripheral to the artificial horizon indicator. If you're already having a helmet fire, these are likely easy to overlook, and there's no horn or anything. (But as has been pointed out here by several other commenters before now - if you're already having a helmet fire, another horn may not do any good.)
I don't think there's any indication that these pilots were under any kind of undue stress when they lined up their approach? If at any point during the several minutes of their final approach they'd looked even once at their ILS they'd have seen that they were way off.
Final approach is the most stressful two or three minutes of an otherwise unexceptional flight; there's a great deal to do, and very little time in which to do it, during a period in which the aircraft's proximity to the ground, and low available energy, make it very easy for what would otherwise be a minor mishap to instantly place everyone aboard in imminent danger of death.
I don't suppose I would call such stress "undue", since every commercial flight ideally ends with a landing. But the probability of helmet fire is directly proportional to the degree of cockpit stress, and the ILS alignment indicators are very easy to overlook. I suspect the user interface could be improved, but I'm not nearly familiar enough with the subject domain to imagine how.
Agreed. Just to be clear, I'm not suggesting that we gang up and witchhunt the pilots for not checking the ILS, my agenda was simply to point out that the technology does indeed exist.
It's irrelevant why they were there, it's a taxi way, it's fine, indeed expected, for planes to hang out there. The important bit was why the AC tried to land there and the crew confirming their approach on ILS would have prevented that.
The main reason is that computers aren't very smart although I know that even some developers think that.
Actually they can autoland but not everywhere and then you still need to tell them where, basically tune them in on the right glide path. Then you really need to be prepared to take over any second, which is a dangerous position on its own. (Just like all self driving cars right now.)
It's still possible to select the wrong ILS beacon or entering the wrong speed or whatever, and something can still show up in front of the plane.
It's just better in low visibility where you otherwise manually have to follow the ILS 'needles'.
Didn't a plane crash there killing some chinese schoolgirls just because the pilots forgot to set the correct speed. Either it was set too low or they set it but did not activate it. And then the monitoring pilot forgot to monitor the speed since the autopilot never fails...
>Actually they can autoland but not everywhere and then you still need to tell them where, basically tune them in on the right glide path.
Why can't the ATC or Pilot guide the computers to the right glidepath? And most importantly, if you're not going in that glidepath, you should get a major alarm, with ATC warned (so they can manually tell the pilot to abort).
They do and they did? But it's primarily a manual process.
There are automated warning systems such as ACAS and separation warnings, but I don't think it would be possible to program something that could understand what's going on with all the traffic all the time and just generally warn if it's not looking right...
Raises an interesting question of which is safer: ILS automated landing with pilots on standby to take over, or pilots manually landing guided by ILS.
On one hand, a fully automated landing leaves less room for human error (you can still instruct the plane to land in the wrong place), but you're more reliant on things working as they should. If the automated systems fail for any reason, you've got pilots who are less practised at landings needing to take over at a moment's notice
I also suspect that a pilot making a active landing would react quicker to things than shouldn't happen (and thus may not be picked up by an automated system), such as a ground plane being in the wrong place, or some sort of mechanical failure.
The A320 can land itself, but pilots regularly approach and land manually to keep their manual flying skills current (as the automated systems -- both the ground components and aircraft components as well as the communications channels between them -- can fail).
The map view seems a bit off to me, UAL1 obviously saw the incoming plane and on the photos it looks to me like there are 4 planes in a straight line, yet UAL1 is oriented in the other direction on the radar.
I think it's the case as depicted in the first radar image, UAL1 turned 90-ish of degrees right on the exit from the taxi-way, and the other three lined up. It could be the angle of the camera is such that it's an illusion they're all in a straight line. That probably still gives them plenty enough view out the side of cockpit to see that the plane they're expecting is heading behind them and not in front of them where it should be.
I think the orientations in the subsequent screens where UAL-1 looks turned around are probably a case of the radar over-extrapolating from its heading from the turn and then not being able to tell its heading while it's parked.
> Lights for taxiway C were also on and set to default settings that included centerline lights (green) along its length. Default settings also included edge lights (blue) and centerline lights (green) illuminating the transition or stub taxiways from the runway to the taxiway.
> Runway and approach lighting for runway 28R were on and set to default settings, which included a 2,400-foot approach lighting system, a precision approach path indicator, touchdown zone lights (white), runway centerline lights (white at the approach end), runway threshold lights (green), and runway edge lights (white at the approach end).
Taxiway: centre green along length, blue edge lights. Can have planes on it.
Runway: centre white along length and edges (also landing zone), perpendicular green at end thresholds. Should not have planes on it.
The planes on the taxiway have white takeoff/landing lights. It sounds like the second in the queue didn't turn theirs on until just before the incident, but if any of the others had theirs on prior, it could contribute to misidentifying the taxiway. I think it's common to have the lights on while taxiing and turning them off at rest.
Wouldn't this whole problem disappear if runways were designed differently, at the minimum one exclusive for landing and one for takeoff with no chance for planes landing and planes taking off to meet, like 2 parallel hockey sticks.
Both runways would be painted and lit up completely differently with large lettering explicitly saying take off and landing.
No. The captain was not attempting to land on a runway. He was attempting to land on a taxiway, which is painted and lit up completely differently, and also covered in airplanes.
I may be putting words in OPs mouth here but I imagine he meant not putting parallel taxiways so close to the main runway.
I know that space restrictions make this very tenuous but I must admit that looking at the overhead on https://ntsb.gov/investigations/Pages/DCA17IA148.aspx that the taxiway marked C looks a lot more like a runway than the one marked F. (due to the width maybe?)
Can't the taxiways be not straight, like have a slight zig zag to them, so that they're easily distinguishable from a runway? Not ridiculously zigzagged, but maybe 3 or 4 angles of 15 degrees along the length or something would do it.
They overflew an Airbus A340 on the taxiway, tail height 55 feet.