Hacker News new | past | comments | ask | show | jobs | submit login
Abstract of the NTSB Report on Air Canada flight 759's taxiway overflight at SFO [pdf] (ntsb.gov)
124 points by CaliforniaKarl 5 months ago | hide | past | web | favorite | 127 comments

The report identifies good preventive measures. Autotuning the ILS would be handy, as would an aural alert for lateral deviation, perhaps announcing, "Localizer, Localizer."

The crew may have caught the closure of 28L and planned better with a review of NOTAMS before top of descent and including them in the approach briefing.

When I'm up there, and sleepy, I ask for a coffee and pull out the NOTAMS for one last look before we pull up the approach in the computer, manually set up the radios, and do the briefing.

On final approach, more precise ground based monitoring would be nice to have in case things are fubar in the plane.

Just for SFO you have 84 NOTAMs.

!SFO 06/079 SFO OBST RIG (ASN 2017-AWP-3368-NRA) 373740N1222224W (0.5NM NE SFO) 41FT (34FT AGL) FLAGGED AND LGTD 1806201400-1811010100

This is a flagged and lighted obstruction half a mile way that is 34 feet above ground level.

Now put in a full route and takeoff airport and maybe if you have had some stops during the day. The NOTAM seems a bit long. Then if are overseas and have to deal with the BS political notams. Check out greece and turkey notams.


I'm curious how many pilots fully read all NOTAMs, locate them geographically to understand where they are etc etc on every flight.

Outputting that in lower case could be a small step to making them easier to parse, and therefore more likely to be read. I find the text you posted tiring even to look at.



The ref (B) Turkish NOTAM A3009/16 LTAAYNYX (111139 Euecyiyn Jul 2016) has no ground, cannot produce any internationally legal effect within Athinai FIR/Hellas UIR and is considered null and void.

Yeah, good luck with changing hundreds of established, life-or-death-systems, most of them actually embedded in planes in the airline industry worldwide, at the same cutover date, to suddenly support mixed-case script.

Have a one-hour watch to understand what kind of beast you are wrestling here: https://media.ccc.de/v/31c3_-_6308_-_en_-_saal_1_-_201412281...

  good luck with changing hundreds of established, 
  life-or-death-systems, most of them actually
  embedded in planes
Seems to me that's precisely the sort of thing the NTSB _is_ supposed to be able to do?

Of course IMHO a mere change of case doesn't go far enough.

You do realize that US planes still have to be able to fly to countries other than the US, and vice versa, right?

How much experience do you have reading all-caps text? It isn't necessarily a burden to people who are used to it, which all airline pilots are.

I'm sure prior experience is a big factor, but I could see how the greater height variances in lower case text can make it easier to read.

Indeed. 'Notices' can be almost anything.

I was doing simulated flights over the continental US recently. Even small airports may have a lot of NOTAMs. Some are important (for instance, warning about objects close to some parts of the runway, or issues with a runway, or even animals or birds), but many others are completely useless.

Noise abatement restrictions ironically add a lot of NOTAM 'noise'. There should be a separate category for these, as these are not safety related.

This was 100% my read as well. NOTAMs are terribly formatted and have no priority distinction between ones which are essentially useless, and ones with meaningful consequences.

Had a project once where we were trying to parse NOTAMs so we could program a GPS with a warning indicator if you were about to fly into a restricted airspace.

Turns out that people type what they want and the parser ended up needing to be almost a full AI. The exclusion zone around DC is a nightmare to parse for example.

The recommendations look like important steps, especially the concept of a system that alerts pilots to runway surface non-alignment. I have to wonder, though, does it always take a close call like this (or worse) to bring up such a proposal? Does the NTSB also study otherwise safe flights to identify risks in pilot, tower, or ground crew human factors and make proactive recommendations?

>does it always take a close call like this (or worse) to bring up such a proposal?

At this point yes. There is a saying that "FAA regulations are written in blood" as a result.

It seems conservative on the surface but at this point the current system is incredibly safe and making changes without clear huge upsides isn't worth the risk of unintended consequences.

Pilot attention is at a premium during landing. The computer shouting "localizer" at you during an approach because you're misaligned might be the straw that breaks the cognitive burden's back while you're trying to deal with another crisis.

Also, what Air France 447 showed us was that excessive automation becomes a problem when it is assumed to always work. The recommendations about more systems are all very well but there are 1000s of international airports who are either not going to have these systems or sometimes they won't work. What is critical is that pilots follow their checklists and that 1 action would have prevented this happening despite the fact that the other systems could have warned about it.

Making NOTAMs and the like more clear just seems like good design. They should certainly be able to prioritise information or only show what is relevant for a particular flight/aircraft/location/approach so you only have to read 3 things instead of 50.

As others pointed out there are fundamental problems with the system and usability of NOTAMs. Its a little absurd that theyre always used to blame the operator (“should have read the NOTAM”) when a flight briefing might have 10s or 100s applicable ones and a VAST number are junk. It reeks of retrospective blame seeking.

That said there is some traction around this. And some interesting alternative takes (https://opsfox.net), which have their own problems.

How difficult is it to see what's happening on the ground at night from a plane? I've never been in a cockpit at night. I'd have to hope that taxiways are marked in some explicit way to make it clear that they're not runways, like with an X or some pattern of lights that's not on runways; or have all runways marked with an obvious pattern that indicates "land here" (so if you don't see that, you know you're not landing on a runway). Don't the runways have their names labeled in lights near the entrances? (Or is that only paint?)

I guess I'm surprised that it took so long for the pilots to see that there were other planes where they were landing.

Is it routine to keep binoculars in a cockpit? Could they have seen what was on the ground sooner if they'd looked through binoculars briefly (after noticing unexpected lights?)

The numbers are not labeled in lights. Although this pic is blurry, this is SFO with both runways open: https://i.redd.it/hguxuyztxzdz.jpg

The night of the accident, the left runway was closed and lights were turned off.. and they were expecting two runways and to land on the one on the right. There's still no excuse for this happening with two ATP-rated pilots that have done this type of landing thousands of times, but you can kinda see how it is possible at least.

This also isn't unheard-of. For example, Harrison Ford had a similar issue where he landed his private plane on a taxiway at an airport with parallel runways.

Large airports have fancy lighting schemes and are pretty obvious in good weather.

Small airports or bad weather at night are a different story - I would not fly into a new airport the first time in those conditions. Think 10x - 100x the risk.

Binoculars might be ok for the non-flying pilot to use, as long as he doesn't drop them on the floor and bind any rudder pedals.

"Localizer, localizer" seems too obscure - "Landing on taxiway" is much clearer.

It's interesting to listen to the radio chatter for this near-miss [1]. The air traffic controller misses a few chances to notice that something isn't right but when the pilot on the taxiway says in plain English, "Where's this guy going? He's on the taxiway" the problem immidately becomes obvious.

[1] https://youtu.be/ZW-ETmZU0u8

As a private pilot, small planes going down occasionally are just a fact of life. But this incident scares me. There were four planes lined up on the taxiway the Air Canada narrowly missed. The worst aviation accident ever was a collision of two 747s resulting in 583 deaths: https://en.wikipedia.org/wiki/Tenerife_airport_disaster - this could have doubled that number.

Is it plausible that that AC759 could have taken out all 4 other planes? I don't really have a sense of the distances planes are spaced out on taxiways at major airports.

Spacing between planes on a taxiway is a lot shorter than you would imagine. Depending on type, a 25-50 meters apart is common.

Keep in mind that these four planes were preparing for take-off, meaning that they were fully fuelled...

The KLM which had barely reached minimum takeoff speed and started to take off still flew 150 meters after it had plowed through the upper half of the PanAm and lost an engine and the main wheels. There's a lot of momentum in a moving plane, and they're stacked very close when holding on the taxiway

Would also double the record for number of vehicles involved in an incident too, I believe

I've been anticipating the release of this report this since I heard about the incident.

I work on an avionics product that's designed to prevent this exact scenario. We use the aircraft's position along with a database of runways and taxiways to determine if the aircraft is approaching the runway the pilot intends. If we determine the aircraft is landing, we issue a visual and aural alert to the pilot ("TAXIWAY LANDING" or "NOT A RUNWAY").

How do you account for winds and pressure differences, both of which will modify an approach. A plane may look like it is landing short, or to the left/right, when in fact the pilot is anticipating a change in wind as they drop in altitude.

Good question. We've run into similar issues while developing a related feature that helps pilots perform a stable approach. Sometimes pilots were performing a circling approach to a different runway, and a naive approach to determining lateral deviations would have caused a nuisance alert.

For this particular system, the alerting threshold is only met when we've determined that the aircraft is "landing". As I mentioned in another comment, I'm vague about this point because it depends on the way the aircraft manufacturer has configured this state machine. Sometimes we use throttle position, altitude, speed above Vref, gear position, height above threshold, etc. You're correct that conditions can slightly modify an approach, but we're confident that we can nail down "we're landing" closely enough to mostly eliminate nuisance alerts.

What's your market? Light aircraft? airliners?

Our market right now is general aviation and business jets. I believe our biggest customer is the Cessna Citation Longitude (max takeoff ~40,000lbs). I noticed the NTSB abstract had some recommendations for inclusion and certification of a system on a broader range of aircraft. Larger aircraft are required to have similar safety-related systems such as terrain, reactive windshear, etc so this could be the next step.

What is the chance of false positives (or negatives) due to out of date databases?

We do get some nuisance alerts from time to time, and part of my job is to investigate them and determine the cause. I did investigate one recently, however, where the cause was a runway that was "one-way". Most runways have two ends and the pilot/ATC determines which one to take off. EDDF (Frankfurt) RWY18 is only used for takeoffs and was causing nuisance alerts due to the way we search for the runway in front of the aircraft.

There's another runway (I don't remember where) where the direction in our database does not match what satellite imagery shows. I don't work in the database group but I was told that this can only be fixed by asking the airport manager to re-survey the runway. Short of that, we can't fix it even if we know it's wrong.

However, It's quite rare for them to be due to a database that's out of date -- most of the time it's due to an incorrect determination that the aircraft is taking off or landing.

Curious: what inputs feed your system? Presumably at least localizer and GPS, but anything else? What accuracy / confidence interval do you require in order to make a call?

We actually don't use the localizer to check for an incorrect runway or taxiway landing because the pilot can choose to do so and we don't want to issue nuisance alerts. Runways don't move often and we've found that a GPS and two databases (runways and airport ground features) are sufficient to determine if the pilot is approaching a runway or not.

We use inputs like GPS position, ground track, ground speed, altitude, and a phase-of-flight state machine that's determined for individual aircraft. This allows us to tailor the criteria for determining if the aircraft is landing for different aircraft. For some, we use throttle and landing gear position and others we use speed and altitude above the landing airport.

As for accuracy, we have minimum levels of GPS figure of merit where there's enough certainty to issue alerts. I don't remember the number off the top of my head, but in the US WAAS has made GPS quite accurate.

This is what is called a 'cascade of failures' in aviation - and the term is used elsewhere. Multiple failures resulting in a catastrophic failure. In your own life and goings about, it's sometimes interesting to do a failure analysis and you'll often find you experienced a cascade.

It's really unfortunate that the cockpit voice recorder was overwritten. I'm guessing they're using very old technology where storage space is at a premium.

They're still using the flashing X on closed runways - e.g. at SEATAC right now. The lights looked incandescent a couple nights ago while passing the X. So not very bright or visible.

This accident would probably have been one of the worst in history. Multiple passenger planes on the taxiway taken out by a passenger jet.

If anyone would like to listen to the ATC radio: https://www.youtube.com/watch?v=ZW-ETmZU0u8

OMG that is just pure gold! The United guy is SOOOO laconic. Like him speaking up on the open freq saved hundreds of lives and he sounds like he's reading yesterday's corn future prices.

Reminds me of the BA Flight 9 announcement:

> Ladies and gentlemen, this is your captain speaking. We have a small problem. All four engines have stopped. We are doing our damnedest to get them going again. I trust you are not in too much distress.

This is the hallmark of a career professional pilot.

I love hearing ATC radio. I think the guy who did the classic SR71 story about upstaging everyone on some California flyby noted this pretty well - even in the times of near peril pilots and ATC will be the cool-as-steel that keep everyone calm.

Another interesting one is the guy that was suffering from hypoxia.

"Unable... to control... altitude" "Unable... to control... airspeed" "Unable... to control... heading" "Other than that, everything A-OK!"

Guy landed safely after ATC realized it was hypoxia and ordered the flight to descend.

EDIT: link https://fearoflanding.com/accidents/accident-reports/hypoxia...

Another close call: https://youtu.be/ag1-7qnkjv0 ("Pinch Hitting a King Air")

And the recent case of Captain Maggie: https://youtu.be/B229-KLudTo

I was just about to post that link! VASAviation has a really good youtube channel with ATC radio recordings and high quality visualizations of all kinds of interesting events: https://www.youtube.com/user/victor981994

https://www.youtube.com/watch?time_continue=57&v=oF7FR7TjnME has an overview shot, showing just how low AC759 got to the airplanes on Taxiway C. (Look at the top of the frame.)

This one is interesting also:


The pilot sounds calmer than the ATC. Former fighter pilot.

I'm surprised that remote video cameras aren't used by tower controllers.

The perspective of the control tower makes it difficult to catch issues like this, but a simple video camera aligned with the center-line of each runway would allow controllers to more easily verify that planes are lined up as well as other possible issues like gear down & locked, etc.

> I'm surprised that remote video cameras aren't used by tower controllers.

There are software and hardware/network setups designed to do this now, in fact London City Airport is about to move to a fully virtual control tower, with a bunch of super high res pan/tilt/zoom cameras on a mast instead of an on-site tower.


what happens when the camera lens gets a smudge?

I'd assume like cameras used at sporting events the lens would be covered by a clear plastic film on a roll, which can be moved to get a clean section. A bit like those toilet seat covers in airports, ironically.

Your description also made me think of clear view screens used in some ships.


> covered by a clear plastic film on a roll, which can be moved to get a clean section. A bit like those toilet seat covers

What? Is that really a thing? Toilet seat covers on a roll? How would that even work?

It's surprisingly effective when it works, even if it looks clownish. Imagine a box at the back of the toilet, out of which the seat extends (full oval, not a horseshoe). The cover is like a sleeve made of shopping bag plastic, and it's fed from one side and pulled in on the other.

real as you and me https://youtu.be/Lfrhgh6ZlQI

These kinds of cameras usually live at the back of a tube that also serves as a sun shade and weather shelter.

If I had to make a wild guess, it's a climbable mast where they send somebody up with microfiber cloth and 99% pure isopropyl alcohol periodically to clean the lenses. Or the glass panel that's in front of the lenses.

> I'm surprised that remote video cameras aren't used by tower controllers.

I'm surprised that forward-looking infrared cameras aren't more commonly used in civilian aircraft, especially given the availability of staring arrays with decent resolution/framerate that don't require cooling.

Even without the requisite certification for use as a for-realsies landing aid, it's nevertheless a useful tool to get situational awareness in inclement weather / at night.

They are more commonly found aboard 'experimental' aircraft because of the certification issues you pointed out.

And high-performance infrared cameras are also probably one of those things that end up on arms control lists. (Since it's not just useful for spotting planes on your "runway", but also for spotting tanks or people on your hill requesting an ordnance delivery)

I have several of these emails (when ordering to Australia):


Your recent SparkFun order contains 1x KIT-13233: FLiR Dev Kit. This item is export controlled by the United States government. By law, we are required to gather the following information from you as the importer:

1) Do you intend to sell or send this item to anyone in any of the following countries: Cuba, Iran, North Korea, Sudan, or Syria?

2) Will this be used in any military applications?

3) Will you be the ultimate end user of this item? If no, please go to 3.a.

3.a) If you are not the end user, who will this item be sold or transferred to? Please include full name, physical address, end use and confirmation that they will not sell or transfer this item to any party in Cuba, Iran, North Korea, Sudan or Syria.

4) What is the end use of this item?

5) Where is the location where the item will be used?

Response to all five questions is necessary before we are able to ship your order. If we don’t hear back from you within one week, your order will need to be canceled.Please let me know if you have any questions. I hope you have a great day!

I've had this exact one for all sorts of kit. Digital pressure regulators, optics alignment hardware, etc.

In aviation these are called Enhanced Vision Systems (EVS) cameras and when paired with a Head Up Display (HUD) are called Enhanced Floght Vision Systems (EFVS). EFVS give ‘credit’ to land in bad weather using the camera and display to see the runway.

Anyways, in the last couple years the FAA has changed the regulations to allow EFVS for credit on Part 121 (passenger aircraft), you can read the rulemaking change proposal that was implemented [0].

The generation of camera that existed at the time of this rulemaking was generally a cryogenically cooled InSb sensor type. The main (certified for credit) cameras here were the CMA-2600/2700 from CMC Electronics (I worked on these) and the Kollsman EVS I and II. These were mostly on BizJets before the rules were changed , except for the FedEx fleet that had an exemption.

As the rules were changed, airport lighting started to change form incandescent to LED, which made those expensive and heavy InSb cameras useless at seeing the approach lights in bad weather.

Over the years microbilometer LWIR tech has improved, and while they won’t help you see any lighting at all, LWIR sensors will help you see the ground and have better weather penetration than other wavelengths.

All that leads to the next gen of cameras being fused multi sensor cameras, 2,3, or even 4 sensors. Rockwell Collins EVS-300 and Elbit’s Clearvision EVS are examples.

The non-certified for credit systems have mostly used microbolometers over the years although some have used SWIR sensors as well.

As others have said the ITAR comes into play with anything approaching usable frame rates and resolution, although these are being slightly relaxed. Also, don’t buy American and you don’t have this problem.

After camera-based systems they are database driven Synthetic Vision Systems (SVS) that would have helped greatly in this particular near disaster. Then you can get even fancier and fuse synthetic with infrared to get a Combined Vision System (EVS + SVS = CVS), while Elbit and Rockwell offer. Sometimes I miss working on these vision systems.

[0] https://www.federalregister.gov/documents/2013/06/11/2013-13...

I'm not operationally familiar with SFO. But based on experience at other airports, a taxiway at night can be a sea of black at even 100' above the ground. I would probably see airplane navigation lights at some point, but would I bet my life on it? No. So this incident has to be taken seriously.

However, I'm still not clear from this reporting how the crew mistook blue taxiway lighting for white runway lighting. This isn't supposed to be subtle, and I haven't ever experienced it to be subtle.

Edit: OK there is a partial explanation here that appears as though at the time they committed to taxiway C as runway 28R, the visual cue they relied on was airplane lighting looking like it was runway lighting. Airplane taxi and takeoff/landing lights are white, navigational lights are green or red. Strobes usually aren't on while on the ground or at least probably shouldn't be.

It's an interesting dilemma. And fatigue is a significant contributing factor as well.

Human consciousness tries to form a coherent model that integrates current sensory perception with knowledge. During an approach at night, scarce sensory input is interpolated to form a full model by the brain. The report uses the term "expectation bias" for this phenomenon. This drive for coherence runs below consciousness and we only notice when the coherence falls apart: When our sensory input cannot be reconciled anymore with the current model. We blink, shift our head, and our model rebuilds anew.

What happened here is that the pilots were locked into a wrong model: In an environment with very little light there is no visual source to compare taxiway lighting against. If we know from experience that a light is white, our brain will use this to calibrate. Without this ever becoming a conscious thought. Color correction happens automatically for us and we don't feel it. This process can easily lead to nominally weird things like blue lights being percieved as white.

Fatigue and stress would contribute to ward off looming decoherence such as "why are the lights on 28L redder than on 28R?" When under pressure, we go with the strongest signal. And that signal may very well come from our knowledge, not our perception.

This process can easily lead to nominally weird things like blue lights being percieved as white.


I've never experience blue taxiway lights being confused as white/runway lighting, even in the very common case of uncontrolled airport where they frequently do leave taxiway lighting on, and runway lighting off, and use pilot controlled lighting to enable the runway lights on demand.

I'd expect at SFO, like any other major airport I have flown into, does have both runway and taxiway lighting enabled as pilots are on final approach. So they do have a reference: the white runway lights versus blue taxiway lights, which on final approach are in the same field of view. And even on the lowest intensity setting, runway lights are brighter than taxiway lights.

Something about this particular situation definitely is off. I don't know if there's something operationally different about SFO's lighting. Or if the pilots really were that fatigued, and this is a seriously underestimated factor.

They expected to see two lit runways. They saw two lit lanes and their tired and distracted brain did the rest.

Obviously it's a rare occurrence otherwise airport lighting would have been adjusted to avoid the mistake.

I don't understand it either.

Here in Seattle, SeaTac airport has a parallel taxiway (Taxiway Tango) with a long history of being mistaken for a runway. The list of things tried so far to resolve the issue is interesting:


The very first thing I thought of upon reading your comment "The list of things tried so far to resolve the issue..." was "What about a big lighted 'x'?" then I clicked the link.

I'd like to know why they removed it.

Exactly my question. It's 9000+ feet long. Put a lighted red X on it every 1000 feet. Or, maybe Xes aren't the answer, but there must be some technique to produce a startling visual warning.

It's probably a situation like Chesterton's fence. There are probably good reasons for not having the Xes. At least I hope so, I hope it's not just "that costs money and then we have to maintain it". The money can't be an issue. How much would 500 or 1000 deaths cost? How expensive are all the other proposals which have been mentioned? Those aren't cheap either.

Indeed, the article does not give any reason for its removal, other than the fact that it was removed. My guess is that it was in disrepair and/or they deemed it ineffective since there were landings in the years preceding the removal.

make them jagged? (I'm half serious, if they are that much dangerous, then sorry for the people taxiing). They could also be made falsely jagged at night with the lighting system.

Who is to say that you won't now have planes coming down on a surface which constantly changes conditions, going from tarmac to grass and back again?

a google search for "taxiway landing" shows that this has happened quite a few times - with full passenger loads in some cases.

Watch this A380 landing at SFO, and watch the pilot's reaction when he's told he's cleared to land visually on 28L, but to expect the ILS for 28R.


What's up with that?

I watched way more of that than I should have. It was way more interesting than I thought it would be.

I would love to be an airline captain.

PilotsEyeTV is a really really neat thing. They sell entire box sets of whole flights. It's pretty cool that the European airlines allow this amount of camera work in the cockpit:


If you really want to see something captivating, watch this clip of a Swiss A340 crew handling an overheating engine in-flight. It's incredibly well done.


Wow, that Swiss A340 video was amazing. The pilots were just so chill. "Let's have some chocolate for the landing" LOL

Yeah, it kind of makes up for all the Romansh flying back and forth in that cockpit. =)

Reminds me how those "Simulator" games are so popular in Germany!

I'll hand it to the Germans. When they get into a hobby, they really get into it.

I give up--why shouldn't airplane strobes be on while they're on the ground?

Harms night vision adaptation of other pilots.

My SOP is beacon (red, top and bottom) on with aircraft battery bus on, strobes on only when on an open runway, landing lights on when cleared for takeoff (or lined-up and waiting in position for takeoff). That seems to be a fairly common SOP for aircraft with separate control of beacon vs strobe.

It communicates a small amount of additional information, but more relevantly, it doesn't leave you sitting next to another airplane with blinding strobes on, killing your night vision adaptation. (Cirrus drives me crazy with their single switch anti-collision lights which light the strobes and beacon together.)

Here's the FAA (recommended, but not regulatory) guidance on the topic: (See "Use of Aircraft Lights") http://www.faraim.org/aim/aim-4-03-14-209.html

I guess that sounds good, though it might sound less good after a mass casualty collision.

Wonder if we couldn't have "ground strobes" visible only to viewers looking downward at a significant angle.

No, because when there's fog, the strobe gets reflected within the cloud everywhere. This is SFO after all.

In the case of fog or other low clouds though, airplanes are overwhelmingly likely to be following a localizer or GPS/RNAV approach, making this type of incident much less likely.

Hmm. So why not use that technique _all_ of the time?

Well, AC759 should have been doing so, even on the visual approach to 28R. On this particular approach (in contrast to literally every other approach in the FMS's database), a manual tuning step was required and skipped.-[0]

If there were another approach that required manual tuning and it was not a visual approach, the crew would detect they didn't have a localizer/equivalent to follow and would identify the missed step, because that's the only way they could navigate to the airport/runway.

It's a specific combination of visual approach and the only approach that required manual tuning (and not having another aircraft to follow visually) that setup this possible outcome. Then, the crew visually acquires the wrong bit of pavement and there's no safety mechanism inside the airplane remaining unless/until the pilots see that they're lined up wrong or get a go-around call from ATC/another aircrew.

[0] - Quote from the NTSB report: The FMS Bridge visual approach to runway 28R was the only approach in Air Canada’s Airbus A320 database that required manual tuning for a navigational aid, so the manual tuning of the ILS frequency was not a usual procedure for the flight crew.

They ruin other pilots' and crewmembers night vision.

Sokoloff's post just before mine is more articulate on this. I learned the same: strobes on when cleared for takeoff.

"The cues available to the flight crew to indicate that the airplane was aligned with a taxiway did not overcome the crew’s belief, as a result of expectation bias, that the taxiway was the intended landing runway."

Damn. It didn't matter that the whole world was telling him it was a taxiway because he was so sure it was a runway.

Gotta hand it to the NTSB - they are thorough. This is just the abstract.

Also that they cite fatigue / circadian low as one of the core causes. I thought they might avoid it because of the potential cost implications for the airline industry.

Worth noting that the Canadian regulations are looser than USA in that respect - the Air Canada pilots were in compliance with home regs, but would have been in breach of FAA regs.

Other point is that the NTSB isn't generally particularly shy about calling out the need for expensive changes (one example being their recommendation for fuel tank inerting systems after TWA 800). They have no power to mandate their recommendations be implemented after all: that's the job of the FAA.

Why does a plane in US airspace not follow FAA regs?

Well, it does pretty much, just not that bit (14 CFR 117). In this Air Canada case it's governed by 14 CFR part 129, which covers operations by foreign air carriers in the USA.


EDIT - removed erroneous reference to 14 CFR 121.480-> which applies to cargo ops.

Yup - copied this to buffer for a comment:

> The flight crew’s work schedule for the incident flight complied with the applicable Canadian flight time limitations and rest requirements; however, the flight and duty time and rest requirements for the captain (a company reserve pilot) would not have complied with US flight time limitations and rest requirements (14 CFR Part 117).

> Cockpit voice recorder (CVR) information was not available for this incident because the data were overwritten before senior Air Canada officials became aware of the severity of this incident.

How long are recordings normally kept? I would assume at least a few days for mundane flights (storage is cheap, isn't it?) but any sort of incident similar to this I would assume they keep recordings for much longer, if only for internal purposes.

It's a loop recorder. When that aircraft was certified, only a 30 minute loop was required. If that aircraft was newly certified, it would now require a 2 hour loop.

Can they not sync the recording elsewhere once they land and every passenger has a high-bandwidth internet connection the second the wheels are on the ground? (Is it too hard to certify that a system that connects to the internet is safe?)

I think it's a matter of privacy rather than capability. The airlines unions don't want their conversations to be easily accessible.

The tower recordings are posted above.

>(storage is cheap, isn't it?)

Not on ancient planes where everything needs a billion certifications from all sorts of regulatory bodies.

There's other issues at play. Pilots hate the idea of more monitoring of the cockpit, and have been vocal in trying to prevent e.g. a video recording of the cockpit being part of the "black box", even though it would have helped to shed light on a lot of aviation incidents[1].

1. https://www.wired.com/2014/07/malaysia-370-cockpit-camera/

It also needs to survive a couple of years at the bottom of the ocean or a similarly extreme environment and still be readable afterwards.

Speaking with a fair level of ignorance, how hard would it be for some onboard computer to be freaking out that a landing attempt is being made way off course of a known runway?

Going to guess this is one of those things where, yes, we have all the requisite technology, but practical implementation is the hard part.

I get excited seeing stories about aviation on HN because it's something I work on every day. I've written a few comments above but I'm developer on the system you described (and more) being used in aircraft today.

With a GPS and onboard databases of runways and taxiways, we can determine if a pilot is attempting to take off or land on a taxiway. The key is preventing nuisance alerts when the pilot is just flying around and happens to be aligned with a runway. The "11 secret herbs and spices" depend on the aircraft but are usually speed and configuration (ie landing gear) based. Approaches are fairly predictable and we can use that to sequence through a state machine that tells us the pilot intends to land.

Here's a video of the system on a G1000 (meant for smaller planes) skip to 0:45ish https://www.youtube.com/watch?v=V2bswm0w4cY

Thank you for sharing. Really cool to see just how much real-time GIS there is in aviation these days.

Automatically monitoring an ILS approach is not that difficult from the cockpit if the frequency can be sent via ADS-B.

Beyond that it gets really hard because visual and circling approaches can be in any heading, as well as large cross-winds.

In this deviation, the pilots literally got lost. Kind of tough to do much about that.

As per https://www.ntsb.gov/news/press-releases/Pages/NR20180925b.a...:

>The full report will be available on the NTSB website in several weeks.

Are there any reports from passengers on the taxiway planes? I'd have to imagine an A320 flying 60' over your roof would cause a fairly large disturbance.

It's kind of baffling to me that there's a system for detecting collisions with other airplanes in the sky (TCAS) and a system for detecting collisions with the ground (EGPWS) but not system for detecting collisions with other airplanes on the ground. This easily could have killed a thousand people.

TCAS alerts about "conflicts" with other aircraft on take-off and landing all the time. It's just that those are almost always false positives, so they get ignored.

There are many aircraft on the ground at airports. Given the high approach speeds, this is probably a non-trivial problem.

SFO has ground radar systems (ASDE-X/ASSC) [1] which allows ground and tower controllers to see potential conflicts on taxiways and runways. Pure speculation, but I don't think that these systems' conflict logic take into account an aircraft trying to land on a taxiway.

[1] https://en.wikipedia.org/wiki/Airport_Surface_Surveillance_C...

Yeah the report mentions that the systems for detecting conflicts and mis-alignments don't really talk to each other.

Different problems, no? Trying to pick out an airplane against the ground is different from detecting an object in the air, which doesn't generally reflect radio back, and detecting the ground, which does reflect. The problem you're describing requires you to differentiate between two things that both reflect signals back to you. It's obviously do-able since the military does it all the time for targeting. However, I doubt it's cheap and there might be cheaper ways to do it that don't leverage the two systems you've mentioned.

TCAS uses active transponders, not radar reflection. https://en.wikipedia.org/wiki/Traffic_collision_avoidance_sy...

TCAS works via the transponders of other aircraft, which they actively transmit.

I think this is really just a case of flying into the ground. It just happened to be in close proximity to other aircraft. There are lots of systems that are available to avoid that (ILS, GPS etc.) which were not used in this instance. And a system for avoiding collisions for taxing aircraft is probably not much use at such high speeds.

Some airports do have a system [0] designed to track taxing aircraft and allow collision risks to be detected.


CVR overwrite is bad news here. The crew knew this was a serious problem when it happened, there's no way they should have allowed the CVR, crucial evidence of exactly what happened, to be overwritten.

Perhaps the full report will have more detail on this, as it stands I have to believe somebody made the decision to overwrite this because they thought it'd be better if it didn't exist. Whether that's the crew, their bosses or other staff, somebody managed to go from "serious incident, preserve evidence" to "maximum ass covering" and that person or persons are an obstacle to effective investigation and thus to safe air travel.

Subject aircraft, C-FKCK, is a 1991 A320, which is only required to have a 30 minute CVR loop. (I don't know which exact CVR it was equipped with, of course.)

By the time the aircraft goes around, is resequenced, completes the second approach, lands, taxis in, and shuts down, much of the initial approach setup conversation may have already been overwritten anyway. I wouldn't jump to the assumption that there was malicious intent in the fact that the CVR was overwritten. (I also wouldn't want the crew to stop the CVR recording only to have a second, more serious incident on the same flight that would then go unrecorded.)

If I understand CVR tech correctly, it's running and constantly overwriting the storage in a loop whenever the plane is powered up. A lot of airplanes are 10, 15, 20 years old. CVR are based on some pretty old late 1980s technology, it's not like a modern flash based storage system with effectively infinite capacity for stereo 44KHz audio (64, 128GB, etc). There's maybe 1 or 2 hours of recording time, max.

They actually do clearly state in this PDF that a degree of fault lies in the crew being compromised by fatigue. Operating for 19 hours straight definitely places an individual in an area where one can be fighting to pay strict attention, even when it really counts.

There's also plenty of communication with Air Traffic Control in the tower, and not a whole lot of gaps to introduce significant mystery, so it's pretty unlikely that there would be anything earth shattering on the cockpit voice recorder.

What I'd expect the CVR to provide good evidence of is the extent to which crew were actually doing the checks that should have caught this error, and their reaction when they belatedly discovered their true position.

Without even trying to mislead investigators a human will report a sanitised version of events when not cued by facts. "Then I scratched my arse and sniffed my finger" is not going to appear in this sanitised version of events even if true. Neither is "I glanced at the panel, I know I'm supposed to scan it, but usually anything important flashes red". The tower log helps, CVR is better, and CCTV better again.

It also really helps to get into the routine of analysing this stuff. If your aircrew routinely skip steps 4, 18 and 23 from the checklist, only finding that out when 200 passengers die in a crash is stupid. Let's find out now. That needs lots of heavy lifting, pilots have to trust the airlines not to fire them for ordinary human frailties, and airlines have to accept that their pilots are human and "But the manual says not to do that" doesn't make it stop happening magically.

You're right that fatigue is an obvious element here, and there are several technical countermeasures mentioned. I definitely don't want to come across as believing in a conspiracy to hide the facts, but on the other hand I struggle to believe that it took over 30 minutes to get this thing back down after the incident and so the overwrite was not a decision. And if we agree it was a decision, it was a seriously bad one that should not be repeated.

  I struggle to believe that it took 
  over 30 minutes to get this thing 
  back down after the incident
If your inferring that a 30 minute offset was introduced, in order to dump the stale audio buffer, before permitting access to the CVR instrument on the ground, I'd offer the idea that there were 4 other planes on the taxiway, waiting to take off, and that the intervening 30 minutes could be eaten up by letting those planes depart, based on an array of other factors.

A concerted effort to dump the recorder buffer would require the participation of more than just the flight crew, since it manipulates the standing traffic pattern. Not an impossibility, and we know that in-group "walls of silence" definitely accumulate in high-intensity industries, but the air traffic control audio suggests a really minor transgression of situational awareness, even if the circumstances are assuredly unforgiving ones.

In the radio audio, the pilot clearly states that he's noticing lights on what he imagines to be his runway, and asks for confirmation as to what they might be. The tower reaffirms that he's all clear. Only then does a pilot from one of the planes on the ground speak up, and complain.

At that moment, both ATC and the plane realize the error. It's arguable that the pilot anticipated a likely go-around, based on intuition that the lights he was observing were definitely other planes, and that what appears to be an extremely close call proves it's own margin of error as reasonable, given that the participants recovered from the error.

No, I'm _implying_ that I don't believe it took 30 minutes to get down. Based on the abstract I would say you're going to find out it came back down in under 30 minutes, and then "oops" they let the CVR be re-used, overwriting the recording anyway.

The text from the linked abstract says: "Cockpit voice recorder (CVR) information was not available for this incident because the data were overwritten before senior Air Canada officials became aware of the severity of this incident"

One of the nice things about working in the Web PKI is that this crap doesn't get you anywhere since keeping records is one of the requirements, and any time the options are "This issuer might be crooked or they might be incompetent" both are disqualifiying anyway, no need to figure out which it is to make a determination.

It lets me sleep easier, were people at Symantec reckless, or were they so greedy that they actively tolerated fraud? Don't care, not my problem now, let their shareholders worry about it.

Oh, I don't know about these particular ideas you're chasing. There are quite a number of independent, redundant resources that would be very difficult to align forgeries across and present as fraudulent data, worldwide, for weeks, months, and even years after the incident.

It just sounds like a fishing expedition for some kind of cover up that exploits an incredibly narrow gap, covered by a myriad of alternative information sources that are more than capable of sealing up an air tight sequence of events by other means.

Open, consumer information sources like flightaware.com alone had the answer to the questions you're asking, long ago.

The radar history for this event was paywalled here for days after the event:


The open, freebie log only gives like two weeks of charts, but the data you mention was lying around, available for like 8 months, to anyone with premium access to that particular service. And anyway, I'm sure it's just a convenience wrapper to public data that's still floating around to anyone willing to dig.

Here's another service, if you've got $500 bucks, and really would like to know right now:


And to reiterate, this is only costing money today on account of being late to the party.

Multiple systems surfaced this data to the world in real time, and openly back-logged for months after. I'd bet good money a wide audience fact checked exactly what you've suggested a long time ago (even here: https://news.ycombinator.com/item?id=14741605). But my opinion isn't very important when the information is available in a general sense, to anyone who really wants it, even now, more than a year later. Feel free to be suspicious about these resources being far downstream and non-official consumer products, but if that doesn't suffice, then other (much, much bigger) doubts occlude any potentially satisfactory assessment which won't get solved in a random thread by amateur internet detectives like us.

Operating for 19 hours straight

Nobody was operating for that length of time. The captain had been awake for 19 hours, the first officer had been awake for 12 hours.

Granted, most people can't function anywhere near their peak performance after being awake for 19 hours. Also there's not enough information in the summary to know the captain's recent history; how much rest time did he have before that flight?

To call attention to my use of the word "operating" in this context, and highlight strict use of the word as ...technical(?) vocabulary, to form a rationale for censure of the general thrust of my statement, is pretty pedantic, and a semantic criticism at best.

Nonetheless, I'll put forward, that even if a direct flight from Toronto to San Francisco only takes about 5 or 6 hours, 19 hours, and rounding out the time of day to midnight as a circadian cue, definitely puts the pilot 3 hours past the usual 16 hour limit for bedtime (right, 12 hours for the co-pilot, okay). Both the pilot and co-pilot probably had at minimum another entire hour before they'd even have a chance to wind down and turn in for the night, so pile up the anticipatory nature of having to deplane and get to a hotel room onto their psychological state.

Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact