John Samford's Blog

I have a number of people asking me as a pilot what could possibly have gone wrong to cause the crash in San Francisco yesterday. Well, it could have been any number of things. But it is possible to look at what actually happened and speculate on a few things that could cause it.

I have gone back and looked at the track of the flight on a site called FlightRadar24. What set up the problem, if the data is correct, was that the plane was doing a very fast descent to the final approach course and for some reason failed to slow the descent and get into a stabilized approach at the appropriate time. It was descending at 1,536 feet per minute when it was only 1,000 feet above the water which would mean that it was only 40 seconds away from hitting the water. Somewhere during that last minute, the pilot pulled the nose up sharply to stop the descent and try to make it to the runway. However, the sharp pullup without any additional power slowed the plane down to 85 knots while it was still falling at 768 feet per minute 75 feet above the water, and a stall warning went off only four seconds before the tail hit the breakwater. Normal final approach speed should have been around 135 knots with a roughly 600 feet per minute descent.

So what could have caused the rapid descent to go on too long? Any number of things. The altimeter could have malfunctioned or been set incorrectly so that the crew, over water, thought their altitude was higher than it was. The crew could have thought the autopilot was set to level them off at a higher altitude and not paid attention as they descended too low. Alternatively, they could have thought the autopilot would automatically join the ILS glide slope to level their altitude, when the glide slope was not functioning for that runway yesterday.

Any of these situations could have been caused by an equipment malfunction, but that is hard to explain with the redundancy built into these planes, and with at least two and possibly four pilots in the cockpit. What is worse, if the problem was noticed 20 or 30 seconds before impact, there should have been time to apply power, pull the nose up, get the gear up, and go around. However, at seven seconds before impact when someone called for a speed increase or at four seconds when the stall warning went off, there was no time left.

It has been pointed out that it is sometimes difficult to have good depth perception or judge altitude on a visual approach over water. But this plane was on short final, and every pilot knows what the runway should look like in that configuration. I have no idea what happened on the flight, but it is very difficult for me to understand why the problem was not noticed earlier.

Update on Sunday, July 7, 2013 at 10:04PM by

John Samford

Here’s some additional tracking information of the flight from FlightAware. Each line shows the time, latitude, longitude, course (West), airspeed in knots and mph, altitude, and vertical speed in feet per minute.

01:25PM 37.5548 -122.2230 298° West 195 224 3,100 -1,140

01:25PM 37.5614 -122.2390 297° West 190 219 2,800 -1,500

01:25PM 37.5669 -122.2520 298° West 187 215 2,400 -1,320

01:25PM 37.5730 -122.2660 299° West 187 215 2,200 -1,080

01:26PM 37.5785 -122.2790 298° West 186 214 1,900 -1,020

01:26PM 37.5847 -122.2940 298° West 178 205 1,700 -1,020

01:26PM 37.5900 -122.3070 297° West 169 194 1,400 -1,380

01:27PM 37.5988 -122.3270 299° West 145 167 800 -1,380

01:27PM 37.6016 -122.3340 297° West 141 162 600 -1,320

01:27PM 37.6045 -122.3410 298° West 134 154 400 -900

01:27PM 37.6073 -122.3480 297° West 123 142 300 -840

01:27PM 37.6103 -122.3550 298° West 109 125 100 -120

01:28PM 37.6170 -122.3740 294° West 85 98 200 120

So you can see at 1:27 that the plane got down to 800 and then 600 feet above sea level while still descending at over 1,300 feet per minute. The descent rate alone is not alarming in a jet, but the descent would need to be arrested at the appropriate altitude. At 600 feet he was close to normal approach speed of 136 but needed power to slow or stop the descent. I don’t understand why a pilot would be pulling the nose up during this final minute without applying power, unless something wasn’t working or he somehow wasn’t aware of his altitude and distance to the runway. It seems to violate the very first principles you learn when you go out in a little Cessna and do stalls.

Update on Tuesday, July 9, 2013 at 10:32AM by

John Samford

Some of the data we are getting now from the NTSB shows that the speeds I got from FlightAware and FlightRadar24 were incorrect. Still, it is clear that power should have been applied much earlier to maintain both descent rate and airspeed during the approach. It is unclear whether the “auto throttle” was disengaged, but it has been stated that the autopilot was disengaged at 1,600 feet.

Much has been made about the ILS glide slope being off for this runway. But that had been widely published in Notices to Airman, and it would have been included in the weather briefing the pilots would have listened to before contacting San Francisco Approach Control.

Much has also been made about the pilot’s lack of experience in this particular plane, but little of the training to get rated in a new jet would cover the basics of a visual approach. When the runway is in sight, every pilot knows what it should look like if you are descending on a proper glide slope. And every pilot knows from their second flying lesson how to use pitch and power to maintain the required airspeed and descent rate for the particular plane. If there wasn’t any clear aircraft or instrument malfunction, it’s looking increasingly likely that the Crew’s situational awareness and resource management failed miserably.

(From Wikipedia: “Crew resource management or cockpit resource management (CRM) is a procedure and training system in systems where human error can have devastating effects. Used primarily for improving air safety, CRM focuses on interpersonal communication, leadership, and decision making in the cockpit. The training originated from a NASA workshop in 1979, which found that the primary cause of most aviation accidents was human error, citing the Tenerife airport disaster in 1977. CRM has since been adopted in different industries and organizations including fire services (to improve situational awareness on the fireground) and the maritime industry, where CRM is referred to as BRM (Bridge Resource Management) or MRM (Maritime Resource Management).”)

Update on Tuesday, July 9, 2013 at 08:37PM by

John Samford

Tonight it is coming out that the crew was relying on the “auto throttle” to maintain the correct approach speed. I don’t have any familiarity with these systems, so I won’t venture any guesses. Apparently, the crew felt they could pull the nose up and the auto throttle would keep speed at the targeted 137 knots. Whether this system failed or was in some kind of manual override position will remain to be determined.

One online news story stated the issue as follows:

Instead, investigators said the plane reached speeds as low as 103 knots and was in danger of stalling because it was losing lift before it hit the seawall.

The pilot told investigators he realised the auto throttle, similar to a cruise control, was not engaged just seconds before they hit. Their last-second efforts to rev the plane back up and abort the landing failed, although numerous survivors report hearing the engines roar just before impact.

Asked if the auto throttle was malfunctioning, Hersman said that is something investigators are looking into as they examine hundreds of parameters of data downloaded from the plane’s flight data recorders.

An overreliance on automated cockpit systems has figured in dozens of air crashes and incidents in recent years.

“Some people, if they believe the auto throttles are engaged and if they are used to flying with the auto throttle engaged, will not realize that the auto throttles are not engaged and will let the plane get pretty slow. That has come up before,” said John Cox, an aviation safety consult and former Air Line Pilots Association accident investigator.

National Transportation Safety Board chairman Deborah Hersman said the training captain who was instructing the pilot flying the Boeing 777 has told investigators he thought the auto throttle was programmed for a speed of 137 knots - the target speed the pilots had selected for how fast they wanted the plane to be flying when it crossed the runway threshold.
Instead, investigators said the plane reached speeds as low as 103 knots and was in danger of stalling because it was losing lift before it hit the seawall.
The pilot told investigators he realised the auto throttle, similar to a cruise control, was not engaged just seconds before they hit. Their last-second efforts to rev the plane back up and abort the landing failed, although numerous survivors report hearing the engines roar just before impact.
Asked if the auto throttle was malfunctioning, Hersman said that is something investigators are looking into as they examine hundreds of parameters of data downloaded from the plane’s flight data recorders.
An overreliance on automated cockpit systems has figured in dozens of air crashes and incidents in recent years.
“Some people, if they believe the auto throttles are engaged and if they are used to flying with the auto throttle engaged, will not realize that the auto throttles are not engaged and will let the plane get pretty slow. That has come up before,” said John Cox, an aviation safety consult and former Air Line Pilots Association accident investigator.