At about 11:28 a.m. PDT on July 6, Asiana Airlines flight 214 crashed at the approach end of runway 28L at San Francisco International Airport (KSFO). Less than 24 hours after the accident, it's way too early to know what happened, but there are some signs that the aircraft was not flying a stabilized approach at any part of the approach to land, and ended up low and slow just before impact. Below, I outline the data that's available at this time, and why I think that's a possibility. But first, a disclaimer: There's not nearly enough data to determine the cause of the accident. The below is speculation, based on the data available to me, which is not even a few percent of the data that will be available to the NTSB. Most speculation this early in the process is wrong, precisely because so little data is available. So anything below that may appear to be stated as a fact is really a conjecture.
First, the conditions at KSFO were ideal. The METAR 32 minutes before the accident was
KSFO 061756Z 21006KT 10SM FEW016 18/10 A2982 RMK AO2 SLP097 T01780100 10183 20128 51005
That is, there was a 6 knot crosswind to runway 28L, with good visibility. The winds aloft were light, so wind shear doesn't appear to be a factor. Twenty seven minutes after the accident, the METAR was
KSFO 061856Z 21007KT 170V240 10SM FEW016 18/10 A2982 RMK AO2 SLP098 T01830100
Still a light crosswind, but with more variable direction. However, given how light the winds were, this would not likely have had a big impact on the landing. So weather doesn't appear to be a factor. Indeed, the weather was near perfect at the time of the landing.
There are reports that the ILS glideslope was unavailable, but that the PAPI (precision approach path indicator) for runway 28L was functional. There are some conflicting reports that the PAPI was unavailable as well. Aircraft were making visual approaches to 28L and 28R.
Using radar data from FlightAware, I plotted the altitude and airspeed of AAR214, as well as that of UAL852, another 777 which landed successfully only 10 minutes beforeAAR214. Both flights were long distance international flights (UAL852 originated at Heathrow, London), probably both full of people but without much fuel, so both aircraft would have had more or less the same weight on landing, and therefore the same reference approach speed, . It's instructive to compare the landing profile of these two flights. The first plot below is the altitude vs. distance from the touchdown zone for both aircraft. UAL852 is in black, and AAR214 is in red.
Note that UAL852 is on a nearly constant 3.2 deg glideslope. The PAPI for 28L is nominally 2.85 deg
, but because a 777 is a large aircraft with a large cockpit to wheel height, it would be typical to fly the approach a tad steeper than the standard glideslope.
On the other hand, AAR214 is 500 feet or so above the glideslope until about 4 nm out. At 3 nm out, the aircraft is quite high, on a 4.48 deg glideslope to the touchdown zone, or about 50% higher than it should have been. From that point, the aircraft descends rapidly, presumably to acquire the correct glideslope. At about 1.5 nm out, the aircraft crosses and then descends through the glideslope.
At the last reliable radar return, the aircraft is at 100 feet, 100 feet below glideslope. Note, however, that the radar returns are quantized to 100 feet, so the result may not be very accurate. Nevertheless, you can see that the descent rate on short final is very high, perhaps twice what would be expected for a stabilized approach.
It gets worse. The plot below is groundspeed for the two aircraft.
Because the winds are light and generally a crosswind, the speeds shown are probably within a few knots of the actual airspeed. Since we don't know the weights of the aircraft, we don't know what target approach speed () was. However, you can see that starting about 6 nm from touchdown, UAL852 slows from about 190 knots to about 145 knots. A typical for a 777 loaded with passengers but not much fuel is about 145 knots, so that makes a lot of sense. So UAL852 flew a stabilized approach, on the glideslope from 12 nm out, and slowing to for full flaps about 3 nm (1 minute) out.
On the other hand, AAR214 was never on a stabilized approach. Until about 30 sec before touchdown, it was high and fast. Only 3 miles out, it's 20 or 25 knots too fast, and 500 feet high. As a result, the pilot no doubt reduced power to intercept the glideslope from above. 1.5 nm out (nominally less than 40 sec from touchdown), he's finally on glideslope and at , but with a high sink rate on low engine power. If he applied power at that point, the engines would take some time (a few seconds) to spool up, and he would further sink below glide slope, slow down, or both.
The situation can be appreciated more precisely (but more technically) by looking at the total energy of the aircraft, that is, the sum of the potential energy due to altitude plus the kinetic energy due to velocity. The total energy is given by
where is the mass of the aircraft, is the acceleration due to gravity, is the height of the aircraft, and is the velocity. Because we don't know the weight of the aircraft, it's convenient to normalize the energy by , yielding the energy height
The plot below compares the energy height for the two aircraft:
Note that the energy of UAL852 decreases at a steady rate until about 6 nm out, where the rate of energy dissipation increases, because the aircraft is slowing. At about 3.5 nm out, the rate decreases, because the aircraft has hit its target approach speed and stops slowing down.
AAR214 has a much different trajectory. At about 3 nm out, the rate of energy dissipation increases a lot, because the aircraft is both too high and too fast. As a result, the power is reduced significantly, perhaps even to near idle, in order to simultaneously slow the aircraft and get it down to the glideslope. At about 1.5 nm out, it has about the right airspeed and altitude (and therefore energy), but the energy continues to decrease precipitously. If the pilot added enough power at this point, a safe landing might have been possible. But it takes several seconds for the engines to spool up, and the pilot may not have added enough power or done so early enough, so both the altitude and airspeed continue to decrease below their desired values. Indeed, at the last radar return, AAR214 would have been near its stall speed, and unable to pull up.
These data are entirely consistent with the eyewitness observations, which indicate that the aircraft approached steeply, and then tried to pull up when it got too low, but was unable to. It's also consistent with observations that the engines were powering up just before impact.
Whatever the reason the pilot flew this approach profile, it's clear that he never had the aircraft established on a stablized approach. FAA Advisory Circular AC 120-71 states that:
An approach is stabilized when all of the following criteria are maintained from 1000 feet HAT [height above touchdown] (or 500 feet HAT in VMC) to landing in the touchdown zone:
- The airplane is on the correct track.
- The airplane is in the proper landing configuration.
- After glide path intercept, or after the FAF, or after the derived fly-off point (per Jeppesen) the pilot flying requires no more than normal bracketing corrections2 to maintain the correct track and desired profile (3° descent angle, nominal) to landing within the touchdown zone. Level-off below 1000 feet HAT is not recommended.
- The airplane speed is within the acceptable range specified in the approved operating manual used by the pilot.
- The rate of descent is no greater than 1000 fpm. If an unexpected, sustained rate of descent greater than 1000 fpm is encountered during the approach, a missed approach should be performed ...
- Power setting is appropriate for the landing configuration selected, and is within the permissible power range for approach specified in the approved operating manual used by the pilot.
It appears that there is no point in the approach where the approach is stabilized. Indeed, the rate of descent at 600 ft was 1320 ft/min, well above the allowable descent rate. Standard procedure at most airlines would have required the aircraft to go around at that point. However, it's not clear that standard practice conforms to standard procedure, and pilots may be reluctant to initiate a go-around at a busy international airport on a clear day.
One other important factor may be that the glideslope signal of the instrument landing system (ILS) was out of service at the time of the accident. It's standard practice to use the ILS glideslope even when on a visual approach. Indeed, FAR 91.129 (e)(2) requires that, "Each pilot operating a large or turbine-powered airplane approaching to land on a runway served by an instrument approach procedure with vertical guidance, if the airplane is so equipped, must: (i) Operate that airplane at an altitude at or above the glide path between the published final approach fix and the decision altitude (DA), or decision height (DH), as applicable." Normally pilots would use the ILS glideslope for this purpose. They could have used GPS for vertical guidance (there is an RNAV (GPS) PRM RWY 28L approach), but may not have done so.
It's far too early to know the cause of the crash of Asiana Airlines flight 214. However, early indications are that it might be due to an unstabilized approach, which is a leading cause of approach and landing accidents. If the cause does turn out to be an unstabilized approach, it will be relatively straightforward for the NTSB to make that determination. The data recorder and cockpit voice recorder will have ample data to determine what decisions the crew made on final approach, what the state of the aircraft was throughout the approach, and what control inputs were applied.
At about 4:45 EDT today, the NTSB held a press conference, and it appears most of our conjecture is correct. The target approach speed was 137 kt (not 145). Just before impact, the power was indeed at idle, and the airspeed dropped "significantly below 137 knots and we're not talking about a few knots." Seven seconds before impact, a pilot called for increased power. At 4 seconds before impact, the stick shaker actuated, indicating incipient stall. At 1.5 seconds, a pilot called for a go-around, much too late, obviously. So it appears (so far) that our analysis is more or less correct.
Below is a screenshot of the last part of the final approach path:
Here is a link to the KML file for those who may want to visualize more of the flight path.