March 2006 Volume 49 / Number 3
Landmark Accidents: Dark and Stormy Night
There are some nights when it's just better to be on the ground. But airline crews often have little choice about when and where they fly. It was just such a night on November 12, 1995, when American Airlines Flight 1572, an MD-80-series aircraft, was scheduled to fly from Chicago to Hartford, Connecticut's Bradley International Airport.
While a light general aviation aircraft probably would not have been out in such conditions, altimeter settings and attention to detail at minimums apply to all instrument pilots anytime they fly in the clouds or at night. The design of instrument approaches is not something we think about, and it seldom contributes to an accident, but on this night it couldn't have been more critical.
The weather that evening certainly would have made one wish to be fireside. The National Weather Service surface analysis showed a deep low-pressure system over Quebec with an occluded front extending south across eastern New York state. A secondary low-pressure center was located over New York City with strong southerly winds ahead of the front and lots of rain. Airmets and sigmets showed moderate to severe turbulence, moderate to severe icing, low-level wind shear, and strong surface winds.
Flight 1572 departed Chicago for Hartford at 11:05 p.m. Eastern Standard Time, an hour and 40 minutes late, delayed by weather and connecting passengers. After takeoff, the American Airlines (AAL) dispatcher, via the Automatic Communications and Recording System (ACARS, a communication system that, among other functions, allows the crew to exchange text messages with the airline dispatch), provided the crew with the latest weather conditions at Hartford. The remark "Presfr" (pressure falling rapidly) was included in the message. This often-overlooked remark was to assume crucial importance.
At the time of the accident AAL's policy was to set the crew altimeters to QFE (an altimeter setting corrected to indicate the height above field elevation, not mean sea level) when below 10,000 feet. At this setting, the altimeter will read zero on the ground, and on a standard Category I ILS, the altimeter always reads 200 feet at decision altitude (DA). The third altimeter, located prominently in the center panel of the flight deck, is then set to QNH, the local altimeter setting that is normally used in the United States to allow compliance with air traffic control (ATC) instructions such as assigned altitudes. American has since changed its procedure but not because of this accident.
During the descent, two ACARS messages were sent by dispatch concerning the Hartford weather. The first message at 12:30 a.m. provided a QFE altimeter setting of 29.23 inches of mercury for the primary altimeters. The standby altimeter indicating the normal feet above mean sea level (QNH) at Hartford was set to 29.42. A follow-up message was sent at 12:31 a.m.: "Attention, [captain's name]. Have information from BDL [Bradley International] Tower that aircraft have been making landings but have advised of turbulence and low-level wind shear on final. Winds are more aligned with runway at BOS [Boston] if that becomes necessary."
At 12:32 a.m., the cockpit voice recorder (CVR) recorded the last part of the automatic terminal information service (ATIS) information Victor for Bradley International with an altimeter setting of 29.50 inches. At 12:33 a.m., the en route controller provided the Bradley altimeter setting as 29.40 inches. The first officer (FO) acknowledged the altimeter setting and the descent clearance. He then listened to the entire ATIS information Victor message at 12:34 a.m. and mentioned that it was about an hour and a half old. At 12:38:45, as part of the before-landing checklist, the FO asked, "Altimeters?" The captain replied, "29.50." The FO stated, "They called 29.47 when we started down...whatever you want." The captain replied, "OK." This was incorrect, as the final setting given by ATC was 29.40.
The crew briefed the VOR approach to Runway 15. At 12:42:48 the captain said, "174's the [airport] elevation so 29.23 [the altimeter setting when corrected to QFE — as provided by the ACARS message]. Set and cross-checked." There was some additional discussion regarding the altimeters, but the QNH setting used was 29.47 — this introduced an error where the altimeters would read about 70 feet too high.
The flight contacted Hartford Approach control at 12:43:41, was told to expect the VOR Runway 15 approach, and was subsequently cleared to 4,000 feet. The winds were 170 at 29 knots gusting to 39 and a vector for the final approach course was issued. At 12:49:41, Flight 1572 was cleared for the approach and the crew was notified that the Tower was temporarily closed. Because of the high winds and heavy rain a window was bowing inward and leaking. The supervisor decided to evacuate other controllers from the Tower but remained in the cab to assist the flight.
The MD-83 intercepted the final approach course at 3,500 feet and 15 miles from the airport while the captain prepared the airplane for landing. He selected the VOR/LOC (localizer) mode for the autopilot. However, it was unable to hold a 30-degree course correction and the captain changed the autopilot mode to HDG SEL (heading select) to manually recapture the inbound course. After crossing MISTR, landing flaps and the gear were lowered. At 12:51:44, the aircraft began the descent to 2,000 feet in VERT SPD (vertical speed) mode for pitch control. Two thousand feet was the final-approach-fix (FAF) crossing altitude as the airplane encountered moderate turbulence and very heavy rain.
At 12:53:27 the FO transmitted, "Hey, Tower. American Airlines Flight 1572. We're on a six-mile final for Runway 15." The tracon supervisor in the Tower cab replied, "Landing is at your discretion, sir. The winds 170 at 25, peak gusts 40, and, ah, the runway does appear clear, you can land and taxi to the gate at your discretion." The supervisor subsequently issued a wind-shear alert.
The autopilot captured and leveled the aircraft at 2,000 feet, prior to arriving at the final approach fix at DILLN. After passing DILLN, the captain told the FO to set the missed approach altitude of 3,000 feet in the flight guidance control panel, and they started down to the minimum descent altitude (MDA) of 908 feet msl using the VERT SPD mode. At 12:54:22, the captain asked the FO to give him "a thousand down." (Program the vertical speed to a 1,000-foot-per-minute descent.)
At 12:55:06, the first officer stated, "There's a thousand feet...cleared to land." At 12:55:11, he further stated, "Now 908 is your, uh...." This was a reference that the aircraft was at minimums (MDA). The captain replied, "Right." The FO later told investigators that he had ground contact "straight down" and he began looking for the field visually. When he looked back at his altimeter and saw that the airplane was descending below the MDA, he stated, "You're going below your...." At this point the captain pushed the autopilot altitude hold button.
A "sink rate" warning from the ground proximity warning system (GPWS) occurred approximately four seconds prior to the first impact with trees. At 12:55:30 the CVR recorded a sound of impact. The captain later stated that he heard a "loud report," followed by severe turbulence. The aircraft struck numerous trees on Peak Mountain Ridge, approximately 2.5 nautical miles northwest of the approach end of Runway 15. The first tree was broken off at 40 feet above the ground, or 770 feet msl.
The captain immediately called for a go around and "firewalled the throttles," according to the report. The crew raised the flaps to 15 degrees, and raised the landing gear. The on-board wind-shear warning system and the GPWS activated again. The turbulence stopped, and the captain called out, "Left motor's failed." The airspeed started to drop and the aircraft began a slow descent. Just when things couldn't get any worse the rain stopped, and the FO spotted the runway.
The right engine was not sustaining full thrust, and at 12:56, the captain said, "Tell 'em we're going down." The FO called, "Tower, call for emergency equipment...we're going down on the runway." The FO then said, "You're going to make it." He asked the captain if he wanted to lower the landing gear and then lowered it on his own.
The captain requested 40 degrees of flaps to create a "balloon effect" in hopes of reaching the runway. It was a nonstandard procedure that worked beautifully, under the circumstances. There obviously needed to be enough energy to overcome the extra drag with lift. The aircraft clipped a tree near the end of the runway, destroyed most of the ILS antenna array located at the end of the safety overrun, and rolled to a stop on the runway, near midfield. The 78 crew and passengers then evacuated the aircraft with only two minor injuries. It was a remarkable recovery.
The 8,000-hour captain had more than 4,000 hours as an FO and pilot in command in the MD-80 series and the FO had nearly 2,300 hours of MD-80 time, so this was an experienced crew. It took $9 million to repair the aircraft since both engines were destroyed from ingesting tree parts with substantial damage to the fuselage, wings, and flaps; $75,000 was required to fix the ILS antenna and associated equipment.
The observations at Hartford made by a contract weather observer just prior to and after the accident: 11:52 p.m., 2,200 feet scattered; measured ceiling 2,800 feet overcast; visibility five miles; light rain; temperature 61 degrees Fahrenheit; dew point 57 degrees F; winds 160 degrees at 28 knots, gusts to 40 knots; altimeter setting 29.42 inches; peak wind 160 degrees at 42 knots at 11:18 p.m.; pressure falling rapidly. This was transmitted to the Tower at 11:53.
At 12:51 a.m., about four minutes prior to the accident: 1,700 feet scattered; measured ceiling 2,800 feet overcast; visibility three miles; moderate rain; temperature 61 degrees F; dew point 58 degrees F; winds 170 degrees at 25 knots, gusts to 40 knots; altimeter setting 29.35; pressure falling rapidly. Peak wind 170 degrees at 43 knots at 12:18 a.m. This observation was transmitted to the Tower at 12:57 a.m., after the accident. Note the altimeter setting, which is 0.12 inches lower than what the crew had erroneously set, and 0.05 lower than the center controller advised. This translates to approximately a 120-foot and 50-foot error, respectively. It was enough to make a difference, but not the only link in the accident chain.
Following the accident, the flight guidance system of an American Airlines MD-80 simulator was used to replicate the VOR 15 instrument approach to determine autopilot behavior in conditions similar to the accident environment.
In rough air, when the altitude hold was engaged, the simulator descended an additional 120 to 130 feet. It then recovered and climbed back to the selected altitude. This is worth noting for instrument pilots who have altitude hold features on their autopilots. Does your system allow the aircraft to "sag" below the desired altitude? How might it react in turbulence with possible downdrafts?
The GPWS in use at the time of this accident was limited in its capabilities and, with gear and flaps down, could only warn of excessive descent. An "enhanced" system that can issue warnings well in advance of a predicted impact is now required on all turbine-powered aircraft with six seats or more.
Approach design considerations
The instrument approach to Runway 15 was first published on February 9, 1989. The highest obstacle that could affect the approach was terrain at 739 feet with 80 feet of tree height added, for a total height of 819 feet msl. That is the ridgeline with trees that Flight 1572 initially struck.
On April 8, 1994, the Air Line Pilots Association (ALPA) sent a letter to the FAA stating that several GPWS warnings had occurred while descending from the five-nautical mile FAF to the 3.1-nautical mile visual descent point (VDP). This is defined as a point on the final approach from which normal descent from the MDA to the runway touchdown point may begin, provided that visual reference is established. ALPA was concerned about the proximity of the ridge. Later that summer the FAA determined that the VDP should not have been published, and it was removed by later amendment. As a result of NTSB recommendations after the accident, the FAA reinstated the VDP and moved it 0.7 miles closer to the runway threshold in August 1996.
The lowest portion of the visual approach slope indicator penetrated the required obstacle clearance plane over the ridge by 55 feet, in conflict with approach design guidance. In designing approaches, the FAA said, "Consideration should be given to induced altimeter errors and pilot control problems in precipitous terrain that may result when winds are 20 knots or more over such terrain." However, "precipitous terrain" is not defined and the FAA did not add terrain adjustments nor were there any concerns from pilots regarding the terrain. At the time of the accident, AAL's instrument approach charts did not contain any graphical terrain depiction, although there was a notation about the height of the obstacle. Several other carriers' charts did show the terrain and that is now the standard to help pilots visualize hazards that may be very close to the approach path.
The Tower and approach controllers obtained current altimeter information from the DASI (digital altimeter setting indicator), a self-contained instrument located in the Tower to provide current altimeter readouts. The DASI updates the automated radar terminal system so that altitudes on the controllers' radar are correct. ATC procedures require altimeter settings to be transmitted upon initial radio contact.
The record of keyboard entries shows that at 12:21:14 a.m., the approach controller entered 29.38 inches. At 12:44:34 he entered 29.34 inches; at 12:48:07 he entered 29.36 inches. The atmospheric pressure continued to drop and was calculated to be 29.24 inches at the time of the accident. Boston air route traffic control center issued Flight 1572 an altimeter setting of 29.40 inches at 12:33:30, about 21 minutes prior to the tree strike, and approximately 11 minutes before handing the flight off to the Hartford Tracon.
This was the last altimeter setting issued to the flight. The tracon approach controller did not issue a current altimeter setting to Flight 1572 upon initial radio contact with the flight at 12:43:49. At that time, the DASI indicated a setting of 29.38 inches.
Minimum Safe Altitude Warning (MSAW) is a computer program to warn controllers when an aircraft descends below a predetermined altitude. The Hartford MSAW was programmed to provide an alarm if an aircraft's Mode C transponder transmitted two signals when the airplane was at or below 1,050 feet msl. However, if the radar does not receive the Mode C transponder signal there is no warning. An MSAW alarm occurred at 12:55:39 when the aircraft was at 800 feet msl and after the first impact. Radar data showed no radar "hits" from the flight between 12:55:23 and 12:55:32. The impact occurred at 12:55:30. The ridgeline shielded the airplane from the radar. FAA technical personnel had attempted to optimize the radar tilt prior to the accident but later said that the ridge still obstructed radar coverage.
Not getting updated altimeter settings, through a combination of ATC's failing to provide them and the crew's not asking, was a contributing factor. The primary cause was the crew's failing to level at the MDA for a variety of reasons. The FO was focused outside the aircraft, looking for the runway during the level off rather than monitoring and backing up the captain. The captain failed to notice and arrest the autopilot's descent when the flight went below MDA.
The NTSB faulted the FAA on multiple points in the design and implementation of the approach procedure that failed to take into account the terrain. The current Hartford VOR 15 approach chart shows the MDA 412 feet higher than it was before the accident (1,320 msl now compared to 908 msl in 1995).
A new approach
As a result of this accident and others related to nonprecision approaches, many airlines began flying in ways that more closely resemble precision approaches. With modern flight management systems it is possible to fly a stabilized approach with a constant glidepath to a point near the MDA. Rather than the "dive and drive" method that has been used for so many years, it's safer to overfly intermediate level-off points at higher altitudes, while maintaining a stable rate of descent, rather than level off at intermediate fixes.
Some carriers add an additional 50-foot safety buffer to the MDA to allow the airplane to "round out" above the actual MDA before a missed approach climb. These are excellent practices for GA pilots to consider, and with the arrival of Wide Area Augmentation System GPS approaches providing vertical guidance, it will be much easier to accomplish. Standard airline procedure is to let the autopilot fly while the crew monitors. Most of the time this will result in lower workload and greater precision than hand-flying. But as with all crewmembers or hardware, it's up to us to verify. Is the other person or the equipment doing what was asked? This becomes vital as we transition to more technologically advanced aircraft.
How about increasing the margin? This was a nasty night and demanded excellence from everyone involved. As we have just seen, humans are not perfect and failure is not a satisfactory option. There is no penalty for being slightly high and the new FAA Instrument Practical Test Standards allow pilots to be 100 feet high at MDA and zero feet below. I'd much rather fly with a pilot who's a hundred feet high at minimums and might occasionally miss a really low approach than trust my life and aircraft to having everything work just right. Murphy's law was written for such occasions.
Bruce Landsberg is executive director of the AOPA Air Safety Foundation.
Links to additional information on Landmark Accidents may be found on AOPA Online.