Photo: Aerial photo of wreckage, from NTSB AAR 10/04, figure 2.
We tend to focus our simulator training on engine failures, systems-related emergencies, and the obvious weather challenges. Weather in the simulator is usually summed up by "hot and heavy" or "ice and rain." We don't do a lot of realistic crosswind training and probably none of that realistically deals with gusts. Wind shear training boils down to a series of programmed problems. In other words, we train to what we are technologically able to train to in the box.
This mishap was caused by a captain who had a lack of finesse with the rudder on a very gusty crosswind which increased unpredictably due to windshear. The crosswind, the gust, the wind shear should not have ended up with the aircraft departing the runway laterally. A pilot with a reasonable amount of experience should have been able to handle it all. But we never test our captains under realistic conditions because we don't know how to do that in a simulator. We simply hope the pilot has seen enough of this before we hand him or her the keys to a multi-million dollar jet and the lives of passengers. The only solution would seem to be this: experience. But if you lack that experience, perhaps researching a few mishaps and talking to pilots who have this under their belts. My advice (for what it is worth):
- When in doubt, wait it out. In this case the winds went from 11 knots to over 25 in just a few minutes. Perhaps in a few more they could exceed the airplane's limits (they did not in this case) or die down.
- Just because everyone else is using the selected runway doesn't mean they are right. If the runway is wrong, say so on the radio. That can change things immediately. I was once at LaGuardia and was given the choice of using the runway everyone else was or waiting an hour. I said, on ground control frequency, "I will have to wait, a fifteen knot tailwind is against the limitations in my flight manual and just about every flight manual I've ever read." Once it was on tape, every airliner on frequency requested the runway be changed. And it was.
- When applying rudder on takeoff for a crosswind, you will usually need most of it when at low speed and a decreasing amount as you accelerate. Put what you think you need at the beginning of the takeoff roll and wait a few seconds before changing it.
- When adjusting your rudder input, add or subtract in small increments and wait a second.
- If you are making rudder reversals you are doing something wrong.
- Most importantly: you need to fly the airplane starting with brake release. Even when stationary, a wind means every air foil on the airplane is generating a force and you need to make sure that force is pointed where you intend.
One last note. The captain had a very good record and was probably genuinely surprised by the outcome of this takeoff. It is vitally important that we pilots acknowledge the fact we "don't know what we don't know." We must always be in learn mode and the best way to do that once you think you have the game mastered, is to research mishap reports from other pilots who also thought they had the game mastered.
What follows are quotes from the relevant regulatory documents, listed below, as well as my comments in blue.
- Date: 20 December 2008
- Time: 1818
- Type: Boeing 737-524
- Operator: Continental Airlines
- Registration: N18611
- Fatalities: 0 of 5 crew, 0 of 110 passengers
- Aircraft Fate: Destroyed
- Phase: Takeoff
- Airports: (Departure) Denver International Airport, CO (DEN, KDEN), USA;
- Airports: (Destination) Houston-George Bush International Airport, TX (IAH/KIAH), USA
Figure: Continental Airlines 1404 Wind / Rudder / Heading / Lateral load, from NTSB AAR 10/04, figure 1.
[NTSB AAR 10/04, ¶1.1]
- The pilots stated that they taxied toward runway 34R without event. About 1812, the DEN air traffic control (ATC) tower (ATCT) ground controller instructed the pilots to monitor the DEN ATCT local controller’s frequency while awaiting takeoff clearance. At 1814:27, the DEN ATCT local controller cleared the accident pilots to taxi into position on runway 34R and hold (to ensure adequate separation behind the airplane that took off on runway 34R at 1814:20). The pilots taxied onto the runway and completed the before-takeoff checklist while they held in position on the runway. According to CVR data, at 1816:16, one of the pilots commented, “what are the winds?” The accident captain noted to the first officer, “looks like...some wind out there.” The first officer replied, “yeah,” and the captain stated, “oh yeah, look at those clouds moving.”
- At 1817:26, the DEN ATCT local controller told the accident pilots that the wind was from 270° at 27 knots, assigned a departure heading of 020°, and cleared them for takeoff on runway 34R. (In their written statements, both pilots noted that although the wind velocity had increased from the 11 knots that had been reported by the ATIS, the tower-reported wind was still within the airline’s published crosswind guideline of 33 knots for a clear, dry runway like runway 34R.) The first officer acknowledged the clearance, and, as they began the takeoff roll, the captain stated to the first officer, “alright...left crosswind, twenty ah seven knots...alright look for ninety point nine.”
The 90.9 refers to the engine takeoff setting.
- At 1817:49, the CVR began recording the sound of increasing engine noise. The captain stated that, as the airplane accelerated, he shifted the primary focus of his attention from the thrust levers to outside visual references, keeping the airplane on the runway centerline. Meanwhile, according to post accident interviews, the first officer’s attention was primarily focused on monitoring the engine instruments, consistent with company policy. At 1818:04, the first officer advised the captain that the power was set at 90.9 percent. The first officer stated that after the power was set, he shifted his attention to monitoring the airspeed so that he could make the standard airspeed callouts, the first of which was at 100 knots.
- During the airplane’s initial acceleration along the runway centerline, information from the flight data recorder (FDR) indicated increasing right rudder pedal inputs, while the control wheel and column and their respective control surfaces were at their neutral positions. At 1818:07, as the airplane accelerated through about 55 knots, the airplane’s heading began to move left, and the FDR recorded the beginning of a large right rudder pedal input that peaked at 88 percent of its available forward travel about 2 seconds later. This 88-percent right rudder pedal input was followed by a substantial reduction, reaching about 15 percent by 1818:09.75. Almost simultaneous with the onset of this large rudder pedal input, the FDR began to record a left control-wheel input. The nose of the airplane moved to the right; however, at 1818:10, as the airplane was accelerating through about 85 knots, the airplane’s nose reversed direction and began moving back to the left at a rate of about 1° per second. This leftward movement of the nose continued for about 2 seconds and was accompanied throughout its duration by another substantial right rudder pedal input. This second large right rudder pedal input peaked at 72 percent of available forward displacement at 1818:11.75 and a speed of more than 90 knots and then decreased again, reaching 33 percent at 1818:13.25.
The NTSB report notes the pilot over corrected during his first application and then took out too much rudder. Things were complicated by the fact the gusty winds were changing during these rudder inputs. But even if the winds were steady, the rudder movement shows very poor technique. For a steady crosswind pilots should apply a little less than they think necessary and wait. If more is needed it can be, if less is needed a portion of the amount used should be taken out, not all of it. For a gust crosswind these techniques are even more important. With the reduction from most of the rudder to almost none of the rudder, the pilot also risks a rapid rudder reversal which can overstress the vertical fin.
- During this second large right rudder pedal movement (at 1818:12), the airplane’s left turning motion slowed for about 1 second, and then the nose began moving rapidly to the left again. A fraction of a second later (at 1818:13.25), the right rudder pedal was abruptly relaxed (reaching its neutral position about 1 second later). At 1818:13.5, the CVR recorded one of the pilots exclaiming, “Jesus,” and, at 1818:13.6, the FDR recorded the beginning of a transition from left control wheel input (consistent with crosswind takeoff technique for a left crosswind) to right control wheel input (crossing the control wheel’s neutral point at 1818:14). Although the pilot briefly made a small right rudder pedal input at 1818:14.25, the FDR did not record any more substantial right rudder pedal inputs as the airplane continued to veer to the left.
The control wheel input to the right is an indication of panic. More about this: Pilot Psychology / Panic.
- At 1818:15, the CVR recorded the first officer saying, “oh [expletive].” At 1818:17, the CVR began to record the sound of increasing background noise as the airplane left the runway, and, at 1818:21, the captain called to reject the attempted takeoff. FDR data showed engine power reductions, as well as activation of the brakes. Thrust reverser deployment began about 3 seconds after the airplane left the runway.
- The investigation revealed that the airplane departed the left side of runway 34R about 2,600 feet from the approach end and crossed taxiway WC and an airport service road before coming to a stop on a heading of about 315° in an area just north of DEN aircraft rescue and firefighting (ARFF) fire station #4. The airplane was still moving at a speed of about 90 knots when electrical power was lost, and the FDR and CVR stopped recording at 1818:27. Post accident interviews with passengers and crewmembers, as well as evidence from the crash site, indicated that, as the airplane crossed the uneven terrain before coming to a stop, it became airborne, resulting in a jarring impact when it regained contact with the ground.
- According to the captain, after the airplane left the runway and he subsequently initiated the rejected takeoff, they were “along for the ride.” Both pilots stated that there were a couple of “very painful” bumps before the airplane came to a stop. They indicated that they were somewhat dazed or “knocked out” for 1 or 2 minutes after the airplane stopped and made no immediate attempts to get up or leave the cockpit. The first officer stated that he could hear activity from the cabin and considered making an announcement, but he was hindered because the cockpit was completely dark. By the time the pilots left the cockpit, the cabin crew, assisted by some deadheading pilots, had evacuated all of the passengers. The first officer and a deadheading captain were the last to exit the airplane.
[NTSB AAR 10/04, ¶1.5.1]
- The captain began his aviation career when he joined the U.S. Navy in 1979, and he had about 4,500 hours of flight experience when he left active duty in 1993.14 At the time of the accident, the captain had flown about 13,100 total hours, including about 6,300 hours in the 737.
- When asked to rate the captain’s flying skills on a scale of 1 to 10 compared to other pilots with whom he had flown, the first officer rated the captain as a 9 and stated that he was very competent. Two other first officers who had flown with the captain indicated that his performance was typical of a Continental captain, one saying that he was “by the book” and that it was good to fly with him.
- A review of Continental’s records for the accident captain revealed no evidence of training or performance deficiencies. A search of the captain’s FAA records revealed no FAA enforcement actions, incidents, or previous accidents and no history of failures or retests for FAA airman certificates and/or ratings. A search of the National Driver Register found no record of driver’s license suspension or revocation.
The captain had a flawless record, but we don't know about his gusty crosswind capability. Simulator technology doesn't really test pilots realistically and it isn't a training item in most curricula.
[NTSB AAR 10/04, ¶1.7]
- A review of the ASOS 5-minute weather observations around the time of the accident showed that 11-knot winds were reported at 1815:31, and, 5 minutes later (at 1820:31), the winds were 24 knots with gusts to 32 knots. A review of the ASOS 1-minute wind data indicated that, at the time the airplane departed the runway, the wind was from 282° at 18 knots with gusts to 23 knots. The maximum ASOS 1-minute wind (277° at 36 knots) was recorded about 1823.
- To detect low-level windshear conditions around airports, the FAA installed basic low-level windshear alert systems (LLWAS), consisting of a centerfield sensor and five additional sensors located around the airport’s periphery at 110 U.S. airports with ATCTs. Since the initial installations, the FAA has improved LLWAS systems, upgrading software and hardware, integrating the system with an airport’s Terminal Doppler Weather Radar (TDWR) and Integrated Terminal Weather System (ITWS), and adding sensors along runway approach and departure corridors.
- DEN is equipped with the LLWAS network expansion rehost system (LLWAS-NE++), the most advanced LLWAS system. The system is designed to continuously collect and analyze wind data collected by 32 remote sensor stations located on and around the airport.
When Federico Peña became Major of Denver in 1983, he set out to close Stapleton IAP in an effort to have the city buy land owned mostly by his family for a new airport. Thus Denver's main airport went from what many considered the epicenter of wind shear activity in the United States to an even worse location. To its credit, the FAA has been pouring windshear detection equipment into the airport.
- At the NTSB’s request, the National Center for Atmospheric Research (NCAR) conducted an in-depth review of the wind data reported by DEN LLWAS sensors around the time of the accident. NCAR produced an animation showing the LLWAS-recorded winds between about 1813 and 1823, which indicated that the winds across the airport were not uniform; the animation showed a band of strong westerly winds over the central portion of the airport, with lighter winds to the north and south. In their review of the accident-related wind data, NCAR personnel emphasized that, because the LLWAS wind samplings do not record wind gusts that may occur during the 10-second intervals between recorded samples, it is likely that peak wind gusts were stronger than the winds that were depicted.
[NTSB AAR 10/04, ¶1.10.4]
- DEN’s LLWAS-NE++ system continuously evaluates wind speed and direction information collected by the airport’s 32 LLWAS remote sensors, and, if windshear and/or microburst conditions exist, alerts are generated and displayed to air traffic controllers on the RBDT in the DEN ATCT. Wind information recorded by the LLWAS sensor #2 is displayed on the RBDT as the airport wind. The airport wind is a running 2-minute average of airport wind direction and speed recorded by sensor #2, with wind gusts, which is updated every 10 seconds and is displayed on the DEN ATCT RBDTs. The LLWAS sensor #2 is about 3,310 feet northeast of the approach end of runway 34R, at 110 feet agl.
- Pilots departing DEN obtain general wind information from the ATIS broadcast by the DEN ATCT ATIS before taxiing for takeoff. Additionally, the DEN ATCT local controllers provide departing pilots with runway-specific wind information when they issue the flight’s takeoff clearance. The controllers obtain the runway-specific wind information (as well as windshear and/or microburst information, when applicable) from the RBDT in the ATCT, which is configured to display wind information recorded by the LLWAS sensor closest to the departure end of each departure runway for which that controller is responsible. If a runway is also being used for arrivals, the RBDT will display both approach and departure runway end wind information.
- The DEN ATCT local controller did not provide the airport wind to the accident pilots when he issued their takeoff clearance; rather, he issued the runway 34R departure end wind information, which in this case was reported by LLWAS sensor #3. It was common practice for DEN ATCT controllers to issue departure runway end winds to departing aircraft.
One can debate the tower's role in runway selection and the role they played in providing wind information that may not have been the most pertinent of several to choose from. The winds were changing so rapidly it may not have made a difference, given the captain's stick and rudder performance. (More about that soon.)
[NTSB AAR 10/04, ¶22.214.171.124]
- The NTSB used available data (measured FDR data and airplane acceleration biases determined from the ground path integration) to estimate the winds that were present during the accident sequence. (Boeing also estimated the wind conditions that were present during the accident sequence, using several different wind estimation methods, which produced results similar to those obtained by the NTSB.) The NTSB’s wind extraction results estimated that the winds at the time of the accident varied between 30 and 45 knots from the west, resulting in an almost direct crosswind for runway 34R and a crosswind component that varied from 29 to 45 knots. A peak gust of 45 knots occurred at 1818:12, about the same time the FDR recorded the right rudder pedal beginning to move aft from a position about 72 percent of its available forward travel, reaching a near neutral position at 1818:13.75. The first recorded main landing gear tire skid marks are also estimated to have occurred about this time.
- Performance calculations indicated that the airplane’s rudder was capable of producing enough aerodynamic force to offset the weathervaning tendency created by the winds the airplane encountered during the accident takeoff roll.
[NTSB AAR 10/04, ¶126.96.36.199]
- After crosswind takeoff procedures were demonstrated by Continental training managers, the simulator was set up to replicate the conditions (for instance, darkness, DEN runway 34R, airplane weight, and outside temperature) of the accident takeoff. The five ATP-rated members of the investigative team then performed crosswind takeoffs with left crosswinds of 0, 25, and 35 knots. The flying pilots were informed about the wind condition before each takeoff and rated the difficulty of each takeoff upon completion. On average, the pilots found these conditions to be “very easy” (the 0-knot crosswind takeoff), “neither difficult nor easy” (the 25-knot crosswind takeoff), and “slightly difficult” (the 35-knot crosswind takeoff).
- After completing takeoffs in all four crosswind conditions, some participants stated that the task did not seem that difficult overall. They also stated that the simulator did not accurately reflect lateral forces, nor did it provide as good of a “seat-of-the-pants” feel for wind gusts as an airplane would.
- Two of the ATP-rated participants tried to take off with a 60-knot simulated crosswind and were able to do so without “crashing” the simulator. They stated that the 60-knot crosswind required more right rudder correction than the other four crosswind conditions, but they indicated that they felt they had more than enough rudder authority available to accomplish the maneuver.
[NTSB AAR 10/04, ¶2.2]
- The NTSB concludes that mountain wave conditions were present at the time of the accident and resulted in strong westerly winds and very localized, intermittent wind gusts as high as 45 knots that crossed the airplane’s path during the takeoff ground roll.
[NTSB AAR 10/04, ¶2.2.1]
- The first of the captain’s large right rudder pedal inputs resulted in an apparent overcorrection of the airplane’s heading after about 1.5 seconds. Because the captain had no way to distinguish the effect of his inputs from the effect of the variable crosswind component, he likely believed that this large right rudder pedal input exceeded the amount of rudder correction that would be required to compensate for the crosswind.
[NTSB AAR 10/04, ¶188.8.131.52]
- Basic air traffic procedures for runway selection contained in FAA Order 7110.65 state that, “except where a runway use program is in effect,” ATC personnel should use the runway most nearly aligned with the wind unless use of another runway “will be operationally advantageous, or is requested by the pilot.” DEN ATC did not have a formal runway-use program; however, according to DEN ATC management personnel, they had an unofficial runway-selection policy, which would use the runway configuration that provided the greatest operational advantage for the airport at crosswind speeds up to 20 knots. This unofficial policy also indicated that DEN ATCT personnel were to consider using a different runway when requested by a pilot or when crosswind speeds exceeded 25 knots. Requests for alternate departure runways were rare at DEN and mostly occurred when crosswinds exceeded 30 knots.
[NTSB AAR 10/04, ¶3.1]
- The captain’s use of tiller and full right control wheel in the 3 seconds before the excursion likely resulted from acute stress stemming from a sudden, unexpected threat, perceived lack of control, and extreme time pressure.
- The unexpectedly strong and gusty crosswinds the airplane encountered as it accelerated during the takeoff roll made maintaining directional control during this takeoff a more difficult control task than the captain was accustomed to dealing with; however, had the captain immediately reapplied significant right rudder pedal input as the airplane was continuing its left turning motion, the airplane would not have departed the runway.
- Although the Denver International Airport air traffic control tower local controller followed established practices when he provided the accident pilots with the runway 34R departure end wind information with their takeoff clearance, he did not (nor was he clearly required to) provide information about the most adverse crosswind conditions that were displayed on his ribbon display terminal; therefore, the pilots were not aware of the high winds that they would encounter during the takeoff roll.
- If the accident pilots had received the most adverse available wind information (which was displayed as airport wind on the Denver International Airport air traffic control tower local controller’s ribbon display terminal and indicated a 35-knot crosswind with 40-knot gusts), the captain would likely have decided to delay the departure or request a different runway because the resultant crosswind component exceeded Continental’s 33-knot crosswind guidelines.
[NTSB AAR 10/04, ¶3.2]
- The National Transportation Safety Board determines that the probable cause of this accident was the captain’s cessation of right rudder input, which was needed to maintain directional control of the airplane, about 4 seconds before the excursion, when the airplane encountered a strong and gusty crosswind that exceeded the captain’s training and experience.
- Contributing to the accident were the following factors: 1) an air traffic control system that did not require or facilitate the dissemination of key, available wind information to the air traffic controllers and pilots; and 2) inadequate crosswind training in the airline industry due to deficient simulator wind gust modeling.
Abnormal Procedures & Techniques / Windshear
Pilot Psychology / Panic
NTSB Aircraft Accident Report, AAR-10/04, Runway Sode Excursion During Attempted Takeoff in Strong and Gusty Crosswind Conditions, Continental Airlines Flight 1404 Boeing 737-500, N18611, Denver, Colorado, December 20, 2008