As the first bodies from Air France 447 are found, investigators suspect a terrifying mix of weather and technological weakness was responsible

When things go wrong at high altitude, one of the deadliest challenges for pilots is a phenomenon known as "coffin corner". This is the point, tens of thousands of feet up, where the margin for error in controlling a sophisticated modern airliner becomes tiny. Investigators are now wondering whether Air France flight 447, which disappeared last week with 228 people on board, may have flown into coffin corner never to escape.

For amid all the speculation and mystery, two events are clear in the worst aviation disaster for half a decade. At 3am BST last Monday morning flight 447, a four-year-old Airbus A330-200, reported that it had encountered "stormy weather with strong turbulence". Ten minutes later, the plane's autopilot disengaged, according to its automatic communications and reporting system (Acars).

Somewhere around 35,000ft, with storm winds raging and the plane buffeted on all sides, the crew found themselves trying to fly 230 tons of electronic wizardry by hand. At that altitude, it is far harder than passengers imagine.

Whoever was in the pilot's seat was looking at two computer screens, a host of other instruments and two rudder pedals - but no traditional hand controls. Instead, an A330 pilot reaches for a small joy-stick to one side. It looks a bit like the control for a games console. Through that "sidestick", the pilot flies the plane with electronic signals, rather than any mechanical linkages.

"It's tricky. At altitude big planes wallow about," said Roger Guiver, a former British Airways pilot. "It's like trying to steer the QE2 with a 2ft rudder." Jean-Pierre Albran, a former French air force pilot, said: "On a [Boeing] 747, you feel things with your hands. On an Airbus, you're just looking at screens."

On a British web forum for pilots, one contributor wrote: "Have any of you hand-flown an Airbus (or other aircraft heavy with fuel) at those flight levels even in smooth air? You are fighting to stay within the flying envelope . . . small margin for error on a good day, let alone a dark and stormy night.

"Take a jet aircraft and put it high, heavy, and run it through rough enough air and the laws of aerodynamics are waiting."

In those laws, speed is a crucial factor. The thinner the air, the more speed needed for the wings to maintain their lift. Too slow: you stall.

At the same time, the faster the air passes over the wing, the more the centre of lift moves backwards, pushing the nose of the plane down. Too fast: you nosedive.

At high altitude the gap between those two critical speeds gets narrower and narrower. That's coffin corner - and that was one of the crises facing the crew of AF447 as the plane plunged through the thunderheads in the early hours of last Monday.

It is now clear the crew, as they fought to stay airborne, no longer knew how fast their plane was travelling. According to Airbus and the accident investigators, the pilots' instruments were giving "inconsistent" readings of the plane's speed.

Did the crew or computer mistakenly think there was a danger of stalling? Did they power up, tipping the plane out of control and tearing it apart in the turbulence? Or did a violent updraft simply drive them too close to coffin corner?

Though no one yet knows for sure what destroyed the plane, investigators are concerned that it was not caused, as first suggested, by a lightning strike or a bomb or a meteorite. Instead they fear it was a fatal collision of high technology and the brute force of nature.

THE passengers who gathered at Rio de Janeiro's airport last Sunday evening for AF447 spanned more than 30 nationalities, including five Britons. Eithne Walls, a young doctor from Belfast, was heading home from a holiday with two Irish doctors. Alexander Bjoroy, an 11-year-old boarder at Clifton College in Bristol, was returning after spending half term with his family. Two Brazilians, Bianca Machado Cotta, a doctor, and Carlos Eduardo de Melo Macario, a lawyer, had married the day before near Rio; they were off on honeymoon to Paris. Silvio Barbato, conductor of the Rio symphony orchestra, was leaving behind his violinist girlfriend, Antonella Pareschi. Later she said of Barbato: "He always used to tell me, jokingly, that he would not simply die, but disappear."

If they worried about flying, they didn't show it. Not so a Swedish family who had a habit of travelling separately in case disaster struck. Christine Schnabl boarded AF447 with her five-year-old son, Philipe; she left her husband to follow a few hours later with their three-year-old daughter on another flight. On such choices lives turn. The route from Rio to Paris passes through an area known as the ITCZ - the intertropical convergence zone - where hot, humid trade winds meet, creating storms with updrafts that can reach 100mph. Weather maps for that night show "numerous cumulonimbus towers rising to at least 51,000ft", with thunderstorms and severe turbulence. But it was not exceptional.

Several hours after take-off AF447 went out of range of land radar and was heading across the Atlantic into the ITCZ. At night, pilots use onboard radar to spot storms ahead and divert sideways round them because they often rise too high to fly over. Did AF447 fail to spot a storm as it tried to find a way through the bad weather?

"Modern weather radar is very good," said Guiver. "You get a good return [signal] off water droplets. The strength of return determines the colour you see on your screen: green, amber or red. Red is the core of the storm."

At high altitude, however, the rain in a storm turns to ice crystals - and radar is much less effective at picking up ice.

Another danger is that, at the top of a storm, strong cross winds can blow turbulence out to one side, down-wind from the main updraft, in a formation known as an anvil.

"You always avoid a storm upwind of the core, or the anvil might catch you out," said Guiver. If AF447 had accidentally hit an anvil, tossing it beyond its normal flying parameters, it could have made the autopilot disengage.

Crew error cannot be discounted. An internal Air France report, seen last week by The Sunday Times, said "the reliability of these [fly-by-wire] aircraft has the consequence of reducing the pilots' appreciation of risk". It warned against "complacency" and recommended that training should include more time on flight simulators.

Yet it seems unlikely to have been pilot error alone, especially since planes regularly cross the ITCZ and the crew of AF447 was experienced. At 58, Captain Marc Dubois had 11,000 hours of flying time. So did a malfunction precipitate or contribute to trouble? On the face of it, the A330 had an excellent safety record, with more than 550 planes built and no passenger fatalities since it went into service in 1993. Nevertheless, it has suffered some unnerving incidents.

Last October a Qantas A330 was flying at 37,000ft over Western Australia when it suddenly "pitched nose-down", in the words of an official report. Henry Bishop, a passenger from Oxford, described the panic: "I feared for my life. It just fell hundreds of feet. It just fell forever and there were people flying everywhere."

One crew member and 11 passengers were seriously hurt, and more than 100 suffered minor injuries before the pilot recovered control and made an emergency landing. A recent report on the incident found that one of the plane's computers, known as Adiru (air data inertial reference unit), had "started providing erroneous data". Back-up systems are in place, but other errors occurred and the "computers subsequently commanded the pitch-down movements". Computers such as Adiru rely on data from sensors all over the aircraft. One that supplies information on airspeed is the pitot (pronounced "pee-toe"), a probe that measures the pressure of air rushing into it. If it gets blocked, it can start supplying incorrect information to the fly-by-wire system.

In 2001 an air worthiness directive for the A330, issued by the US Federal Aviation Administration, noted: "Unreliable airspeed may be caused by a radome [radar housing] destruction or obstructed pitots."

The danger was illustrated only weeks later when a different model, an Airbus A319, suddenly found its instruments giving different airspeeds as it flew into Heathrow. At 6,000ft the autopilot disengaged without warning and the captain had to take manual control.

Though suspicions fell on the Adiru, no faults were found. Instead a pitot was discovered to have blockages, causing false speed readings.

There are also problems with the probes icing up in the freezing air at high altitudes, despite a heating system supposed to prevent it. One contributor to a pilot's web-forum last week alleged: "The A330 is a beautiful aircraft but it has shown, again and again, very susceptible to probes icing."

Did a pitot ice up and confuse the fly-by-wire system? Did the computers wrongly order more, or less, thrust?

What is clear is that the autopilot of AF447 disengaged and massive system failures rapidly followed. One minute later an Acars message reported "multiple faults regarding Adiru".

Two minutes later flight control primary computer one failed, then flight control secondary computer one. Both those systems, however, have back-ups. Something far more drastic was also happening and the plane was out of control.

Four minutes after the autopilot disengaged, the cabin suddenly depressurised, perhaps with explosive force.

Although planes are designed to withstand enormous stresses, those caused by turbulence can be huge. That was demonstrated when an Airbus A300 flew into the wake of a Boeing 747 just after take-off from John F Kennedy airport in New York in November 2001. The turbulence - and the Airbus pilot's attempts to correct for it - sheared off the A300's rudder and vertical stabiliser. Without them the plane was doomed, and 265 people died.

Had AF447 suffered a structural failure? Did a window break or wing shear off? Whatever it was, the passengers must have been terrified. It was night over the Atlantic, lightning splitting the sky, the aircraft jolting in the turbulence, systems failing. Then massive decompression, cabin air gone and, outside, the temperature -30C or below. Mercifully they may not have suffered long.

As Philippe Juvin, a French doctor, explained: "If depressurisation is extremely brutal, you lose consciousness and a deep coma sets in. It would have been like falling asleep."

INITIAL reports that wreckage from the plane had been spotted floating over a wide area have proved false, with the material turning out to be detritus from ships. Last night, however, the Brazilian air force reported that it had found two male bodies from AF447.

The search will continue with a French submarine heading for the area. The aircraft's two "black box" data recorders - one for flight data; the other for cockpit voice recordings - can withstand immersion up to nearly 20,000ft and emit an audio beep for up to a month. But the chances of finding them must be slim in an area where the ocean floor is mountainous and up to 9,000ft deep.

Investigators admit that, without the black boxes, the full causes of the crash may remain elusive. But yesterday Airbus revealed that the Acars messages had pointed to 24 errors in the fly-by-wire system. It said: "There was inconsistency between the different measured airspeeds." It also emerged that Air France is now replacing pitot tubes on all its medium- and long-haul jets - which it had previously been advised to do but had failed to carry out.

That will come as little comfort to Marie-Noëlle Linguet, whose husband Pascal was on AF447. He had just posted a card to his wife. Shortly before he boarded the aircraft, he rang his wife to say he would be home before the card arrived.

When news broke that the plane was missing, Marie-Noëlle could barely comprehend it. "I didn't believe it," she said. "I keep seeing him on the plane." He never made it home. All his widow now hopes is that his last words will.