Disquieting news in the Wall Street Journal today; the paper reports today that the official inquiry into the disappearance of missing Malaysia Airlines flight 370 is riven with disagreement:
Ongoing differences of opinion between five teams of experts that include Boeing Co. and the Australian military have led to search vessels being deployed in two different priority search areas. These zones overlap in some places but in others are hundreds of miles apart, highlighting how efforts to solve one of modern aviation’s biggest mysteries remain little more than educated guesswork. Searchers may only be able to scour around 80% of the probable crash sites before government funding runs out.
For its part, the Australian Transport Safety Board (ATSB) issued a response that essentially confirmed the gist of the WSJ article:
[ATSB chief commissioner, Martin] Dolan said that earlier there had been consensus amongst the five groups, based on the data available at the time, but once the data had been refined, “the results from the methodologies did not coincide exactly. There is no disagreement, just the deliberate application of differing analysis models,” said Mr Dolan.
One would like to think that, nine months after the plane went missing, that the experts would have ironed out any loose threads in their understanding of the plane’s final trajectory. Especially given the fact that Inmarsat scientist Chris Ashton told the BBC program Horizon that the company had cracked the nut way back in March, saying: “The graphs matched, the data worked, the calculation was solved.”
But if we take a closer look at the history of the accident investigation, it’s not surprising disagreements exist. For all the confident press statements that the authorities have released, behind the scenes investigators have always struggled to make sense of the data in their possession. It’s not a matter, fundametally, of a difference in opinion between experts; it’s a matter of inconsistencies within the data sets themselves.
Allow me to explain. As we all well know by now, everything we know about the final six hours of MH370 comes from seven electronic handshakes, or “pings,” exchanged between the plane and a geostationary Inmarsat satellite. Each of these pings, in turn, provides two data points. The first, called the Burst Timing Offset, or BTO, is a measure of how far the plane was from the satellite at any given time; this data is well-understood, reliable and accurate, with an uncertainty of only about five miles. The second, called the Burst Frequency Offset, or BFO, measures the wavelength of the signals and both harder to understand and much more inaccurate, with an inherent uncertainty of hundreds of miles.
Taken alone, each of these data sets provides only a rough idea of where the plane was at each moment in time. But the hope has always been that, if combined properly, they would be indicate the plane’s trajectory and final resting place, in the same way that a line of latitude and a line of longitude can be combined to specify an exact spot on the Earth’s surface.
Unforunately, this turns out not to work in practice. No matter how much Inmarsat and the ATSB have tweaked their algorithms, they have been unable to find any routes that provide a satisfying match to both the BFO and the BTO data.
In its June report, the agency tried to make the numbers gybe by running them through three different types of analysis, none of which made sense on its own but which overlapped in a way that suggested they might somehow be right collectively. In a subsequent report released in October, the ATSB tried a new approach, this time creating one set of routes that fit the BFO data and another set of routes that make sense in terms of how a plane might actually be flown. These areas lay near each other hundreds of miles from the previous search area, but did not overlap much; the ATSB resolved the dilemma by calling them one big area and hoping to search as much of it as possible. This solution is illustrated in the picture at top; the “data error optimization” area attempts to minimize BFO (and BTO as well) while the “constrained autopilot dynamics” ignores BFO and tries to match BTO observations with aircraft performance and autopilot functioning.
Obviously, there must be some gap in our understanding of the Inmarsat data and how it relates to the aerodynamic constraints of a real aircraft. Someday we’ll figure out where we’ve gone wrong; the plane must have gone somewhere, and when we find it, the nature of the shortcomings should become clear. At the moment, my suspicion is aimed at the BFO values, because the algorithm used to explain them are so complicated, and the inherent uncertainty is so large.
We’d have to think carefully about throwing out the BFO data entirely, however, because after all, it’s the only reason we believe that MH370 wound up in the southern Indian Ocean.