We're now entering the fifth week of the search for MH370, and here at Fast Company Labs, the fourth week of developing our Monte Carlo data model in order to figure out where it went. We determined that wherever the plane ended up, it's most likely course rules out the possibility that the plane was flying by autopilot. So were we right?
- First we Narrowed Down The Location Of Malaysia Air Using "Monte Carlo" Data Models.
- Then after some feedback from Redditors, we added More About Our Methodology: Tracking MH370 With Monte Carlo Data Models.
- Finally, we improved upon the model, showing MH370 Could Not Have Flown "Accidentally" To Its Destination.
When we read about the disappearance of MH370, we thought it would be a great real-life, real-time example of how to use computer modeling to answer the question: Where is MH370 most likely to be located?
We still can’t say for sure, but we ought to be able to tell where MH370 is more or less likely to be, because we know several key data points—all of which are public information. We know:
1) Where MH370 deviated from its flight path, and we know where it was last visible to radar. And therefore...
2) We know the direction it was headed during that time period.
3) We know the approximate hourly arcs of where the plane could be, from reconstructing the Inmarsat satellite ping information. (In the spirit of our approach, we model these arcs as probabilities, not fixed locations, to account for the known error.)
In our model, we simulated the MH370's location at each hourly interval, determined from a combination of its tendency to move toward the satellite ping arcs—our known geography constraints—and not doing so in a way that would be unrealistic for a plane's flight path (for example, we ruled out the plane quickly doubling back on itself.) We assumed it was traveling at Boeing 777 cruising speed, Mach 0.83.
Here's what we predicted, and here's how close it has been to what we know so far.
- In the first version of the model, we ran thousands of simulations using different error margins, which all pointed to the same conclusion: MH370 is much more likely to have gone on the Southern track than on the Northern track. Additionally, we anticipated that the plane was off the West Coast of Australia. We did this just using the last ping arc for each update step, instead of all of them. Just after the model was finished, Inmarsat released Doppler analysis which was largely in agreement with our conclusions.
- In the second version of the model, we updated it to reflect new information since the model’s development (6 pings, instead of 5) and by using each ping at its time step. We also reduced our error margins in line with published analysis showing the error to be within 2.5% to 5%, instead of 5% to 20%. Our final MH370 locations were now even closer to Inmarsat’s analysis.
From this work came a second key conclusion: Given the possible spacings of the ping arcs, MH370 did not ever fly in a Great Circle route from its last location, to its most likely final locations off Australia. This means that the typical autopilot setting (which uses Earth's irregular sphere geometry to cover distances as efficiently as possible) could not have been used, unless waypoints were purposefully set; alternatively, the plane was flown manually.
On the day of publication, Malaysia classified the investigation as criminal, and several days later, this past Sunday, an unnamed senior government official believed that the southern route was designed to intentionally avoid Indonesian radar. That suggests foul play, presumably not at the hands of an autopilot, as we determined in our models.
More recently, there have been reports of black-box pings (separate from the satellite ping arcs) heard in the water further north than where prior search areas off Australia were, where we and Inmarsat believe the plane is most likely to be located. However, those unconfirmed black-box pings have not been heard in the several days since, and the black-box batteries will soon expire.
Next, we will further examine the range of places the plane could be using the most likely scenarios, and explain the limitations which prevent us from narrowing down the location further—stay tuned for more updates.