[Peetwo]

Captain's Log 19.10.16: MC & the MH370 'testable hypothesis' -

ATSB’s Testable Hypothesis for MH370

Posted on October 18, 2016 by Mike Chillit

The title is a bit tongue-in-cheek. ATSB does not have a testable hypothesis for its MH370 search effort. May not have known how to do it; may not have even thought about it.

A Testable Hypothesis? It’s simply a way of quantifying or measuring how we are going to measure success and failure in predicting where the plane is located. For example, a crude testable hypothesis could be “not where ATSB is looking”. So far, that is correct, but it is open ended and doesn’t help us that much. Nor will that approach help us know when to move on.

The only MH370-related Testable Hypothesis suggestion I’ve found so far is one proposed by Duncan Steel on April 6, 2014, four days after China’s Hai Xun 01 reported a large debris field at -25S, 101E near Batavia Seamount, roughly 500 km south of Zenith Plateau. Duncan did not actually come up with a testable hypothesis, unfortunately. The closest he got was a proposal to employ a surrogate: i.e., a flight simulator. That isn’t really testable at all because we have no data that would help us set up a simulator to replicate any part of MH370’s flight; but Duncan gets an A+ from me for being the only one since the plane vanished to grasp the need for a Testable Hypothesis.

The first thing we need for a Testable Hypothesis is a hypothesis with limits. ATSB did some of that when it proposed that MH370 came down at -38S, 88.5E. But it did not take the next step, which is: “How will we know it is NOT at -38S, 88.5E? Consequently, the agency has been wallowing in a quagmire it has no idea how to get out of except to say, “we came, we looked, and we didn’t find anything”. A testable hypothesis has to be able to measure success AND failure. It can’t be open ended.



So let’s turn ATSB’s “guess” into a testable hypothesis. The agency states it expects the plane to turn up on the seafloor at -38S, 88.5E. Is that realistic? Of course. When we don’t know anything at all, almost everything is realistic. But, we need to set some pass or fail parameters. Fortunately, we can use those coordinates to test the correctness of that location as the search progresses (starting back in September 2014).

There are a number of ways of doing this, but the way I prefer is to use ATSB’s flightpath assumptions. Let’s face it, if anyone makes incorrect assumptions about where the plane was at any given moment prior to the crash, all of it’s assumptions from that point forward will almost certainly be wrong. And we don’t have to test every single possibility. In fact, we can just take the last in a sequence of assumptions and test it. (Because if anything prior to the last is wrong, this one is wrong, too.)

How do we do that? One way, and the way I prefer, is to simply work back from ATSB’s terminal location expectation. To do that, we need a bit of additional information. For example, we need to know the plane was assumed to be flying at 829 kph. At that flight speed, the plane crossed each ping ring at known times: we have the “exact” time each ping was recorded.

For this simple yet robust metric, all we need is the 6th ping, which was recorded at 00:11 UTC March 8, 2014, and the 7th ping, which was recorded at 00:19 UTC. Those two values tell us how long the plane was in the air at measurable points just before it hit the water. And its speed tells us how far it flew, about 111 km.

Since ATSB has given us the coordinates for the plane’s final location (they hope), it is easy to calculate where it had to have been at 00:11 UTC. That turns out to be -36.99S, 88.5E, or about 111 km due north of ATSB’s terminal guess.

Next, the question becomes, How far might the plane have deviated from ATSB’s terminal location due to normal variation during that 111 km final leg of its journey? In order to determine that, we need to find the standard deviation for the headings ATSB believed the plane was on during its final 3 or 4 hours of flight. That metric is available and it turns out to be almost exactly 180 degrees south, with a standard deviation of 0.9798 degrees.

How does that help us? Nicely. Simply find the end location for 180 degrees, plus or minus the standard deviation. It turns out to be a true heading of anywhere between 179.02 degrees and 180.98 degrees, or about +/- 3.8 kilometers. That is also known as an error of the estimate, or a 1-Sigma event. It is not very robust yet, but we can make it so.

Now let’s make that potential error of the estimate testable with a simple normal distribution, or anything else that works in this particular instance. I tend to use a normal distribution because most people understand a little about it, and it is as important to communicate with those who are following this as it is to be reasonably accurate.

To make it testable, we simply have to decide if ATSB should search what is known as a 3-Sigma area or a 6-Sigma area. The first is a pretty small area (~11.4 km, plus or minus from ATSB’s endpoint). The second is twice its size in terms of radius. From a probability point of view, we chose the 6-Sigma metric criterion because it eliminates almost all chance we will miss the plane if it is really there.

And that’s all there is to it. So, in terms of radius around ATSB’s terminal location estimate, a 6-Sigma search requirement means it needs to thoroughly search (3.8 km * 6 =) 22.9 kilometers, a diameter of 45.74 km around -38.5S, 88.5E. It has done that in spades.



Now someone might say that is a pretty small area. True. It is small compared to what has already been searched. But this metric tells us if it isn’t within our 6-Sigma search area, it almost certainly isn’t anywhere near where we are searching because our hypothesis about its flight path is WRONG. And the way we fix our estimate is to stop searching in an empty hole long enough to take another look at our math.

So, obviously, there are some very clear advantages to having a “Testable Hypothesis” that includes an unambiguous “Exit Plan”. For example, it would have taken a month or two to determine there was nothing at all related to MH370 in the far Southern Indian Ocean if ATSB and others had simply followed Duncan Steel’s suggestion and adopted a testable hypothesis before heading off to a remote part of the planet where the plane certainly never flew.

Other pluses? Yes. The total area in that 6-Sigma search zone is only 1,647 square kilometers: 1.4% of the 120,000 square kilometers Australia has nearly completed to find nothing. Assuming proportional costs, that means instead of a $180 million tab for the search, Australian tax payers would have had to shell out just $2.5 million to get exactly the same results they now have.

MTF...P2

Ps In addition to above MC tweeted:

#MH370 Have added cost-benefit to the earlier article. Would save Australia $175 million. Any use for it? http://www.seventharc.net/2016/10/18/atsbs-testable-hypothesis-for-mh370/ …

Hmm...that would be an excellent question to ask Hoody at Estimates -