The previously discussed new information from CSIRO answers a lot of questions about debris drift in the northern area that I have had ever since I’ve developed the MH370 Location Probabilities Model. Probabilities north of Latitude 26S have always been subject to large error, due to my dilemma of simply having to guess at what drift results there might be.
Figure 3.2 of our 2nd report to ATSB showed the probability, as a function of crash site latitude (along the 7th arc, from 42°S to 26°S), and subsequent time, of the flaperon being in the vicinity of Reunion (centre panel). Informed by drift tests of a genuine 777 flaperon, and an ocean ‘reanalysis’ model fed with satellite data and carefully compared with trajectories of buoys drifting freely on the surface of the ocean, this showed that crash sites from the southern limit of the search area to about 30.5°S were all consistent with the observed arrival date of 29 July 2015. Latitudes from 30.5°S to 26°S were less consistent but not inconsistent. Figure 1 extends the latitude axis all the way to Java (8°S) and shows that in order to explain the flaperon evidence using our model, the northern limit of plausible crash sites was reached by Ocean Infinity’s search. Crash sites farther north of the searched area, to 23°S, have high probability of the flaperon reaching Reunion – but only on dates many months earlier than observed. For crash sites north of 23°S the modelled trajectories of the flaperon do not enter the vicinity of Reunion (defined fairly broadly as spanning 10° of latitude). Stepping through the simulation we see this is because the trajectories pass far north of Reunion. According to our model, the flaperon’s arrival at Reunion is not consistent with crash sites north of 25°S.
My results do not allow me to reach quite the same conclusion as Dr. Griffin for the 23S to 24S area as he is still indicating a slight potential for the flaperon to arrive roughly “on-time.” However, as previously discussed, my methodology does not fully quantify the “too early in Africa / Reunion” problem. Rather, it is somewhat agnostic regarding the arrival timing other than requiring debris arrival by the first observed date. I call these results “interim” because I now believe there are enough puzzle pieces available to modify the methodology to take into account both the “too early in Africa / Reunion” and “no confirmable debris in Australia” issues in a rational manner. By defining a “detection window” along with a reasonably accurate estimate of the contrast in detection probability between Africa, Reunion and Australia, these questions can be handled in a quantifiable manner, and the model refined to to a mathematical probability rather than hand-waving. This update will be tedious and have to wait for another time. The above results are about 80% of the answer for 20% of the effort. The remaining 80% effort will provide another 20% of the result per the Pareto Principle, so I don’t expect dramatic change, but I want to squeeze out every quantifiable ounce of signal possible out of the CSIRO drift analyses.
Griffin proceeds to describe what the results of other non-flaperon debris imply with regard to the probability the plane lies in the northern area:
Crash sites north of 33°S are only consistent with the debris evidence if one assumes that arrivals occurred before December 2015 but remain undocumented, which is possible, but increasingly unlikely for longer intervals of non-detection and non-reporting. Extending the latitude axis farther north as above, Figure 2 shows that the required non-reporting interval is greatest (more than a year) for crash sites between 23°S and 21°S. As described in our reports, Figure 2 shows the probability of being within a defined region. Another way to present the model results is to look at the statistics (Figure 3) of the dates on which modelled debris items are blown onto the coast by wind. This shows that for all potential crash latitudes south of 24°S, the median modelled beaching date lies within the range of dates on which debris was found, and the 10th percentile date was no more than 6 months prior to the earliest observed date. For latitudes south of 25°S these measures of the fit of the model to the observations gets progressively better. For latitudes north of 24°S, in contrast, the mismatch becomes much greater, with more than 50% of the modelled beaching preceding the earliest finding date.
Non-flaperon debris provides a weaker argument, in my opinion, but there is definitely a change in character for points north of 24S that indicate arrival times are simply too early. Unfortunately Dr. Griffin did not extend Figure 3 all the way to 40S, but the graph tends to argue for southern origins as most likely the correct ones.
First, the spreadsheet has been revamped. Gone is the big caveat regarding the northern area. It is no longer a huge question. Also, the MH370 residual location probability (Columns N and O) have been modified. Specifically, Column O now references the the probability that MH370 lies outside the +/- 25 nm corridor, so we can begin to see where we would have the best success if we were to extend the search width, and that is emerging as the corridor from 34S to 38S. There is now an estimated 32% chance that MH370 lies beyond 25 nm from the 7th arc. Whether it is economic for an Ocean Infinity to search there depends largely on the probability as a function of distance beyond 25 nm of 7th arc.
The remaining hotspots in the north area are now limited to 23S to 24S. As discussed previously, Dr. Griffin simply underestimates the shift in probability created by null results in the large area searched to date and mostly writes them off. The revised model, however, does indicate that 22S and points north are out of contention in agreement with Dr. Griffin and contrary to the satellite and debris based conclusions of IG members.
The Waypoint Hypothesis (40S), discounted as low probability by CSIRO, and apparently zero probability by the IG is now dominant by Bayesian math at 26% probability. Again, the discounting by both entities is believed due to the mental fallacies that result when probabilities shift dramatically due to null results in the subsurface searches of their preferred areas. The probability the Waypoint Hypothesis is correct has grown by a factor of over 4 since the underwater search began based on the current updated model. It’s hard to mentally calculate that in your head, which is part of the explanation why Bayes Theorem Finds Missing Planes (and other missing items) – you simply have to do the math. I was always aware, since 2014, that the quantifiable data ranked the Waypoint Hypothesis as somewhat remote, but my development of the Waypoint Hypothesis has always been based on what seemed to me to be very logical, but unquantifiable (in terms of location probability) human factors pointing at a particular spot, before anyone outside Inmarsat knew how to calculate BFO or the flaperon first reached Reunion. Later, I found that spot was supported independently by more unquantifiable coincidental events identified by independent researchers. Now, by virtue of underwater search elimination, even analyses of the quantifiable data is concluding 40S is the most likely location of MH370.
I am now convinced that any future search not informed by the insights that come from a Bayesian analysis of all available data is simply going to be wasting time.