Contributed by Don Thompson
We now have confirmation that Ocean Infinity’s Armada 86-05 has been mobilised with the equipment necessary to conduct a renewed search for the wreckage of MH370. Armada 86-05 is docked at a marine engineering facility in Singapore with three containerised AUV ‘garages’ fastened over the vessel’s stern with other, minor, necessary features evident.
Recently, Carl A. Allen’s Allen Exploration/AllenX team publicised that an Ocean Infinity Armada 86 vessel would join their maritime wreck survey planned under the auspices of the National Museum of the Philippine’s Maritime and Underwater Cultural Heritage unit.
This AllenX survey is to explore the San Bernardino Strait for WWII and Spanish colonial era wrecks. The Strait separates Luzon and Samar islands. Considering the allotted duration, it’s expected that the survey will be exploratory, perhaps identifying and mapping potential wreck sites for later investigation. One of the AllenX vessels is also docked in Singapore.
The Philippines archipelago is experiencing late season typhoons, while typhoon Kalmaegi has passed through, south of the strait, a second building storm in the Pacific may add delay to this project.
While speculation has been voiced in other quarters that an MH370 search mission will have ROVs and or offshore recovery cranes to hand, that is unlikely. Very unlikely. The Hugin AUVs may be tasked to revisit potential targets, acquiring higher resolution side-scan sonar imagery and, if necessary, acquire visible spectrum images using the AUVs digital camera. Planning for any physical recovery will come later, as it becomes evident what might be achieved.
The wider Armada fleet is now engaged on projects throughout north Atlantic sea areas: A78-01 in the UK North Sea; A78-02 in the Gulf area, south of Louisiana; A78-03 in the German Bight; A78-04 has returned to Vard, Norway; A78-05 & -08 with A86-01 & -03 working off Portugal on a survey for three offshore floating wind generating sites; A78-06 in the Netherlands North Sea; A78-07 in the Stabroek block, off Guyana. Only A78-07 is conducting operations with AUVs at this time: two vessels are conducting ROV operations using deep water/heavy duty ROVs; the remaining are engaged in geotechnical and geophysical operations.
Armada 86-02 is en-route to Vancouver while A86-04 and A86-06 remain at the Vard jetty in Vung Tau undergoing trials and completion. Two other ‘legacy’ construction support vessels in the wider OI group fleet are engaged on long term projects. In contrast to some other offshore operators, it’s impressive that Ocean Infinity’s vessels are now seeing such high utilisation.
Considering the AllenX project, sIO weather prospects and repositioning time, I expect the soonest Armada 86-05 might get to the sIO is likely to be late December.
Tags: AUV, debris field, MH370, Ocean Infinity, subsea
Thanks for the insightful update Don, and to Victor for hosting and maintaining his valuable forum.
Yes, it appears that Armada 8605 was designated for the search of MH370 wreckage last Summer and from October 28th when it was at Vung Tau, we have reactivated our webpage to track the ship.
Armada tracking : https://www.mh370-caption.net/index.php/armada-tracking/
Today the ship is in Singapore. Destination: Batam, Indonesia which is few nautical miles away 🙂
The ATC-Caption-2025 dataset is now published at: https://www.mh370-caption.net/index.php/malaysian-civil-atc-radar-data-atc-caption-2025/
It description notice is included in the .zip package.
We invite you also to watch the descriptive video at : https://youtu.be/-hOy5lAKK2A
@Don
When data was published on this forum in 2018 and 2020, representing a step forward, you asked us to trust, I guess to protect your source, which we did. Today, we are proceeding in exactly the same way. … Why such mistrust from the outset? Are you turning the debate into a public trial and a trial of intent before you have even reviewed the data? Is there a specific reason for this? … By publicly discussing the origin of the data, you increase the risk of reducing the protection of that origin. Your concerns seem to relate to details that we initially planned to remove from the final report; we changed our minds when we later realized that these details were already public, including on this website. We fully comply with Bellingcat’s guidelines regarding source protection and have obtained all the necessary permissions. I will therefore say no more on this matter… except it is another helping step forward.
We are pleased to make enhanced data available to the community, as we have done for Geoscience data last year, which, as sk999 demonstrated on this forum, also represented progress at tat time. The ATC-Caption-2025 data is anonymized and can be used to support new studies.
Since this forum is dedicated to discussions and technical challenges, I will respond to your previous message, as well as Tim’s, from a technical standpoint in a following post :-).
Sorry for the long reply below….
@Don Thompson @Tim
I think you will find the answers to most of your points in the report if you have time to read it (so I will refer you to the page number when appropriate 🙂
« I’m glad that … However, attribution of source appears absent. »
As you can guess, the source wants to stay anonymous. The radar data previously published on this Forum weren’t attributed either, which is understandable as it is the same. (cf my earlier post)
« As has been repeatedly stated, … that you have been accorded with permission to share, publicly, the records you have to hand. »
Yes, of course our source granted us the authorisation to publish the data subset concerning MH370 and surrounding traffic, which we just have done at https://www.mh370-caption.net/index.php/malaysian-civil-atc-radar-data-atc-caption-2025/ »
« 1/ My extraction process …were extracted, completely. »
It is not clear to me of which dataset was the source of your process (the 2018 or the 2020 radar files), but while you may have had all the data yourself, some relevant data were never published (like the Genting ADS-B data, which is more precise and complete than the KL radar data for example)
« 2/ We certainly “carefully examined”
all the extracted data from the available sources. » We have no problem with this, but in the case of the 2018 file, some “coasting” or “no detection” points were put amid the real echoes, and some BW points were missed.
« 3/ Our “decoding exploited all available bits“. Nothing synthesised or artificial. »
I basically agree on the 2020 file, but in the 2018 file it seems that some slant range modification was done in the KB data, and the points look like they come from filtered data (i.e. cartesian coordinates). I refer you to page 68 of our report and the list of detected issues with IG 2018 and page 69 for IG2020.
« 4/ My processing of the WMKB radar correctly accounted for what you term as the “mismatch between service messages and target report messages. … (within the accuracy of this remote radar head’s clock). »
If that is the case, I hope you can find the time to double check the calculated timestamps; Referring to our detailed analysis on pages 69 & 70 of the report, the main issue in the IG 2020 dataset is that timestamps seem erroneous. The most visible issue is in BW timestamps, where, for example, you find a point at 17:52:55.688, and the next one at 17:53:01.773, which is 6s later, while the antenna makes one rotation in 4s. Actually, it seems that all the BW timestamps have been calculated with half the rotation period. Please, could you confirm or infirm this hypothesis ? Thanks.
« 5/ It appears the utility of the “obscure function” remains a mystery to your team. »
This is specifically addressed page 16 in sect 2.2.3 of our report. Please, if you want to share knowledge on this particular topic, you are more than welcome 🙂
« 6/ The final WMKC mode-S EHS messages and Kuala Terengganu WAM/ADS-B messages received from 9M-MRO were definitive in demonstrating the transponder mode had been transitioned to standby. This is not breaking news, your claim that the NUCp parameter sets this conclusion in stone is risible. »
I think Tim’s comment might indicate that some people have a different view on this. You and I might already be convinced by the meaning of the last two messages presented on this forum before, but it might not be universally accepted. Let’s take for example the confusion about what the last two messages’ altitude fields contain; some people seem to think that they contain a baro altitude of 0ft, which is not the case. Considering the ICAO specifications in relation with the last two values of NUCp, any question about the possibility of a total loss of baro altitude vanishes. The NUCp (and its technical specifications) proves that the baro altitude WAS available to the transponder (see p39->42 Sect 3.3.2 of our report), and the only remaining explanation is the ALT OFF mode at that moment. This DEMONSTRATION cannot be challenged, while our initial assumption/conviction could. Hence why this might be critical to some who wish to obtain proof which did not exist before.
« 7/ Concerning weather data, the most informative environmental information to hand was the contemporary radiosonde data reports. …the ADS-B messages had been received. »
For any calculation needing the parameters for all layers (like for estimating the refraction profile and potential ducting as you mentioned), RAOBs indeed are useful. That being said, the low spatial and temporal resolution makes them less interesting (the closest measurements in time would be at 12Z, basically 5 hours before the U-turn). Wind speeds at FL250-400 are important, for example if you aim at making a precise fuel consumption calculation, because the high Mach number means that a small variation in TAS speed has a significant impact on the required thrust level. This is why it seems that mode S derived winds and temperatures, as well as good weather models dataset like ERA5 are important.
« I presume that now you understand how DCA-MY used the ASTERIX message SAC and SIC fields. Or, rather, they didn’t use these fields? »
Exactly, it seems that the ICAO attributed values were ignored. For example, official KT‘s SAC and SIC are 42 and 72, while other codes were actually used in the messages (Nothing secret as this info was already available on the internet).
From a broader perspective, after analyzing the raw data, we can confirm that the datasets previously published on this forum have some imperfections (particularly the 2018 radar data), but that they closely reflect the raw data and proved extremely useful at the time. It also appears that even the 2020 radar dataset exhibits slight time lags (less than 3 seconds), this should only impact GS error-sensitive calculations.
We have redone the study on the 4D trajectory between IGARI amd LSTRP (10Nm west of MEKAR). A revised most realistic trajectory has been reconstructed and fully simulated. It used ATC-Caption-2025 data and Geoscience data as guidance material and fully considered the aeronautical and operational contraints.
It updates our previous analysis especially at Kota Bharu and the subsequent descent.
The report on is available at:
https://www.mh370-caption.net/wp-content/uploads/MH370-KLumpur-MEKAR-realistic-trajectory-ATC-Caption-2025.pdf
The explanatory and simulation video is available at:
https://www.youtube.com/watch?v=1AcvtkuqyDU
@Jean-Luc Marchand: Others may choose to read your report, but I will not, after your and Blelly’s batshit crazy accusation that our search recommendations were based on your work. Compounding that is your complaint that I used your ship tracking data in a graphic without attribution, even though there were no restrictions placed on the data, and even after I have provided MANY data sets for public use without restriction.
My takeaway from those events is to keep my distance from you and your colleagues.
@JL-M wrote ‘you asked us to trust, I guess to protect your source, which we did. Today, we are proceeding in exactly the same way. … Why such mistrust from the outset?‘
Not only did we strive to protect our source by explicit identity, we sought to explicitly mask the characteristics of the source data and explose only the products of our analysis. A very small cadre of individuals would have had access to the data in the format you have set out. Exposing the format of the source, as you have, is an overt pointer to the source of the data.
We not only had a concern for potential repercussions against individuals but also cybersecurity risk in exposing the data in its fundamental form.
Both issues may matter less at this point-in-time than it did in 2019 but an undertaking is an undertaking & we have, in the past few days, clarified the matter with our source. I understand that the same request was made to you, you interpreted that request more loosely.
Concerning Tim’s opinions, I concur with John’s comment, ‘the way the switch operates was a possible cause of an independently observed event. Tim’s responses invariably propose the speculative occurence of a litany of alternative and/or additional unobserved events.
As for your comment concerning the leading post: a number of events indicated preparations became focussed on Armada 86 05 to undertake the search. Its arrival in Singapore with three AUVs confirmed its intended missions. A86-05 remains in the Benoi Basin, its AIS destination recorded by two independent tracking services is ‘PWBGB’, i.e. the Pilot Western Boarding Ground “B” passed on its entry into the ASPLU anchorage before reaching the dock in the Benoi Basin. Singapore port information service is not yet listing a departure.
Armada 86 05 has departed Singapore for the San Bernardino Strait (PH MTN/Matnog)
@airlandsman
It looks as if you have limited understanding of basic antenna theory. On the other hand, you read the instruction manual for the system carefully. This combination makes it difficult for me to explain why your conclusions are wrong, since it leads you to think that my explanation disagrees with the manual saying S/N variations are up to 10 dB. I do not disagree with the 10-dB variation, but my claim is that it is mostly calculable and far from random.
In addition, I disagree with your statement that S/N is not increasing. Look at the INMARSAT paper in Journal of Navigation, table 1. S/N goes up by roughly 2.5 dB. The decimal depends a bit on the analysis, but the increase is steady and significant, while basic theory clearly predicts a decrease.
There are several things with small influence on the received S/N level. The most obvious one is simple (Fraunhofer) diffraction. We agree that this one gives a decrease of 0.2 dB over the relevant flight distance from shortly before 21:00 to roughly 24:00. The next one is the relative load on the system. This is higher when the satellite is in the shadow of the earth (no power from solar panels) and during the daytime when there is more communication. Particularly in the morning when many flights login at the relevant satellite and some make initial contact with their destination airport or local office through satellite phone before direct radio contact becomes possible. During the dark period the satellite is mostly powered by battery. I doubt this type of satellite has a nuclear reactor. That is practically only used by Russian military satellites (typically using Sr90) and a few NASA satellites (e.g. to Mars and beyond) typically using Americium (I forgot the isotope number).
Typically, the battery is sufficiently over-dimensioned to ensure stable operation under normal conditions, so the variations are no larger than a few percent. That means the effect is a bit smaller than the simple diffraction effect.
The next small effect is noise. That can either come from external sources (e.g. solar eruptions or cosmic events), or from dense traffic. As far as I know there were no significant external events at the relevant time, so noise changes were predominantly related to traffic. That would have been increasing during the relevant time interval, so the impact would have had the same sign as the simple diffraction (worse S/N). My estimate is that it was comparable or a bit smaller than simple diffraction.
Finally, there is the eigen-movement of the satellite. This makes the southern route a bit better than other straight routes, but less than the 0.2 dB coming from simple diffraction and with the opposite sign.
There are also some minor effects from turbulence, wind and weather, but I choose to disregard those, since they are small and hard to calculate.
In conclusion, we can effectively ignore all small effects since the problem is dominated by the bigger effects described in the following.
There are 3 larger impact factors for the S/N ratio. The most obvious one is differences between the efficiency of the communication channels. This seems to be around plus/minus 2 dB which is relatively large compared to naïve expectations. I presume it may partly have to do with aging. For instance, it only takes a small amount of dirt on the connections to reach this level of variation. Fortunately, all relevant communication during the relevant period took place through the same channel (number 4), so we can be sure it had no impact on S/N during the relevant time interval. Outside this time interval it had significant impact in some cases.
The next important effect is Fresnel diffraction. This has a simple component due to the effective projection of the antenna area perpendicular to the direction towards the satellite. The phase optimization is unable to compensate for this effect because of curved wavefronts. It just ensures that the pointing direction is correct. However, the power efficiency of the pointing is not equally good in all directions, partly since the antenna shape is elongated along the body of the airplane to ensure smooth airflow with minimum additional fuel consumption, and partly because the antenna dimension is not infinitely large compared to the wavelength.
This means that the pointing efficiency is best along the body of the airplane and worst perpendicular to it due to the curved wavefronts included in Fresnel diffraction (in contrast to simple Fraunhofer diffraction).
That is the reason the curved paths originally suggested by Inmarsat lead to a smaller drop in the expected S/N than any straight flight (along a great circle). As the airplane approaches arc six with low speed, it gradually turns to point dramatically more perpendicular to the arc(s), thereby facilitating the phase correction (and thereby power efficiency) of the pointing (by the end almost completely along the airplane flying perpendicular to arc six). This means a smaller deviation between the theoretically expected decrease in S/N and the observed increase in S/N. I estimate the maximum deviation reduces from 5 sigma to 3.5 sigma when the last effect (see my following contribution) is also included.
However, it should not be forgotten that any normal flight to larger distance from the satellite (leveled, without height changes during the handshakes) has a dominant Fresnel term leading to falling S/N. The curved wavefronts always make the pointing less effective with distance, since the antenna size is not infinitely large compared to the wavelength. Gradual alignment of the flight direction with the body of the airplane just reduces this effect somewhat. Abrupt turns or height changes may change this conclusion for other flights, but both seem irrelevant here.
I am quite sure INMARSAT knew all that, and it was one of two reasons to insist for a long time on the unusually curved solutions. The other reason was the position of the holding pattern for the Singaporean radar plane. It was easier to bypass with curved routes.
From a purely academic point of view INMARSAT did an excellent job with their first solution. However, they forgot to think out of the box, and take things like deliberate data spoofing or corruption into account, because they thought the pilot had no detailed knowledge about their system.
@Viking
“INMARSAT did an excellent job with their first solution. However, they forgot to think out of the box, and take things like deliberate data spoofing or corruption into account, because they thought the pilot had no detailed knowledge about their system.”
Wow even I assume pilot was conceivably aware that SATCOM system logons probably generated data, and he may have wondered if SATcalls generated GPS data. But I am pretty much OK with BTO/BFO data as reported, it works for me, so I think I am OK with Inmarsat downplaying spoofing/corruption. If the data is spoofed, they did one heck of a good job providing a data set consistent with flying to the SIO. If it was corrupted, sure fooled me, as my experience it looks like solid data, and I am generally not fooled like that, because if you got a bad data, you can usually see it. In fact I probably would not have tried my hand and working a flight path solution if I thought the data was very fuzzy. That is one example where it is “hats off” to IG for helping Inmarsat refine the BTO/BFO data for the rest of us to use well-refined data. I do feel BFO may have some minor drift as expected, and unfort that drift is probably in the direction of making more westerly flight paths look better than actual up to Arc5.
@Viking
Once again, you display an incredible level of ignorance (and pure made-up nonsense) about how the Inmarsat Satcom system works, how the Ball AES antenna system works, and on and on.
Regarding your ridiculous statement: “It looks as if you have limited understanding of basic antenna theory. “…I have to laugh out loud! I was designing complex HF amateur radio antennas in high school (1960s). I founded a company in 1975 (Synergetics International) that manufactured over 6000 antennas for radiosondes, satellite ground stations (including the first commercial L band GOES DRGSs), rockets, navigation sats (2 of my 1/2 wave QFH antennas provided Transit navigation for the Double Eagle II Balloon), and more. I also founded Skylink Corporation in 1982, which was one of the first 2 companies to file with the FCC an application to build, launch and operate the first commercial mobile satellite service (MSS) in the western hemisphere. At that time, Inmarsat was 1 year old. I also founded American Mobile Satellite Service, which was a consortium of 8 companies including Skylink, which was finally licensed by the FCC to build, launch and operate the first commercial mobile satellite service in the western hemisphere. I was on the BOD at AMSC. In 1995, AMSC launched its first MSS spacecraft (using the same L band frequencies as Inmarsat). At the time, it was the most powerful (6000W solar array) commercial satellite Hughes had ever built. Later, I became VP of Engineering, and then CEO/president of Radiometrics Corporation, where I designed antennas for microwave radiometers, including L band, K band, V band and a 183 GHZ offset parabolic scanning antenna. Since retiring, I have been having fun again with amateur radio, building several HF and VHF antennas for SSB, FT-8 and WSPR.
So, I’m just wondering…what experience do you have that is relevant to this discussion?
@Jean-Luc Marchand
I am taking a look at the report and videos. Thank you for the effort. I suggest the IGARI turn could have started out as L HOLD command which is 30% bank if constant speed; presumably the pilot might have manually taken over after the start of the circle.
Regarding the proposed descent after IGARI (shown in the sim video), I am slightly more open these days to the alternate case of no depressurization at IGARI. However for the moment sticking with the case of intentional depressure, the air flow out of the outflow valves would start as “choked” (sonic) flow such that a decent, eg; to FL260, could probably be started without slowing down out-gassing too much. At some point the choked flow will revert to regular flow out, at which point ascent would be required to reduce cabin pressure to higher altitude.
Overall, if we accept the path to MEKAR+, the key question becomes fuel estimate considering the maneuvers and possible non-standard operating modes. Another question might be potential for structural or internal cabin damage.
Fantastic news! Thanks for the update.
@airlandsman, @All
You should try to do a Google search on the following:
power efficiency of phased arrays as a function of angle
If you read the AI answer it takes less than 30 seconds to realize that you are qualitatively wrong. The received power does fall as a function of distance for level flight because of the curvature of the earth. The finer details on curved wavefronts take more work.
My experience is predominantly in optics, but the fundamental equations are the same.
@Viking,
While you conduct further conversations with your chosen AI, I will ensure I have some popcorn supplies to hand.
I will suggest familiarising yourself with the arrangement of the Ball Airlink HGA conformal antenna installation on 9M-MRO.
@Viking
You should try to understand how the entire SYSTEM works (SDU, HGA, Ball Antenna, s/c antenna, s/c transponder, s/c C band backhaul link margins, GES Antenna, GES LNA, Square Peg demods, etc.). You clearly have no clue!
As previously demonstrated, the L band uplink link loss varies only 0.25 dB over the entire MH370 flight. Meanwhile, a dozen other factors (values unknown) in the total link vary more than 10 dB depending on things unrelated to location. Thus, it is impossible to infer anything about the location or direction of flight from the C/N0 values.
The maximum gain produced by the Ball antenna (like all phased arrays) is orthogonal to the planer array. Minimum gain is occurs 90 degrees from the orthogonal direction. This is the opposite of what I understand you to claim.
Regardless of the AES antenna orientation and gain, the HPA Pout is adjusted by the SDU in 2 dB increments to maintain the required EIRP = Pout + Ant Gain dBW. Thus, the C/N0 at the GES remains approximately constant (+/-2 dB) regardless of the aircraft location and orientation, assuming all the s/c and GES factors remain constant.
What we observe in the C/N0 data is 100% consistent with these facts. Nothing about the location or orientation of 9M-MRO can be inferred from the C/N0 values.
Some data on the Ball antenna is available here: https://bit.ly/4nKSEyd I encourage you to look at these files. AI is useless in this case.
BTW…Pat Westfeldt was a good friend for 40 years. He passed away recently. I had the opportunity to discuss the Ball antenna on 9M-MRO with Pat ~10 years ago. I met Pat at Ball while spending 100’s of hours working on several Ball antenna ranges in Boulder and Broomfield where I tested antennas for various missions and products, including antennas for MSS between 1973 and 2000.
@airlandsman
The short answer to your description of the system is that it is correct, but irrelevant for this particular case. The reason is that 2-dB steps are too large to affect the conclusions.
You can see why in Figure 6 of my publication. The bottom curve is a calculation of the expected S/N if MH370 flew the route calculated by Inmarsat in their publication in Journal of Navigation (close to the consensus solution). Except for the tiny correction due to the satellite Eigenmovement, it agrees with your statements. However, it also shows that no 2-dB adjustments were necessary during the relevant period.
The middle curve is my calculation of the expected S/N for the curved route initially proposed by Inmarsat. I chose the most extremely curved one (out of 3 possibilities), since that comes closest to the measured values. In that case the curved wavefronts actually dominated the result, but still the S/N change was too small to need any 2-dB corrections. If anything, there could in principle have been need for a negative correction.
The top curve is the curve originating from my solution including a temperature drop to almost exactly the outside temperature at the estimated flying height. I used average weather data for the area at the relevant time of the year for estimating the outside temperature, and assumed an open door for the thermal calculation of the temperature drop inside the cabin. The agreement is obviously good.
Some smaller details need more explanation, but I postpone that discussion to keep this contribution relatively short.
Finally, it may be relevant to discuss if the SDU could have made a 2-dB step because of some error? I find that quite unlikely, since the system would try to make no more upward steps than needed to save power and prolong the lifetime of the amplifiers. It also plays a role that pushing the system to operate at higher amplification pushes the noise-floor up. Usually power amplifiers are developed to come close to the 3-dB limit for a certain (normal working) level. Any increase in amplification and output power increases the noise. Ultimately a multistep power amplifier turns into a maser. In optics we call it a laser. Narrow-band operation helps reduce the risk, but it is still present.
A much simpler (and stronger) argument against a faulty 2-dB step is that the S/N curve is smooth with no jumps during the relevant period.
@All
It may be relevant to repeat the link to my paper:
http://arxiv.org/abs/1811.09315
My comments on Viking’s paper are here: https://bit.ly/4ozbwkV
@ALSM
I am thinking depressurization via outflow valves may not cause too much immediate cabin cooling by itself, the question becomes how long if the pressure was lowered and to what extent warm air conditioner air could be used. My understanding of the quartz crystal in the SATCOM has an oven to xx degC (50-100?). Thus the apparent depower of SATCOM until 18:25 will presumably turn off the SATCOM oven, so there will be at least that much warm up needed as we see at Arc1 BFO transient behavior. From a practical sense, BFO data quality looks good to me allowing for some minor frequency drift as expected.
@TBill
Yes, the oven is off for an hour. I would expect the cool down of the AES equipment over that 1 hr (~17:27-18:23) to be maybe 30C or so. Regardless of the cooldown amount, the OCXO oven restores the oven setpoint temperature in just a few minutes, as evident from the 12:50 and 18:25 data. It would only take a couple of extra minutes from -30C vs. +30C ambient to reach equilibrium.
@alsm, could you clarify. If the oven returns to set temp, why is there any dependency at all between ambient temp and the stabke frequency attained (per compliance docs that showed a coefficient approx 0.45Hz per degree centigrade)?
@Tbill. My understanding is that hot bleed is mixed with cold to heat tge cabin, and if pressurisation (packs) is off, then heating is off. The question then is how rapidly the cabin loses heat.
@Paul: Sorry. I don’t understand your question. Please clarify.
@Paul: I’ll take a stab at it. If you asking why there is any TC at all, the reason is that there are internal temperature gradients on the PCBs inside the oven oven chamber, and temperature gradients across the entire package surfaces inside and outside the oven. These gradients change as the inside/outside temperature changes. The result is tiny residual changes in the Xtal temperature even though the temperature at the discrete point of measurement inside OCXO is constant. IOW…the entire circuit and mounts are not thermally coupled perfectly. They are very near perfect, which is why the TC is so low in the test article. But as I noted in the Viking review, we have zero information about the exact TC in the 9M-MRO TCXO. It may have been similar to the test article, or quite different. We have no way to know. Thus, it is ill advised to infer any BFO bias change due to changes in ambient temperature.
Note also, the output frequency is primarily determined by the Xtal, but every electronic component in the circuit has some small effect on the frequency. All the components are located at different places on the PCB, not all perfectly thermally connected to the XTAL and thermistor.
Thanks Mike – you interpreted the question correctly and thanks for the response. Re your point on what inference may be drawn on possiblebias perturbation, my take would be slightly different. We have good grounds to believe the plane was depressurised at diversion and quite posdibly/probably never repressurised. So a signifucant change in ambient temp is a strong posdibility. Temperature level and rate of change a matter for conjecture. We also have ex ante reason to expect significant ambient temperature delta to impact on frequency. Magnitude is uncertain because we neither know the temp delta nor the equipment-specific temp coefficient. But it would be imprudent, IMO, to assume zero impact!
@Paul: It was definitely depressurized at IGARI. I think the evidence, scant as it is, and “pilot logic” more strongly suggest the plane was repressurized at 18:23, at the same time the L AC Bus was repowered. I think the PF restored power and pressure at that time to take a look in the cabin.
As a pilot, I have never understood why so many non-pilots think there was some big advantage to turning off the packs/air-conditioning to save fuel. It makes no sense to me. Why? It does not matter one bit if you save 5 or 10 minutes of fuel. You still end up in the middle of nowhere to be found. Of course it is possible, but there is no evidence that happened. This is especially true if you are a fan of Z going all the way to the 7th arc.
If you want to depressurise, surely outflow open AND packs off. Presume fuel not the primary consideration or motivation.
@airlandseaman,
You said: “As a pilot, I have never understood why so many non-pilots think there was some big advantage to turning off the packs/air-conditioning to save fuel. It makes no sense to me. Why? It does not matter one bit if you save 5 or 10 minutes of fuel. You still end up in the middle of nowhere to be found. Of course it is possible, but there is no evidence that happened.”
Actually, there IS evidence that the air packs were off for at least 4 ½ hours and probably longer after 19:00. That’s because, for all routes ending circa 33-36S, there is otherwise insufficient fuel to achieve MEFE circa 00:17:30.
The UGIB (2020) fuel model demonstrates this conclusion. No one has ever demonstrated a route ending in that area that allows the air packs to be on after 19:00 (and that is also consistent with all the BTOs and BFOs and MEFE).
I agree with @Paul Smithson. The primary reason for turning the air packs off (either once or twice) and opening the outflow valves was probably not to conserve fuel, although that might have been a secondary consideration if they were turned off a second time (which I doubt happened). The primary reason was to hasten the demise of the people on board.
If the UGIB descent to FL100/FL50 occurred, there was never a reason to restart the air packs circa 18:25, since cabin repressurization would happen naturally soon thereafter due to the descent to low altitude whether or not the air packs were turned on again.
If that UGIB descent did not occur, then someone must demonstrate an alternate path which saves comparable fuel as the low altitude run. Even if you fly a holding pattern, you can’t save enough fuel at cruise altitudes prior to 19:41 to match the low fuel flow of the low-altitude and slow-speed UGIB path.
In summary, to match MEFE , BTOs, and BFOs, the air packs need to be:
1. OFF at diversion,
2. ON for <1 hour between 18:25 and 19:41 (if there was a descent to FL100/FL50),
3. OFF between 18:25 and 19:41 (if there was no descent then), and
3. OFF from 19:41 to MEFE.
These fuel considerations imply the pilot could not have survived the southbound leg to Arc 7.
@Bobby: I know you think the fuel model predicts the packs had to have been turned off. But how accurate is the model? 1%, 2%?
I can believe they were turned off at IGARI, but it’s hard to believe they were not turned back on at 18:23, especially if Z remained alive until the end. No way he flew 5 hrs with a pressure mask on.
@DrB
I assume straight line holding configuration to connect an early turn (18:30) with many different 19:41 positions (through varying FLs to create path continuity) and find a high number of flight paths ending in the S32-S37 range, both for air packs on and off after 19:41. Perhaps I miss some fuel consumption as I ignored the surplus for the two turns and the descent + ascent, but that would be minor. I evaluated an equal amount (2 million) of paths with air packs on and off.
CTT, packs on: 10.1 summed probability
CTT, packs off: 12.5 summed probability
I realize, my BTO/BFO statistics may be less restrictive than yours.
@DrB
Same for LNAV:
LNAV, packs on: 8.1 summed probability
LNAV, packs off: 10.2 summed probability
Reflections on Search Assumptions and the Possibility of a Piloted End Scenario
Background: Prevailing Assumptions in Search Planning
From the very beginning, the investigative approach has been shaped by the Malaysian position, which maintains that the disappearance of the aircraft was not a result of any wrongdoing by the crew. This standpoint established an early presumption: at the time of the aircraft’s final moments, no one was in control. This belief has had a lasting impact, continuing to influence the methodologies and boundaries of the search operations conducted to date.
In particular, current search parameters remain limited to distances from the 7th arc that align with scenarios where the aircraft did not glide after engine failure. These constraints reflect the persistent influence of the original assumption regarding crew involvement and control during the aircraft’s final phase.
Conflicting Evidence from Debris and Statistical Analyses
Subsequent developments saw debris washing ashore, which spurred a variety of drift studies and discussions. While not an expert in the field, I find two conclusions most credible: first, that the crash occurred along the southern, rather than the northern, route; and second, that the most probable location lies north of the 7th arc.
However, this latter conclusion appears to be at odds with the outcomes derived from Bayesian and other statistical analyses. Specifically, the probability distribution based on satellite data does not intersect with that from drift studies. Such a discrepancy suggests that at least one of these foundational conclusions must be reconsidered or rejected, as both cannot simultaneously be true.
Exploring the Piloted End and Maximum Glide Scenario
To address this logical conflict, one potential approach is to hypothesize that the aircraft was piloted during its final moments and that a maximum range glide was performed. While it is possible to weigh the likelihood of varying glide distances, an initial focus on the maximum glide potential offers a starting point for further exploration.
Implications for Future Discussion
Considering the possibility of a piloted end and associated glide raises important questions. This line of reasoning could broaden the scope of analysis and discussion, potentially influencing future search efforts and interpretations of existing data.
@ALSM
I agree with you, highly likely cabin depressure at IGARI, but how can you say definite? Admittedly it seems like that was implied from the start with the early news reports saying FL400. Qualitatively there is not much new, except debris, since Aug_2014 when Ewan Wilson/Geoff Taylor pretty much had it all down on paper in the first book Goodnight Malaysian 370. Quantitatively we have come a long way on data over the years.
@Paul Smithson. A possibile reason for leaving packs on would be to delay alerting passengers, crew and co-pilot, presumably in the cabin, to depressurisation, turning off the packs raising that risk.
Even so if the aircraft did indeed simulate an emergency descent on completing the turn back after passing Igarii, as concluded by Jean-Luc Blanchard and Patrick Blelly in their 4th Nov flight assessment, then that would not arise.
@Sid Bennett
Re “From the very beginning, the investigative approach has been shaped by the Malaysian position, which maintains that the disappearance of the aircraft was not a result of any wrongdoing by the crew.”
Sid, if you are rolling ICAO Annex 12 accountabilities for search and the Annex 13 accountabilities for investigation together under the concept of “the investigative approach”, then you are manifestly and demonstrably wrong in positing that thinking was shaped by the “Malaysian position” that a flight crew member was not implacated in the disappearance.
Let’s start with what the then “Malaysian position” actually was. If the then Australian Prime Minister, Tony Abbott, is to be taken at his word, the then “Malaysian position”, as it was conveyed to him from “the highest levels” of the Malaysian government, was that the disappearance “… was almost certainly murder suicide by the pilot, mass murder suicide by the pilot.”
Having spoken candidly with some of the ATSB personnel responsible for formulating the underwater search areas, I can assure you that they proceeded at that time on the same understanding as that conveyed to Prime Minister Abbott.
We all need to remember one of the key constraints with regards to conducting a deep-ocean seabed search that applied at the time, viz the technology. Back in late 2014 – early 2015, conducting a deep-ocean seabed search meant using a tow-fish that was suitably instrumented with sonar. And that meant a preliminary bathymetry survey had to be conducted first. The conduct of any search would necessarily be a) slow and b) expensive on a per square kilometre searched basis.
For fear of stating what should be obvious, there were not unlimited resources available for an underwater search. Accordingly, the search area needed to be constrained.
The logic underlying the determination to prioritise search length along the 7th arc over search width from the arc is well documented in the ATSB’s MH370 – Definition of Underwater Search Areas dated 26 June 2014 (updated 18 August 2014). An extended, and therefore piloted, glide after fuel exhaustion was never rejected outright, rather it was deprioritised for a number of reasons, one of which was to maximise the length along the arc that could be covered.
The possibility of an extended, piloted glide subsequent to fuel exhaustion was most assuredly not deprioritised as a sop to Malaysian sensibilities, and I know that the personnel involved in that decision would be rightly offended by the suggestion.
@Mick Gilbert
My views on the subject are based primarily on the impressions that I recall of Duncan’s blog and early discussions of the IG. It seemed to be impolitic to suggest that the pilot did it. I recall it being called a “zombie flight”. I believe that the initial studies were done with statistical techniques and that it was only much later that it was accepted that a good portion of the early portion of the flight was by waypoints.
Surely you would agree that the early working assumption was that there was no human control during the later portions of the flight (except for the hijacking to Kazakstan).
Recently the possibility that the later portions of the flight were CTT became acceptable. Yet how could this have happened except by a person setting a switch on the control panel?
Rejecting the glide on the basis that it would lead to a impossibly large search area is indefensible. Whether there would be the resources to search such an area is a separate matter.
It still comes down to the situation that, as defined, the satellite data and the drift data do not overlap.
@Sid Bennett
Re: “Surely you would agree that the early working assumption was that there was no human control during the later portions of the flight …”
There was most assuredly a working assumption of no pilot inputs at the end of flight. Surely you are not conflating “no human control during the later portions of the flight” with no “wrongdoing by the crew”?
Re: “Rejecting the glide on the basis that it would lead to a impossibly large search area is indefensible.”
“Indefensible”? That’s debatable but in any event moot, as a piloted glide was never rejected, rather it was deprioritised in favour of extending the length along the arc that could be searched. As illustrated at Figure 7: Search area width priorities in MH370 – Definition of Underwater Search Areas dated 3 December 2015 (updated 10 December 2015), the search width required to address what was deemed the “realistic distance for an unpowered glide” was accorded Priority 3.
Frankly, what purpose does this decade-on quarter-backing of the original decisio serve? The decision to prioritise the search effort to the inner zones in order to maximise the extent of the 7th arc that could be searched was the most logical approach to the problem at the time. It is a basic tenet of search operations to start with the Last Known Position and work outwards.
@Mick Gilbert
It still comes down to the situation that, as defined, the satellite data and the drift data do not overlap.
@Sid: You say “…satellite data and the drift data do not overlap.” What is your understanding of the two arc segments associated with each?
@Sid Bennett,
You said: “It still comes down to the situation that, as defined, the satellite data and the drift data do not overlap.”
That is incorrect.
Figure 12.1-2 in Ulich and Iannello (2023) demonstrates that the MH370 debris drift probability extends from 26S to 36S.
Figure 14.2-2 demonstrates that the route probability (based only on BTOs and BFOs) extends from 29S to 40S.
Thus, the drift and route probabilities overlap from 29S to 36S.
Figure 14.2-2 demonstrates that the product of the route and fuel probabilities extends from 29S to 37S.
Taking into account the drift, route, and fuel probabilities, the overlap range is from 29S to 36S
@airlandseaman,
You said: “I know you think the fuel model predicts the packs had to have been turned off. But how accurate is the model? 1%, 2%? I can believe they were turned off at IGARI, but it’s hard to believe they were not turned back on at 18:23, especially if Z remained alive until the end. No way he flew 5 hrs with a pressure mask on.”
I agree completely on your last point. No one could have survived and glided the aircraft at MEFE if the cabin were depressurized after 19:41.
On your first point, the UGIB (2020) fuel model predicts with a high (but not 100%) probability that the air packs were off for more than 5 hours.
Here is a summary of those fuel modeling results:
1. The 1-sigma uncertainty in the fuel flow model predictions is estimated to be about 400 kg, or 0.9 % of the total fuel consumed after the last fuel report at 17:06:43, which was 43,800 kg.
2. The average 2-engine fuel flow nearing MEFE is about 92 kg/minute.
3. The 1-sigma uncertainty in the predicted MEFE is about 4 minutes and 15 seconds.
4. All nine MH370 flight scenarios evaluated in UGIB (2020) have a fuel shortfall, ranging from 1,221 kg to 157 kg at 19:41.
5. Having the air packs OFF from 19:41 to MEFE saves about 480 kg of fuel, which is somewhat larger than the fuel prediction error.
6. Scenario 9 incorporates 5.6 hours of air packs OFF time and no descent. It has the smallest fuel shortfall of 157 kg at 19:41, which is within the prediction error of 400 kg.
7. Scenario 8 is a “descent” route. It also has a total of 5.6 hours of air packs OFF time, plus the right engine is shut down for about an hour. Case 8 has a fuel shortfall of 257 kg at 19:41 and a predicted MEFE within the expected prediction error. Therefore, the 1-hour INOP period saves about as much fuel as the net fuel cost for the descent to FL10 and climb to FL390.
The fuel model predictions indicate it is much more likely than not that the air packs were OFF after 19:41, but this is not 100% certain.
@DrB Thanks for that summary. I appreciate the effort you put in to perfecting that fuel model, and all the other UGIB work. FWIIW…My thinking on the ending has evolved. In the beginning I was more confident that the descent was uncontrolled. But the more I poured over all the Boeing simulations and the ones Paul and I did Nov 2, 2014, the more I questioned that assumption. Based on the BFO data and missing 00:21 IFES logon, I don’t think there is much chance of any glide, but a deliberate high-speed dive is consistent with the unusually short time after MEFE (~3 min) and the shreaded debris.
I meant to add…none of the Sims ended in less than 5 minutes. They end in a very high speed descent, but they all ta.ke a few minutes to wind up
@airlandseaman: In the simulation that you and Paul did to look at both engines flaming out at the same time, there is a re-start of the left engine that starts producing significant thrust about a minute after fuel exhaustion. The asymmetric thrust from the re-start causes the bank to increase to around 50 degrees and the descent rate exceeds 10,000 fpm at which time there is a “violent” correction with left yoke that leveled the wings, resulting in a phugoid oscillation until the aircraft crashed around 13 minutes after fuel exhaustion. I believe the correction is an artefact of the simulation. If the simulation did accurately model the re-start, that might have caused the high downward acceleration suggested by the final BFO values and resulted in a crash relatively close to the 7th arc.
Sid Bennett said: Rejecting the glide on the basis that it would lead to a impossibly large search area is indefensible.
Given the imprecision in localizing where MH370 crossed the 7th arc, given the final BFO values suggesting a steep descent, and given the missing IFE log-in, I believe the decision to limit the extent of the search away from the 7th arc and to extend it along the 7th arc is very defensible.
Of course, today we are in a different position, having searched extensively near the 7th arc without success. I may be in the minority to say the choice of where to search next is not very obvious to me.
@Sid
My feeling would be that the Razak admin was not so averse to finding the aircraft, for example we got the leak of (incomplete) sim data which gave some SIO coordinates, also per former PM Tony Abbott we had that Australia was told apparent deliberate pilot in the early days (which most of us did not hear until the 2019 disclosure). Many did not want to believe that, or perhaps did not want to the public to hear that unless proven fact. Thus we were stuck with general reluctance/defensiveness to assume the the worst case scenario, whereas that is probably what we observed.
@Victor: Re: the “…both engines flaming out at the same time…” simulation, 2 things can be true at the same time here. The simulator trainer stated at the time that that reaction was caused by a simulator glitch…some parameter “out of bounds”. But he also said that restarts can and do happen, so the fact is, a brief restart could explain an upset and a more rapid descent, like we see at 00:19:27. Or, it was a deliberate piloted maneuver. But either way, the 00:19 BFO data, shredded debris, and missing IFE logon are hard to explain without accepting the fact that the descent was quick and close to the 7th arc.
@airlandseaman: I agree. There was likely a simulator glitch causing the wing leveling, but the engine re-start and the subsequent steep, banking descent was what we would expect to see during a re-start, so there is no reason not to accept the realism of the simulation to this point. If the re-start actually occurred, no pilot inputs would be required for a high downward acceleration with a crash close to the 7th arc. The simulation would have shown this had the glitch not occurred.
On the other hand, the failed subsea searches to date make a crash close to the 7th arc less likely. That doesn’t mean it didn’t occur, as we can’t be certain that the search didn’t miss the debris field. That’s why I say it is not obvious to me where to search next, especially since the search team has high confidence that the debris field was not missed.
@Sid Bennett, @TBill: As @Mick Gilbert explained, the search area was not defined by the official investigation team with the assumption that the diversion was not intentional. In fact, privately, the team believed that the pilot deliberately diverted the plane to the SIO. The search area was defined as most likely using the data at hand and within the limitations of the resources available. What Malaysian and Australian officials were saying publicly was almost irrelevant to the search area.
@All
I am sorry that my broad-brush comments on the drift studies were a bit intemperate. I did re-browse the cited paper and accept that the study appears to be thorough and pertinent. My recollections were based on some of the early studies. Thanks for setting me straight.
@TBill
That the Malaysian position on pilot responsibility was publicly known until 2019 (and indirectly) is consistent with my recollections. I believe that the misleading silence by the officials led to a constrained viewpoint as to the impact point and unnecessarily influenced independent studies.
@airsealandman
To my point above. The public discussion led to a substantial uncertainty as to when the human control of the flight ended. If we had been certain that there was a pilot in control when the fuel ran out, would you expect this experienced person to have just let the plane go out of control?
Sid: No. If there was a live PF at the 7th arc, the data and facts point to a deliberate very high speed descent.
@Victor @all
I might be at the height of my Dunning-Kruger curve and also boring, but the best next place (places) to search would be everything that was missed and left off along the 7th arc, including data gaps and everything else within the area recommended by ATSB.
Victor, it is true that the search team was confident about their search results. In Peter Foley’s words, ATSB’s sonar data was checked four times by multiple entities. I am sure those areas can be dismissed. However, that applies only where data was available, and that does not include ATSB data gaps and certainly not the far southern part of the search area. Otherwise ATSB would not highlight them in their Final Report.
One person or another can have a preference for a certain area, however, only the systematic approach is the right one. Let’s search everything that is larger than 200×200m in size ±25nm from the 7th arc all the way to 40°S. Even if the plane is not there, at least we would know with high confidence that it was a controlled descent and would provide another starting point for future analysis.
@airlandseaman
I beg to disagree. If there had been a “deliberate” high-speed descent it would still could be in a controlled manner and there is plenty of evidence from other incidents that such an initial high-speed descent could recovered into a long-range glide.
@Sid Bennett said: No doubt any pilot can recover from a steep descent with a modest loss of altitude, as it’s part of training. But in this case, the BFO indicates a large downward acceleration of around 0.7g, which in controlled flight indicates an aggressive nose-down yoke input. Why would a pilot choose the glide-dive-glide sequence you believe occurred? It would seem that a pilot in control would either glide or dive.
@Sid: You misunderstand my comment. I agree with you. It could have been a deliberate (controlled) high-speed descent. That was my point all along.
Marijan: You are proposing a logical progression that makes sense. However, I don’t think there will be much support for this because the search team believes the areas missed are either too small to hide the debris field or too improbable to revisit.
We also don’t really know what to expect for a debris field. Some of the small man-made objects were never investigated, but could have been large parts of MH370 that were not part of an identifiable debris field.
Andy Sherrill provided this comment two years ago:
https://mh370.radiantphysics.com/2023/11/02/man-made-objects-detected-on-seabed-are-possibly-from-mh370/
“Typically, if there were small isolated objects that appeared to be man-made and marked as a target, but nothing else was of interest within several kilometers then we did not investigate further.
We certainly took into account if the debris field did not look like AF447 or any others, however there still needed to be enough debris to be at least a fair amount of the aircraft to warrant further investigation.
Sure a small part of the plane could have drifted and sunk, but we were looking for the main field. A decision was made to focus on finding the main field of debris, not just one small piece – and likely all of those “potentially man made” contacts are from passing vessels given there was no associated debris within several kms.
Having said that, there is always a chance it [a tagged contact] could be from MH370, but based on our assessment the time it took to investigate each of these small contacts was not worth taking vs searching new areas.”
@airlandseaman
Thanks for the clarification….
I’ll add that we have evidence suggesting but not proving that there were no pilot inputs after 19:41 (statistical consistency with automated flight) and a steep, accelerating dive at the 7th arc (last two BFOs, no IFE log-in). However, if there were pilot inputs before and after fuel exhaustion, it becomes very difficult to define a high probability search area based on the available data without introducing biases (i.e., guesses) about what a pilot may or may not do, especially since in the pilot diversion scenario, the pilot was not behaving like a typical pilot. That’s fundamentally my frustration. The areas that were most likely to contain the debris field were already searched.
@All. I think Victor post above “…it is not obvious to me where to search next” voices the feeling of many.
However, the glaringly obvious response is that there is a viable path solution ending in a section of 7th arc that has NOT yet been searched. I have already demonstrated that the area just south of 39.0 is actually the PEAK probability terminus by BTO path-fitting; that this predicts a waypoint-route MEKAR-SANOB-IGEBO-[RUNUT]-4085S; that this route matches the first available left-turn along an airway at the NW extremity of Sumatra; that the route is feasible with the fuel on board at M0.84, FL360.
Perhaps folks are unpersuaded by the last bit. so allow me, please, to appeal to your scientific curiosity and willingness to examine the evidence in an open-minded way.
Like Bobby and others, I have taken as gospel the fuel on board at 17:07 ACARS. I accept the Boeing fuel flow tables, the 3% per 10C TAT penalty and the 1.5%PDA and the plane’s zero fuel weight. So no dispute about fundamental fuel modelling assumptions.
Where I do differ is in pointing out the following:
1. There are good grounds to believe that we have significantly underestimated the fuel cost of bleed air. UGIB estimates this as 1.56% at FL390, combining packs-off with reduced drag from ram-air door. This produces a saving (in UGIB case 9) of roughly 700kg if the packs were off for nearly 6 hours.
There are multiple authoritative sources online that indicate the fuel cost of bleed air is more like 5%. First example comes from Lehle W, 2006 and refers to an Airbus 330 with Trent 772 engines, FL390. The author worked for the company that designed and supplied the Airbus environmental control systems.https://tinyurl.com/Lehle-2006.
Martinez (Madrid, thermodynamics professor) and Faleiro et al (2004) also provide 5% as a ballpark number. Bender (2018) goes for “3%-5%. Sinnet M (2007) suggests NET fuel savings of 1-2% by replacing bleed air with on-board power generation in the case of “more electric aircraft” and the B787 in particular.
That brings us to the second body of literature – which has quantified the energy requirements of substituting parasitic bleed air by compressed air produced by generated electrical power. The estimates of the power requirement are clustered pretty close together, at ~1.3kVA per passenger (range ~1.0-1.5)(Voth et al, 2024) and this is driven by the FAR 25.831 standards on air-per-passenger. Now, scale this to the maximum passenger configuration of the aircraft type and convert to a fuel equivalent. The arithmetic is as follows:
777 200ER design person-load approx 360 pax + 15 crew. Say 375. Power requirement per head = 1.2kVA; apparent power factor 0.8. Thermal efficiency of power generation 30%. Specific energy density of jet fuel 43MJ/kg. Over 6.9 hours (from 17:22 to the end). Produces a fuel equivalent of 1075kg – or 2.6% of fuel on board. Now, based on various “more electric aircraft” sources, we know that the actual bleed air fuel cost is about a percentage point higher. That takes us to 3.6% (of fuel remaining at 17:22 which I estimate at 42,079kg), or 1,535kgs. If it were 5% (per Lehle, Martinez etc) the figure would be 2,100kg.
Long story short: it is entirely credible – and supportable by published literature – that the fuel cost of bleed air is somewhere in the vicinity of 3-5%, producing fuel savings of 1.3 – 2.1 metric tonnes.
2. ADDITIONALLY, my previous post proposed the following further savings:
a) There is an estimated 215kg (range estimate 160-270kg) of fuel trapped in pumps, manifolds, lines and engines. This is not measured/included in the FQIS indicated fuel quantity but it is by definition accessible/usable by the engines. Source: ChatGPT, assembled from various sources, not Boeing specific.
b) Reduced electrical load on the IDG due to lights off and non-operational galleys reduces power offtake from the engines. Using the same “energy-fuel equivalence” methods as before, I have estimated that reduction in IDG load of 100-150kVA to produce savings of 240kg-350kg (midpoint 295kg).
That’s another 505kg, that I consider “highly likely” to be available.
Then we have some “also ran” contributions such as the “error budget” of ~400kg; a longer period under 1 engine (saves 30kg per additional minute, up to 8 mins more than assumed in UGIB – so 240kg equivalent); and maybe a little fuel below/beyond FQIS calbiration point zero (at the most unfavourable aircraft attitude for fuel pickup). These “also ran” possibilities represent a total fuel quantity of ~700-800kg. However, I haven’t even counted them/relied on them for the route simulation.
At FL360, M0.840 throughout I estimated a fuel deficit of 1852kg (roughly 19 minutes) before taking any “credit” for savings. Against this, using the MIDPOINT rather than upper-end estimates, we have:
1. Bleed air/packs-off from 17:22 saves 1683kg (using 4% as the midpoint of the 3%-5% range)
2. Entrapped fuel gives you 215kg
3. Lower IDG power offtake gives you 295kg
TOTAL = 2193kg.
Without relying on error budget or other less-likely scenarios, we already easily cover the deficit, with 341kgs to spare (2193 savings minus 1852 deficit).
Ergo – there is demonstrably enough fuel to do it.
Comments, please.
@Paul Smithson
Best post in months! I agree 99.99 percent.
Final waypoint was (41.896667°S 85.00000°E)
Final BFO’s were ‘deliberate rapid descent through a cloud gap’ to get visual with ‘sea state’ for the end game.
@Paul Smithson
PS:
There was a second very logical reason for the final BFO’s.
When left engine failed, there is no more air for pressurisation.
Cabin altitude would begin to rise (depending on cabin leak rate).
Zahari did not want to have to go ‘mask on’ again at this critical time.
@Ventus45
I have proposed a final waypoint of 40S 85E based on the fact that it is “easy” to manually create. The BTO errors for a flight at exactly M0.84, FL360 – and no messing around with invented turns or jiggles – from IGEBO towards 4085S are: -8, -29, 33, -36, -5, 19
4085S happens to lie within 0.5 track angle of IGEBO-RUNUT and is also in the vicinity of the old waypoints that used to exist before those SIO oceanic routes went flexi.
But it doesn’t really matter if that final WPT assumption is precisely correct or not because BTO fit dictates that you arrive at the same terminus anyway. Peak probability path termini are: 39.58 (from RUNUT), 39.60 (from IGEBO), 39.60 (towards 4085S) and the tolerance range is less than +/-0.2 degrees of latitude to maintain BTO fit within 2 sigma.
Since you have a similar magnitude of uncertainty associated with the final descent (ATSB suggest 15NM radius) it doesn’t make any practical difference. You would still need to search pretty much from 39.5 to 40.0 along the arc. In practice, you could prioritise an area that is roughly a quarter of the size: 39.5 to 39.8 with Arc 7 defined at (say) 10,000ft and a margin of 10NM either side.
You will note that the scenario is not reliant on active pilot. In fact active pilot is ruled out by reliance on packs being off throughout. So the unpiloted descent after final fuel exhaustion (per ATSB end of flight scenario) is part and parcel of the route hypothesis I have proposed.
For anyone wanting to look at the proposed route in Google Earth, the .kmz of the model path in 5s steps is shared here: https://tinyurl.com/yn7rak9j. All positions are at height 36000ft.
@Paul: How do you reconcile a POI that far south with all the drift data and modeling? All the best models point to a segment of the 7th arc further north…S30 to S35. None of the models to my understanding support a POI ITVO S40.
@airlandseaman the elephant in the room is that even after more than a decade MH370 debris has yet to appear in sufficient quantity anywhere to substantiate the verisimilitude of drift models. I think from the onset it was accepted that debris drift endpoints fed back Into a drift model with speculative wind, swell and wave action inputs would point to a very expansive zone of possible decampment. I think drift analysis has had more value in ruling out some northern EOFs rather than value in honing in on an actual EOF.
@Paul Smithson: Perhaps you can get Boeing to weigh in on fuel savings from no extraction of bleed air.
@Victor
I agree with everything what was said about the areas where reliable sonar data exists. However, I don’t agree that all data gaps are too small to contain debris.
Here is one example from the “MH370 Data Review – Final Report”: https://www.atsb.gov.au/sites/default/files/2024-02/mh370-data-review-2022-final-report-v2.pdf
Figure 3 on page 9 shows the map of data holidays (both data gaps and lower probability detection areas – LPD-s). Since the map contains the distance scale, it is possible to roughly estimate the size of areas. For example, the “shadow zone,” which crosses the 25nm circle, is roughly 2.8km x 1.4km in size. For that particular area there is no sonar data, and the debris field could well fit into it. There are a couple more of similar data holidays at that are at least 1km wide just next to 25nm circle. This is just one example from the searched area close to 33.2°S and for sure there are more.
I believe that with some help from the ATSB the current Ocean Infinity search team can relatively easily filter out those zones of interest (from previous search maps) that are large enough to contain debris and their total area should not be very large and manageable for the equipment OI has.
@Paul Smithson
Paul, if the plane is not in one of those data gaps, I am with you. We should stick to the 7th arc and not so far from it since all data point out in that direction
@All
I feel it is pretty darn obvious active pilot was alive and flew away from Arc7, with SATCOM off. The descent at Arc7 is almost obvious active pilot, as pointed out here by @Gysbreght years ago. I believe pilot was probably trying to hide his track, flying descending to the cloud layer below in the early morning light. The BFO/BTO data tell us it was effectively “straight” til Arc5, with descent slow down next and curve Southeast. If we realize sim data is 180S CMH not NZPG LNAV, then the pilot probably adopted sim path at Arc5 and headed in the general direction of Broken Ridge.
Blaine/Chari actual debris finds are based on drift calcs from a 32.5s/96.5e hypothetical end point, with Chari originally saying it could be +-70nm from there horizontally (East/West). A few years back I contacted Blaine/Chari to ask if their drift model could be greater than +-70nm offset from Arc7, and I was told yes.
We are very lucky that the SATCOM reboot at Arc7 actually gave us Arc7, and tells us pilot was active, so the distance from Arc7 may be a little less than it could have been if Arc6 had been the last Arc. We are lucky to have the sim data, lucky to have debris work by Blaine/Chari. But we are in disdain and denial of all of that seemingly obvious apparent evidence.
@Victor – I have p.m’d you regarding the possibility of going to Boeing with this question. I have exhausted all avenues of published/public domain data as well as the subject-matter experts with whom I have been able to establish comms. I hope that there are other colleagues here who can help with directing specific questions to Boeing or other subject-matter experts.
But even if that is not the case, it seems to me that the evidence I have laid out above makes a prima facie case for the fuel penalty of bleed air (at cruise) being somewhere in the region 3-5% of total fuel burn.
@All,
All this talk here of a controlled path is not based on any proper evidence. The facts that we have show this could be a hypoxic ghost flight all the way back to the IGARI turn. The last evidence of any pilot input was the IGARI turn, after that it just became a meandering, autopilot FAILED ghost flight. Just like all the other accidental hypoxic accidents. Only this time the autopilot had failed too.
A random flight path can match the primary radar, and arc data. Then an uncontrolled final descent fits the debris damage caused by a final rapid out of control descent.
It should be near the 7th arc, unless the arc calculations have some sort of unknown inherent error
The aircraft must be close to the 7th arc
@TBill: You conflate your opinions with facts. Here’s a short list:
>I feel it is pretty darn obvious active pilot was alive and flew away from Arc7
>The descent at Arc7 is almost obvious active pilot
>The BFO/BTO data tell us it was effectively “straight” til Arc5, with descent slow down next and curve Southeast.
>We are very lucky that the SATCOM reboot at Arc7 actually gave us Arc7, and tells us pilot was active
>But we are in disdain and denial of all of that seemingly obvious apparent evidence.
The last in the list is particularly irksome because you ascribe disdain and denial if we don’t agree with what you consider to be obvious. This is classic MDS behavior that I’ve observed among many people with very different opinions.
As for Chari Pattiaratchi’s drift predictions, we’re all still waiting for how he predicts a POI so much further north than the CSIRO model. A very basic question never answered is what Chari used for windage, which has a big effect on where along the 7th arc the POI is predicted. The fact that Blaine Gibson found debris where Chari predicted is meaningless, as just about any model would predict landings in East Africa, including Reunion Island and Madagascar. Since you have discussions with Chari, why don’t you ask him. In an ideal world, Chari and David Griffin would work together to compare and contrast their findings. David has been extremely open in sharing all details and results of his model.
@Tim said: Just like all the other accidental hypoxic accidents. Only this time the autopilot had failed too.
I previously asked you whether in the history of aviation we’ve ever seen an aircraft with no autopilot, no augmented flight control, and no pilot inputs fly for many hours before crashing. You never responded.
@all
I know that many here will not take too kindly to the following, but here goes.
It’s clear (to me) that Zahari’s last flight was both deliberate and meticulously planned, which means two things. He had an objective (he definitely had an ‘end point’) and he had a methodology to achieve that (he definitely had ‘a route’). We have to accept those facts as baseline, and get into his ‘mission planning head’, or we are on ‘a perpetual hiding to nothing’, because eleven plus years has shown that 9M-MRO can’t be found by mathematics alone.
But therein lies the fundamental problem.
No one wants to publicly admit the blindingly obvious, not the politicians, not the official investigators, not the mathematicians etc. If we are ever to find 9M-MRO, we have to ‘get over the denial hurdle’ and work out what ‘Zahari’s end point’ most probably was, by looking at both his ‘motivation(s)’, and his ‘objective(s)’. In other words, it is first necessary to ‘profile’ Zahari, and that is the poison pill that none of the above are willing to swallow. Even Tony Abbott remained ‘silent’ for years (way too many in my view). Anyway, TBill can give you ‘chapter and verse’ on “denial”.
Moreover, ultimately, it is irrelevant whether or not Zahari was alive much past about 19:30utc, because by that time, he was way outside both all radar coverage and the maximum interception range of any TNI-AU fighters (even if there had been an AWACS or a Naval Warship with an ASR in the area to vector them – which there wasn’t).
Those not in denial, have proposed many ‘final waypoint theories’ over the years (the first prominent one being GlobusMax) but all have been dismissed, because of denial (as above).
So let’s swallow the poison pill and look at Zahari’s LONG history in MAS.
Back in the day, MAS used to fly WMKK-FAOR-WMKK and WMKK-YPPH-WMKK (still does YPPH). It is highly probable that Zahari had flown both routes, possibly regularly. If a MAS aircraft had ever gone AOG at either FAOR or YPPH, it is possible that another MAS aircraft may have needed to be repositioned (by non-revenue ferry flight) across the SIO (either FAOR–>YPPH or YPPH–>FAOR). MAS Operations must have had ‘contingency plans’ for such a possibility, and they would have practiced them in the sims (on occasions), and since Zahari spent a lot of time training people in the sims, he would have been very familiar with the details. As Paul Smithson noted (a few posts above) prior to the introduction of FANS, there were published great circle air routes across the SIO, which had mid ocean waypoints spaced 15 degrees of longitude apart. Both Paul and I have a list of them, indeed, I posted them on this forum a long time ago.
Zahari certainly knew that the route between FAOR and YPPH is virtually open ocean all the way (except as below), so much to consider in training for PNR’s and ETP’s (‘points of no return’ and ‘equal time points) etc. As Nick Huslan (Datuk Nik Ahmad Huzlan Nik Hussain, a retired chief pilot for Malaysia Airlines) so pointedly stated (with emphasis) in an Australian ABC 4 Corners interview: “Zahari made DETAILED notes”. If such an event had ever actually occurred in reality, and if Zahari had been ‘on hand’ at the time, he would undoubtedly have ‘jumped at the chance’ to do it, so he may just have flown such a ferry flight. Even if he had not ever done so (highly likely) he was certainly ‘mentally very well prepared for a one-way trip to the SIO’.
When planning his final flight, Zahari knew that the most southern waypoint between FAOR-YPPH is RERAB (43.000333°S 75.000000°E). He knew that he would not have enough fuel to reach it, but even if he had enough fuel, he would not select RERAB, because it would have taken 9M-MRO way too close to Amsterdam Island (only about 50Nm abeam) and Isle Saint Paul (less than 30Nm abeam) for comfort. Moreover, there is a French civilian scientific base on Amsterdam Island, but significantly, all ‘base technical services’ are provided by the French Navy, which, among other functions, acts as a comms and control base for patrol craft with ELINT gear that the French Navy regularly sends into the area protecting their EEZ and keeping illegal fishing boats at bay. On the other hand, tracking towards IGPOL (41.896667°S 85.000000°E) totally avoids that risk set, and it may be just reachable (depending on fuel), and most importantly, it is a far more ‘personally satisfying’ destination to him (from his ‘political’ point of view) even if 9M-MRO did not actually ‘reach it’ on the day (think about it).
@Mike. Thanks for your question about the drift model compatibility of a terminus as far south as 39.6S. Although the “common consensus” points to areas further north, I think it would be unwise to rule out a southern terminus based on drift. There are basically two kinds of reasons why. One is short and slightly “dismissive”. The second is longer and more “rebuttal”. Take the short one first.
TL/DR – drift modelling is a highly-uncertain science that provides “fuzzy” answers at best. It should be applied to MH370 terminus discrimination with extreme caution and full cognizance of baked-in model assumptions.
Anyone “skilled in the art” will recognise that behind any drift model there is an ocean surface current model, a meteo model for wind speed and direction, a coefficient for wind forcing that may or may not separate out stokes drift from from “leeway”, and a “dispersal” factor. If you take the underlying models and parameters as gospel, you will – of course – end up with a deterministic result.
It is quite another thing to believe that this actually reflects reality of the integrated current+wave+wind vector that underpin the unique drift journey of an specific bit of debris. The reality is that apparent “determinism” is an illusion and ought to come with health warnings about the inherent uncertainty applicable to parameters, underlying models and random effects. That is before we even get into debates about the likelihood of any particular stretch of coastline being “sticky”, or the likelihood that beached debris stays that way, or is witnessed, recognised and reported by a human being.
The reason that drift modelling was invoked at all was not because anybody believed it would be an accurate predictor of crash site – only that it MIGHT be helpful at the margin to narrow things down. As it turns out, its chief application was to indicate (to the First Principles Review) which of two residual areas should be preferred – both of them lying at the 5% probability boundary of the original DSTG pdf.
Long story short, drift modelling is relevant to the discussion but it would be unwise to place a high level of confidence in specific predictions.
If folks want to invoke drift models to EXCLUDE any particular hypothesis, then I would like to assume that the modelling approach, implicit assumptions and inherent uncertainties are properly understood.
@Ventus45. As you know, I subscribe to the idea that the route was planned and flown by waypoint in LNAV. I don’t believe there was any pilot intervention much beyond repowering shortly before 1825 – and there is no direct evidence that there was any.
When you shared the details of those oceanic waypoints I tried them both for compliance with the BTO data and I believe that I reported back here. Perhaps the “Arc 2 onwards” school of path-modelling could find a solution, but if there is one it would require manoeuvres and time offsets before Arc 2. There’s no simple/no-loiter path towards RERAB or IGPOL.
The “no-loiter” solutions can end anywhere 38-40S, but the peak [BTO] probability solutions are found between 39.0 and 39.5 using “default assumptions” of a single turn – and the best-match waypoint route to that spectrum of path solutions is the one that I have proposed. Because it involves a 2-stage turn (MEKAR-SANOB-IGEBO) that cuts a little distance off the single FMT model, it also produces a peak probability terminus a little farther south, at 39.6S.
Re:
“… there were published great circle air routes across the SIO, which had mid ocean waypoints spaced 15 degrees of longitude apart.””
Correction:
They are TEN degrees of longitude apart – not fifteen degrees of longitude apart
@Paul Smithson
I said above that “… it is irrelevant whether or not Zahari was alive much past about 19:30utc …”.
The reason I think he would not have set his final FMS input before about 19:30utc, is that he had to remain alert to the possibility of ATC having woken up to his escape, and putting out calls for other aircraft to report any sightings. If anything, to ‘play safe’ he could have initially edged a little further west (perhaps only 20 to 30nm) of your path (thanks for the kmz) towards ISBIX, or even towards the “FIA Corner (2S92E), before actually setting or activating his final waypoint.
In the end, it would not move the 7th arc intercept very much, only aa few nm.
@ventus45 – I hear you. I tried, and I couldn’t find a BTO-comiant route.
@Mike. It is sometimes supposed that the UGIB drift model favours 34S. In fact, the probability distribution exhibits a monotonic increase from 20-something to 37S – and then abruptly falls off a cliff. Hosever, the text makes clear that model results south of that are uncertain due to the small number of particles in the source drift models.
DrB, could you remind us if you drift model is VALID south of 38S? Should that x-axis have started at 37S or 38S rather than indicating probability zero for points south?
@Paul Smithson,
You said: “DrB, could you remind us if you drift model is VALID south of 38S? Should that x-axis have started at 37S or 38S rather than indicating probability zero for points south.”
The starting points used by David Griffin and analyzed by Ulich and Iannello (2023) vary from 8S to 44S near Arc 7 (see Figure 3.3-1). The drift probability calculations are therefore valid over that full range of Arc 7 latitudes (as plotted in Figure 15.1-1). Figure 15.1-2 is an enlarged portion of the latitude range where the probability is not zero, but the probabilities on both sides are virtually zero as shown in the previous plot.
The fact that drift model probabilities are “less certain” at far south latitudes is not because there was a shortage of drift trials there, which there was not (see Figure 3.3-1). It is, in fact, due to two different factors: (a) there were reduced numbers of predicted landings (i.e., lower probability) at the MH370 debris sites, and (b) the time window was insufficiently long to capture all possible landings from the most southerly bins on Arc 7 to the most distant landing sites. So, some late-arriving landings were excluded by the finite length of the CSIRO time window (the duration of the predicted tracks). However, this last effect is not large, but it makes already small probabilities from 40S somewhat smaller (and noisier in a relative sense).
I will also say that the CSIRO drift model is calibrated to match instrumented drifter tracks. So, there is an inherent forcing function to make the drift model predictions statistically match actual drifting debris tracks. I would expect the match to be best for aircraft debris resembling the instrumented drifters. It certainly won’t match sealed tissue packets. The flaperon is unlike the drifters, but CSIRO conducted sea drift trials with a replica flaperon and adjusted their model parameters to match the observed speed and direction. So, I think CSIRO did as good a job of drift modeling as possible given their time and budget constraints.
@Ventus45
If I may, I have two comments on your post November 18, 2025 at 6:05 pm. I have these comments can be seen in a positive way, not as criticism.
You said:
“It’s clear (to me) that Zahari’s last flight was both deliberate and meticulously planned…”
For the record I think like I am sure many do that most likely Zahari’s last flight was both deliberate and meticulously planned. However, no matter how clear it may be to you (or anyone else), until such time that there is irrefutable physical evidence of the state of Zahari’s animacy after that time, the notion he was not alive remains an assumption, and should be acknowledged as such.
You said:
“Moreover, ultimately, it is irrelevant whether or not Zahari was alive much past about 19:30utc…”
Whether or not Zahari was alive past 19:30utc certainly would have been relevant to Zahari. And since you said Zahai’s planning was deliberate and meticulous, it follows that the condition of Zahari being dead or alive after about 19:30utc would be relevant, both in hind-casting his actions prior to then and forecasting his actions afterwards if he was still alive, or the consequences of his actions prior if he was not. If he knew he would not survive after approximately 19:30utc one would expect his behaviour to be likely different to if he believed he would be alive after about 19:30utc., and of in the meticulous planning Zahari’s expected personal endpoint would be a deliberate consideration.
Be careful not to throw out the baby with the bathwater.
@Victor
You asked “I previously asked you whether in the history of aviation we’ve ever seen an aircraft with no autopilot, no augmented flight control, and no pilot inputs fly for many hours before crashing. You never responded.”
I’m struggling to think of any examples. But that doesn’t mean it did not happen this time. The fact that 370 is still such a mystery may be just because it is the first of its kind.
If it had just carried on with a functioning autopilot in HDG or LNAV, we would not be here 11yrs later debating what might have happened.
@Tim: There are cases of incapacitation of a pilot flying without autopilot, but those flights all ended within minutes and never hours after incapacitation.
Your scenario rests on a series of improbable events occurring sequentially.
The inability to find the MH370 debris field does not prove that MH370 was flown without an autopilot. Rather, we are reconstructing hypothetical flight paths with scarce, imprecise data, leading to multiple interpretations based on flight path assumptions.
@ventus45: You and @TBill both claim that the search has failed because we are “in denial”. Yet the two of you disagree on where to search next by thousands of kilometers.
First, what exactly are we denying? That most likely, the captain intentionally diverted the plane? The vast majority of contributors here believe that to be the case, as do key members that comprised the official search team.
Your scenario involves a series of guesses you’ve made. @TBill’s scenario also involves series of guesses. That in itself doesn’t make either of you wrong because we don’t know how the captain chose to navigate to the SIO. What both of you fail to recognize is there are MANY possible guesses, and it is not “denial” to question the uniqueness and data compliance of your proposed solution.
@PaulS
If Boeing were willing to discuss, I would be more general, besides bleed air I suggest it is possible Rt IDG is off from IGARI. I’d also be curious, for sake of argument, let’s say CAPTION is correct about radar flight path and most power off until 1825, what is the expected fuel consumed to Arc1 in a case like that (basically an update to Boeing’s appendix in the SIR).
You want to know if there is fuel to get to 40s. I want to know how much fuel is remaining at Arc7 @~30s to make some estimate of how far beyond Arc7 the aircraft might have flown at reduced altitude.
@Victor
I am trying to be impartial and evidence-based. There is always some need for logic to connect the dots. You criticize educated “guesses” but your suggestion is that intentional ghost flight after Arc2, no maneuvers, fuel exhaust at Arc7, crash in random sea floor location, disregard of sim data to Southeast of Arc7, etc. are somehow the approved guesses, but I do not personally believe any of those assumptions are correct at this juncture. And if all of us had known that Australia knew it was likely pilot deliberate from the start, there might have been less support for a deliberate ghost flight with zero creativity: Pilot planned to just sit there motionless in the cockpit? maybe with a glide at the end.
Re: debris drift, part of the problem is, I suspect the crash could be some distance from Arc7, at least 100-150nm. Citizen investigators, including some contributors here, have used CSIRO/drift models to suggest further off Arc7 to the the Southeast fits other models besides Chari’s. I think Chari could be correct that a point ~70-100nm East of Arc7 at 32.5s fits quite well to give parts like ROY and the flaperon a head start to make the gyre’s U-turn at OZ to get going towards Africa in time for the debris finds. It seems to fit BTO/BFO and sim data as well.
As far as Ventus and the believers in 38-40s, they have the ear of JohnG of Flight Safety Detectives, I tried to comment that I mostly agree philosophically, but I feel the far Westerly end point is probably not correct.
@TBill: The preferred path for UGIB was based on these assumptions:
1. Automated flight with no pilot inputs after 19:41.
2. The BTO and BFO data sets are correctly interpreted.
3. The fuel model is accurate within a defined error band.
3. The CSIRO-generated drift paths are accurate.
4. The contemporaneous weather data is accurate within a defined error band.
You might call those “guesses”, but I would call them our best estimates at interpreting of the data sets used in a way that allows us to define a high priority area for searching. You’ll note I’ve never call it a high “probability” area.
If you remove assumption (1), you are left to guessing the pilot’s intentions, and it becomes impossible to objectively define a point of impact. The fact that you and @ventus45 try to guess the pilot’s intentions and arrive at such widely spaced points of impact is demonstration of this.
@Victor you said If you remove assumption (1), you are left to guessing the pilot’s intentions, and it becomes impossible to objectively define a point of impact.
How strong is the evidence of an inanimate pilot after 19.41?
If no pilot inputs after 19:41 is an assumption (guess) for the purpose of defining a path to and endpoint then doesn’t that make that path highly questionable?
@Dr B
Thank you for your earlier response. I remain uncertain how to interpret the absence of results from southern origins in your dataset. Is it because they take so long to arrive that they “didn’t get counted”?
On the subject of arrival time, I revisited Griffins reports from 2016 and 2017 I-III with a view to quantifying the effect of a) change in “wind” coefficient b) change on directional offset to wind.
Regarding a) you need to compare Figure 3.2.1 from 2016 (non-flaperon, low windage) and Figure 3.2 from 2017 (Part 2). Predicted arrival time at Africa west of 50E (ie Madagascar E coast) from source latitudes 25-35 is shifted EARLIER by approximately 10-12 MONTHS by altering wind coefficient from 1.2% to 3.0%. That is a very large effect for rather a small change in coefficient.
Regarding b) the best-match origin latitude for the flaperon moved roughly 3 degrees southward – from about 35S to about 38S once the model uses directional offset 20 degrees left of wind rather than 10 degrees left. This is because the larger offset takes a tighter loop/shorter distance around the SIO gyre. CSIRO Part II describes the 20 degree offset as being a better match to observed behaviour of the “actual” flaperon as well as a better model for predicting arrival in Reunion. So, again – a rather large effect from only a 10 degree delta in directional offset to wind.
I don’t believe it is possible to claim with any confidence that the totality of objects found is better represented by 1.2% wind coefficient than 3.0%. In fact I would EXPECT it to be at the higher end. This is not because of “direct leeway” exerted on the portion of the object above the waterline, nor because of wave action but because of the vertical structure of currents in the uppermost layer. There is a dramatic (inverse log) relationship between depth and current strength within the first meter. As a result, objects with minimal draft (like flat honeycomb panels) drift a great deal faster than things with 15cm draft (like an undrogued drifter). Near zero draft (seeds, oil slicks) exhibit wind coefficient closer to 5% while more submerged objects (say 10cm-30cm deep) might drift at 1%-3%. There is also a ROTATION of current direction as you go deeper – which means that for everything with at least some draft, an offset to wind direction – to the left in the S hemisphere – is the rule not the exception.
In short, I am saying that small changes in parameters have a demonstrably large effect on drift model outcomes. And that parameter change that would favour more southern origins (higher coefficient and directional offset to the left) are not only possible, but should be EXPECTED.
As for everything the increased likelihood of Australia:Africa for origins south of 36 in the high windage scenario, it is clear that this would be materially altered by even a 10 degree offset to left of wind direction (ref CSIRO Part 2, Fig 3.1 for a visual depiction of impact on drift trajectory).
Finally, it is worth recalling that Griffin notices the (surprisingly) powerful influence current direction at the time of the crash along different sections of the arc. Segments with NW-flow (30S, 35S) had a markedly accelerated westward transit compared to other origins. I don’t think it is a coincidence that the other section of 7th arc that had a strong NW-flow at the time of the accident was 39.5S. You can see this clearly in Fig 3.3.1 (CSIRO 2017, Part 2).
I do not wish to denigrate in any way the excellent work done by Griffin and colleagues. What I do wish to point out is how sensitive the models are to parameters, and the strong possibility that wind-forcing coefficient was higher (and more “left”) than modelled – and that will demonstrably have a large impact on modelled trajectory, arrival time – and ultimately probable origin.
For experimental results on appropriate coefficient for drifters of different types, see Sutherland, G. et al, 2020: Evaluating the Leeway Coefficient of Ocean Drifters Using Operational Marine Environmental Prediction Systems. J. Atmos. Oceanic Technol., 37, 1943–1954, https://doi.org/10.1175/JTECH-D-20-0013.1.
This corroborates the appropriateness of a higher coefficient ~3% for drifter types with minimal depth (iSPHERE or OSKER) and more like ~2% for drifters with depth profile 10-20cm (Roby or SCT), and nearer ~1% for drogued CODE or SVP drifters.
There are multiple other references dealing with similar models and experiments and qualitatively similar results: less depth => greater leeway.
This table from Lodise et al 2019 provides a summary from literature review of leeway results. As well as documenting leeway coefficient of similar magnitude this also illustrates how large the directional offset to wind is (clockwise/to the right in northern hemisphere)
https://os.copernicus.org/articles/15/1627/2019/os-15-1627-2019-t01.png
Again – this substantiates the view that CSIRO may have underestimated the leeway coefficient for flat buoyant items and ought to included directional offset (and sensitivity test) for non-flaperon debris.
@John: We don’t know the probability of (1), which is why I called the recommended search area high priority versus high probability. However, without that assumption, the probability of crash sites gets spread out along the 7th arc and we are left to guessing, which is why those that try to guess (while refusing to admit they are guessing) are recommending points of impact far apart. You could argue that the simpler solution (no pilot inputs after 19:41) has fewer solutions, and therefore has a higher probability of being correct, than any single solution of higher complexity, which is a way of stating Occam’s Razor. That doesn’t guarantee that the simpler solution is correct or even highly probable. Just more probable than any single solution of higher complexity, which is why the simpler solution should have higher priority when searching.
@Paul: Let’s suppose when we estimate a parameter, we have equal chance of the estimate being higher or lower than the actual value. The BFO, fuel model, and drift are three sets of data that all say that 34.2S is better than 39S. The chance of all three being wrong in the same direction would be 1 out of 8.
@Victor. Allow me to correct your framing a bit – I’m not sure I’d go with your 0.5^3. Moreover, you have omitted BTO fit and search results from the compound probability.
34 and 39 are “equally bad” for BFO fit, the peak probability being obtained around 37S.
34 and 39 are also *both* unfeasible for fuel endurance without a non-normal explanation. UGIB proposed packs-off saving of 1.6% based on indirect inference based on B777 documentation remarks regarding recirc fans inop, associated delta pack flow & fuel planning. I proposed a larger saving based on published literature.
I agree that CSIRO’s drift studies favour 35S over 39S. I have set out above why this conclusion is vulnerable to parameter assumption and the basis on which I believe the parameters used may be in error.
So out of those 3 factors, I’d say drift strongly supports 34 over 39, BFO equally supports 34 and 39, fuel endurance (without significant non-normal savings) supports neither.
What you have missed out of your compound probability representation is BTO and search results.
BTO fit favours 39 over 34 by a factor of about 10. I have demonstrated this using relative probability of paths from Arc 2 using log likelihood and DSTG (sans BFO) got the same result.
And search results (which renders 34S vanishingly improbable) while the Smithson spot at 39.6 is virgin territory.
@Paul Smithson: As @DrB has explained, lower BTO error doesn’t mean higher probability if the error is less than expected. Furthermore, by my calculations, the minimum BFO occurs around 34S-35S.
I will concede that I did not include the search results, which I believe makes it difficult to recommend where to search next.
@Victor, lower BTO error does mean higher probability as long as it is not improbably low. So a path with an individual error of (say) 3 sigma is plainly less likely than one with an error of 1.5 sigma. I have already demonstrated that the probability of the good-fitting paths I have identified is NOT improbably good because the probability of obtaining the sample SD is about 0.8. Thus Bobby’s earlier assertion that the BTO fit was “too good” / errors “too small” is emphatically and empirically incorrect.
Your principal measure (z-test) of BTO fit is NOT a goodness of fit metric (probability of observing this collection of values), it is a test for a difference of the MEANS. That is a different thing, and it is much less sensitive to poor-fitting path models with at least one value >2 sigma. UGIB “best fit” path (trial 815, fig 16, p24) has a BTO error of -63 2.17 sigma at Arc 5, in a sample of just 6 values, of which the first, second and last have been forced close to zero by the path-fitting process. It is not a good fit, and paths with much better fit are obtained for LNAV paths ending near 39S.
Correction: in a sample of jusst *5* values [not 6].
@Paul Smithson said: “@Victor, lower BTO error does mean higher probability as long as it is not improbably low.”
When testing the standard deviation of the sample, the probability (p-value) of the chi-squared test for variance is maximized when χ² = n−1, which occurs precisely when the sample variance s² equals the hypothesized population variance σ₀².
@Victor – quite so, we agree! By that measure, my model SD is only mstginally small (p=0.8, two tails) while yours is improbably large (though not p<0.05).
@Paul Smithson,
I did a comparison of two routes – one to 39.5S and the other to 34.3S at Arc 7.
You can download a copy at:
https://drive.google.com/file/d/1-tQYR6DEwTjmkAQH_Ne5wFAo8LmcPqzm/view?usp=sharing
In this table I computed the BTOR probability using both the mean and the standard deviation. The 39.5S Route has a considerably higher probability for the standard deviation, and the 34.3S Route has a slightly higher probability for the mean.
You are correct in saying that the standard deviation for the 39.5S route is very close to the expected value. Indeed, it is not “too low”. There are a few routes with BTOR standard deviations down around 24 microseconds. These results are “too good” and their BTOR probability suffers as a result, but S39.5 is not one of those.
Overall, the BTOR probability for the 39.5S route is about 2.8X higher. For the BFOs it is lower, and for the fuel it is much lower. Thus, overall, the 34.3S Route is more likely, and I don’t believe the S39.5 Route is even flyable.
This table also shows the additional fuel deficiencies required (beyond air packs off and some electrical load shedding) as 6.2% for 39.5S and 1.1% for 34.3 S. I can see how to get 1.1% with a greater reduction in the electrical loads, but how to you get an additional 6.2% to make 39.5S flyable?
All the Boeing and Rolls Royce information we have for the B777 consistently indicate that the air packs off can save about 1.6 % – 2.0 % depending on altitude. Nothing we have found which is applicable to the 9M-MRO aircraft type suggests 3%-5% for the air packs alone. None of the “generic internet” values of 3%-5% you mentioned are known to apply to a B777.
You also said that :” BTO fit favours 39 over 34 by a factor of about 10.”
My table demonstrates that the ratio is about 3X, not 10X.
In another comment you said: “I remain uncertain how to interpret the absence of results from southern origins in your dataset. Is it because they take so long to arrive that they “didn’t get counted”?
No, the primary reason is simply that they did not make many landings at the sites where MH370 debris was found. The “clipping” of the CSIRO time window creates at most a factor of 2 reduction in probability for far-south crash latitudes. That is not enough to make 39S a viable candidate crash site based on debris drift.
You may have forgotten that we already explored the transit speed sensitivity in Ulich and Iannello (2023). We applied a transit speed correction factor to the predicted drift tracks to speed them up or slow them down. We tried fitting this factor separately for each debris, because one expects to have differences in the leeway of various items. We also expected to see the probabilities of the more distant crash sites improve when we allowed the drift speed to increase, but that did not happen. Instead, they just missed at an earlier date. We also wanted to see if some debris became more precise in their crash latitude prediction using this method. While there were some cases of slight improvement, overall, it was not possible to clearly segregate the debris into leeway classes. I suspect this is mostly due to the fact that we had a fairly modest number of recovered MH370 debris items and a smaller number of sites. Also, for many debris the time windows for a match to a find were very wide because of highly uncertain arrival dates. So ,the spatial distribution of the tracks was the most important discriminator for the non-flaperon debris. Another contributor is that It was not possible to simulate the effect of an offset drift direction using the CSIRO drift tracks, but I would not expect this to be a major factor for almost all of the debris. Clearly this happens with the flaperon, but CSIRO measured this and did separate drift track calculations for it. As a result, the flaperon provided the most discriminating crash site information, both spatially and temporally.
Eight years ago, I created this graphic for LNAV paths after 19:41, It shows the general trends of BTO and BFO errors as a function of crossing latitudes. It’s not exact because only one altitude (FL350) was considered, but it helps to frame the problem. Removing the altitude constraint only marginally improves the BTO and BFO errors.
https://mh370.radiantphysics.com/wp-content/uploads/2017/11/GCBTOBFO2.png
The graphic shows that the lowest BTO errors occur near 40S, but at the expense of high speeds (Mach 0.842) which would lead to fuel exhaustion. On the other hand, around 34S, you get better BFO fits, acceptable BTO errors, and lower speeds.
@DrB.
1. Thank you for checking and now confirming that BTO probability is indeed higher at ~39 than at ~34. When you use a “goodness of fit” test rather than a test for difference of the means, you will find the larger probability differential that I referred previously (and have shared previously with you by email). Specifically, using relative probabilities calculated from log-likelihood, the “190” track to 39.35S had a relative probability for BTO “fit” of 0.090 while the “180” track ending at 34.25 had a probability of 0.011 – so a ratio of 8.4 For probability of the sample SD, I get p= 0.78 and 0.28 respectively for 190 and 180 and a ratio of 2.8 – same as you. If you chose (which I don’t) to compound those two probabilities, you would get a probability ratio between 190 and 180 of 23.1:1. I don’t use z-test difference of the means as a test of BTO fit because it is a poor discriminator of goodness of fit.
2. You said “All the Boeing and Rolls Royce information we have for the B777 consistently indicate that the air packs off can save about 1.6 % – 2.0 % depending on altitude.” That is an over-statement of what is described in the manuals you are referred. The fact is that you have made an indirect inference based on recirc fan(s) inop – pack flow rate – fuel planning; then scaled number of fans, flow rate by altitude – to arrive at a fuel delta. What I have done is to consult a wide range of literature, covering a range of different aircraft types – including a specific estimate for an analogous aircraft (A330 with RR Trent 772B). If fuel penalty of bleed air was as small as you suppose, how is it that you can get to a NET fuel saving of 1%+ after substituting bleed air by onboard power generation – including the fuel cost and additional weight of doing that – and why would anyone be investing $millions in developing “more electric aircraft”?
3. I hadn’t forgotten your TSCF. I appreciate that you have attempted to use this to estimate sensitivity to wind parameter. In practice, TSCF is not able to represent this because the wind vector and current vector are different (the latter spends much more time meandering around eddies). Thus a small change in wind coefficient does much more than marginally speed up the “playback”. Specifically, CSIRO demonstrated the very pronounced impact of a change in windage from 1.2% to 3.0% of wind (non-flaperon debris). I will juxtapose the relevant plates for ease of reference in my next post.
I promised earlier to illustrate the effect of small changes in wind parameter on drift model outcome using the CSIRO papers (2016 and 2017).
The first page summarises the impact on non-flaperon debris drift of changing “leeway coefficient” from 1.2% of wind to 3.0% of wind, with no directional offset in either case. From origin latitudes 30-35S, this brings forward “Africa” arrival time by about 6-11 months, representing a ~30%-60% reduction in total transit time from origin to beaching.
The second page summarises the impact of changing “leeway directional offset” on the flaperon’s model trajectory. Moving from 0 to 10 to 20 degrees left shifts the origin latitude with time-matched peak probability from 33S (zero offset) to 35.5 (10 degrees left) to 38.0 (20 degrees left). Arrival time shifts forwards by about 6 months (0 offset to 10 offset) but is little changed between 10 and 20). Across the origin latitude range of interest, the relative probability (Africa vs Australia) shifts in favour of Africa as directional offset increases. After testing the “real flaperon”, CSIRO concluded that the larger directional offset was the better representation of observed drift behaviour.
Dr B says above “As a result, the flaperon provided the most discriminating crash site information, both spatially and temporally.” If that is the case, then by CSIRO’s models, 38.0 is the latitude of maximum probability, and the chances of originating at 39.5S or 34.2S are roughly comparable (and both relatively high).
Conclusion: Drift-modelled “crash latitude of highest likelihood” is critically dependent on wind parameters. Leeway coefficient magnitude and direction have a demonstrably large impact on the predicted transit time, the relative probability of different origin latitudes, and the relative probability of drifting towards Africa vs Australia.
@Victor, referring to your post above. We agree that there is a probability peak circa 40°S for BTO, and that south of there requires speeds >0.84 at altitudes FL350 or above.
You have used the RMS BTOR metric, characterising the fit at 34°S as “acceptable” without quantifying relative probability.
As you know, log likelihood also uses the square of the errors, but it has the merit of using reference population sigma to allow quantification of relative likelihood between models with the same number of parameters.
Using the spectrum of LNAV paths at FL350, in one degree increments from 175° to 195°, starting at Arc 2 and optimised for log-likelihood across Arcs 2-6, we obtain the BTO and BFO probability distributions shown here.
If speed is capped at M0.84 (altitude FL350), the distribution still peaks between 39°S and 40°S, but falls off very abruptly south of there. In contrast, the path to 34°S is closer to the tail of the distribution, with relative probability very much lower than the mode.
As stated previously, I obtain BFO fit maximum around 37°S with more or less symmetrical distribution either side. The distribution will obviously get narrower the smaller BFO sigma you elect to use.
@Paul Smithson said: You have used the RMS BTOR metric, characterising the fit at 34°S as “acceptable” without quantifying relative probability.
Acceptable means the null hypothesis should not be rejected. It is independent of relative probability. Meanwhile, @DrB has shown there is enough fuel for 34S, while paths around 39S fail in that regard.
You throw out our best estimates at modeling drift and fuel consumption. Of these, I think the shortage of fuel is most problematic.
@Paul Smithson,
You said: “Dr B says above “As a result, the flaperon provided the most discriminating crash site information, both spatially and temporally.” If that is the case, then by CSIRO’s models, 38.0 is the latitude of maximum probability, and the chances of originating at 39.5S or 34.2S are roughly comparable (and both relatively high).
We showed the flaperon-only probability in Ulich and Iannello (2023) based on the CSIRO drift tracks, in Figure 8.7-1. This shows two peaks, with a higher one at 34S and a lower one at 38S. the drift probabilities at 39S and 40S are considerably lower.
You also said: “I don’t use z-test difference of the means as a test of BTO fit because it is a poor discriminator of goodness of fit.”
Why do you say the Z-Statistic with a gaussian distribution is a poor discriminator of goodness of fits? It’s fine for that. In fact, if you look at the formula for log-likelihood (which you say you are using), it is this:
LL = ( -N/2)*LOG(2*PI*sigma^2) – (1/2)* SUM{ [(X – mu)/sigma]^2 .
Since Z = (X-mu)/sigma, the second term in the log likelihood equation is simply the sum of the squares of the Z-statistics for each data point. So, you are using Z-statistics even though you don’t realize it.
Here is what I get using log likelihood:
39.5 S, N = 5, STDEV = 27.7, RMS = 24.8, sigma = 29, SAMPLE MEAN = -1.8, LL = -1.46
34.3S, N = 5, STDEV = 37.7, RMS = 33.7, sigma = 29, SAMPLE MEAN = 0.02, LL = -2.70
These Log Likelihood (LL) values are just the second term in the equation above, because the first term is just the same constant for these 2 cases.
Their relative probability is = EXP(-1.46+2.70) = 3.4. I will compare this result with two other methods below.
You seem to have an error in your calculations of log likelihood. It may be that you also used STDEV incorrectly.
The SUM[ (X-mu)^2] in the log likelihood equation cannot be estimated with the STDEV, because that uses the estimated sample mean rather than the known mean of the distribution (mu). Instead, you should use N*RMS^2.
So, a correct formula for LL is, when mu = 0 :
LL = -(1/2)*SUM [ (X – mu)^2 ] / sigma^2 = -(1/2)*N * RMS^2.
Thus, it’s actually the ratio of the squares of the RMS values that counts in Log Likelihood, not the ratio of the squares of the standard deviations of the BTORs. That’s why I usually try to match the RMS value when fitting, not the STDEV value. In effect, that also forces the sample mean to match mu. So matching the RMS is, more or less, equivalent to matching the STDEV to sigma and the sample mean to mu.
I also did the chi-squared probabilities as follows:
CHISQ = (N-1)*(STDEV/sigma)^2
39.5 S, N = 5, STDEV = 27.7, sigma = 29, SAMPLE MEAN = -1.8, CHISQ = 3.66, p = 45.4%
34.3S, N = 5, STDEV = 37.7, sigma = 29, SAMPLE MEAN = 0.02, CHISQ = 6.76, p = 14.9%
Their relative probability is .454 / .149 = 3.05. So CHISQ uses the STDEV ratio rather than the RMS ratio, as done in Log Likelihood.
From the Table I recently posted, the ratio for the standard deviation probability using the Z statistic =96.6% / 30.9% = 3.13.
So, when I do it, these three methods give consistent results of close to 3X for the ratio of the standard deviation probabilities:
Log Likelihood 3.4X (based on RMS)
Z-statistic 3.1X (based on STDEV)
Chi-Squared 3.1X (based on STDEV)
Thus, the slight differences in relative probability arise from one method (Log Likelihood) using the variance about the distribution mean, and the other two methods are based on the variance about the sample mean. That doesn’t change the probability ratio much (3.4X versus 3.1X).
If you are getting a different ratio, I would suggest checking your formulas against the equations I listed above, because you may have an error.
Your plot of BTOR probability versus Arc 7 latitude appears to be infected with this error. I have demonstrated three methods of calculating the relative BTOR probability of 39.5S versus 34.3 S, all being about a factor of three. Your plot shows a factor of 8.2X. Your plot is wrong. If you used the same process for the BFOR probability, it is probably wrong too. It has another problem in that you assume sigmas for the BFORs of 5 Hz and 7 Hz. Those are much too large. In Section F.3.4 in UGIB (2020) we demonstrated the route fitting BFO residuals, with samples about 1 hour apart, has a standard deviation of 2.3 +/- 0.4 Hz. This includes the contributions due to read noise, trig noise, and the OCXO Allen Deviation. It is also consistent with Inmarsat’s criterion that the correct route will have BFORs within +/- 7 Hz. In addition to the calculation error described above, your assumed BFOR sigmas much larger than the MH370, MH371, and MHXX data support. That error in sigma suppresses the probability variations seen across Arc 7 latitudes in your BFOR plot.
DrB,
1. Re Flaperon drift model, your Figure 8.7-1 seems to differ from CSIRO (Part II) Figure 3.2 where the mode is very clearly 38S.
With reference flaperon drift characteristics, CSIRO remarked: “Figure 3.1 shows that leeway angles of 10° and 20° both (but especially the latter) put the most common path of the simulated trajectories fairly close to Ile de La Reunion (which lies at 21°S), while Figure 3.2 shows that July 2015 is close to being the most likely arrival time, for all potential crash sites from 40°S to 30.5°S. The actual flaperon arrival time and place is thus more consistent with the new model than our earlier one…”. In Ulich & Iannello (2023) you said “One set of 86,400 trials used the drift parameters CSIRO determined for the flaperon by sea trials with a cut-down flaperon.” Which version of flaperon drift did you use?
2. Re possible error in my LL numbers – thanks for taking a closer look. I have checked mine and so far have found no errors. When I am done I will reconfirm one way or the other and am happy to share the spreadsheet so that you can examine the formulae and identify any error I have overlooked or otherwise satisfy yourself that they are error-free.
3. On appropriate BFO sigma, I need to “disagree agreeably” with you. DSTG established empirical “noise” standard deviation based on data from 20 preceding flights and published this at table 5.1 of Bayesian methods. The values are: inflight only 5.5Hz, 4.3Hz (incl/excl outliers); including values on tarmac 5.0Hz, 4.0Hz (incl./excl. outliers). Their words: ““The statistics show that even when outliers are discarded, a SD of about 4.3Hz is applicable. As discussed above, to be conservative and allow for potential variation in the bias value on the accident flight, our model assumes a noise standard deviation of 7Hz.” I recognise that you elect to use tighter uncertainty boundaries than this.
@Paul Smithson,
1. You said: “1. Re Flaperon drift model, your Figure 8.7-1 seems to differ from CSIRO (Part II) Figure 3.2 where the mode is very clearly 38S.’
I did my own independent analysis starting with the same drift tracks used by CSIRO. I purposely did not compare my probabilities with the CSIRO results during the process. I also did double blind testing using simulated data. In addition, my probability calculations are done with time and spatial windows which are optimized for each debris to maximize crash latitude discrimination. . There is general agreement with CSIRO’s results, but where we differ, I think my method does a better job of extracting the information in the debris reports and providing better latitude discrimination.
2. You asked: “Which version of flaperon drift did you use?”
I used the flaperon trials recommended and provided by David Griffin. The flaperon parameters are described in Chapter 5 (Summary) of CSIRO’s report:
The search for MH370 and ocean surface drift – Part II
David Griffin, Peter Oke and Emlyn Jones report number EP172633 13 April 2017
which states: “Modelling the flaperon’s drift as being 20° left of the wind, and 10cm/s in excess of the Stokes Drift . . . .“
This is based on their field tests of an actual B777 flaperon, not the initial replica.
3. You said: “On appropriate BFO sigma, I need to “disagree agreeably” with you. DSTG established empirical “noise” standard deviation based on data from 20 preceding flights and published this at table 5.1 of Bayesian methods. The values are: inflight only 5.5Hz, 4.3Hz (incl/excl outliers); including values on tarmac 5.0Hz, 4.0Hz (incl./excl. outliers). Their words: ““The statistics show that even when outliers are discarded, a SD of about 4.3Hz is applicable. As discussed above, to be conservative and allow for potential variation in the bias value on the accident flight, our model assumes a noise standard deviation of 7Hz.” I recognise that you elect to use tighter uncertainty boundaries than this.“
Regarding the BFOR dispersion, DSTG could not justify using a value greater than 4.3 Hz, or 4.0 Hz excluding outliers. Your values of 5 and 7 Hz are too large. By “being conservative” DSTG needlessly degraded their crash latitude discrimination. You are doing this too, but seemingly for a different reason (possibly personal bias in favor of a particular crash latitude).
My independent analysis of the BFOR statistics segregated the three noise components and provided a useful measure of their combined effects (2.3 +/- 0.4 Hz for samples at 1-hour intervals). DSTG does not provide any error bars on their values. I suspect DSTG’s data set may have some data with ROC contamination, which I am unsure if they compensated. The only BFO in the MH370 data set which may be contaminated by ROC is the one at 19:41, and then only by a few Hz.
4. You also said: “When I am done I will reconfirm one way or the other and am happy to share the spreadsheet so that you can examine the formulae and identify any error I have overlooked or otherwise satisfy yourself that they are error-free.”
In a previous comment I demonstrated three methods of computing the BTOR standard deviation probability (including equations and a numerical example), which gave consistent results. Your answer varies wildly from those. Clearly you are doing something wrong. It’s not my job to fix your error.
I am curious, if someone can explain me or tell me where I can read, how did the ATSB treat the “problem” with fuel amount and southern latitudes, cca. 39.6-40.2S?
That area is still included in ATSB’s PDF and basically the only large area which was not covered, and which is, I presume, still valid place to search according to their analyses.
I guess that would mean that the turn south happened just after 18:25, or there is something else? I tried to find myself that specific answer, what would be the track and the speed of the aircraft in ATSB’s work, but I couldn’t.
@Marijan: In the DSTG report, in several places, it states that the speed was limited to the range Mach 0.72 to 0.84 with no further fuel limitation. For instance, in section 8.3 which lists the assumptions of the particle filter model, it includes:
4. Infinite fuel: the fuel constraints on the aircraft can be applied to the pdf afterwards. In the simplest case, maximum reachable ranges could be used to censor impossible trajectories. However, analysis of candidate trajectories has indicated that the majority are feasible. Broad information about the fuel consumption rate of the aircraft has been used to inform the range of allowable Mach numbers.
7. The aircraft air speed is limited to the range Mach 0.73 to 0.84. Fuel consumption becomes very inefficient at speeds higher than this and at lower speeds the aircraft is not able to match the measurements. In practice it is likely that the viable range of speeds is actually much narrower than this.
The absence of a true fuel model is generally recognized as an important limitation of the DSTG analysis.
Thank you, now it is clear
@Marijan, I have been championing the path solution that ends around 39 6S. The proposed route is: MEKAR-SANOB-IGEBO-RUNUT and the towards (but not reaching) 40S85E. The route produces excellent agreement with the BTO data at M0.84, FL360. In an earlier post I have discussed the model fuel deficit and means by which the apparent shortfall could be explained.
@PaulS, DrB, VictorI, Marijan
FYI: my fuel calculations* give 8.2 higher probability for 33.0-36.0 vs. 39.0 – 40.0, after compensating for the number of available paths is the bins (they tend to decrease in number towards S40 because of a latitude cut off applied for 19:41).
(*Based on Bobby’s model, with different 19:41 assumptions and Monte Carlo approach)
The problem seems to be that we have to mainly consider three unlikely options:
1. A crash in the S32 -S37 range close to the arc, but wreckage was missed in search efforts
2. Glide, away from the 7th arc, while BTO/BFO + fuel analysis are consistent with fuel exhaustion and no pilot inputs in the final hours.
3. A crash location at a latitude that seems less consistent with fuel, BTO/BFO and drift models
I would want to make sure that the indicated few percent probability that the wreckage was missed in a previous search effort is correct. Typically, if you are looking for something and it is difficult to find, it is really important that you know what you are looking for (in terms of shape and size). So, one question I have is how debris fields could look like depending on type of impact, local water depth, currents, wind and waves.
@Victor
I do not think we are going to find the aircraft without correctly guessing the pilot intentions.
Let’s face it: we all are making that guess. It is a trial-and-error problem, and the correct answer (which I do not see us iterating to get there) is going to be, who makes the correct guess? Tim is adamant of a speculated O2 cylinder accident causing a meandering Ghost Flight. IG/ATSB have been advocating “Hybrid” Ghost Flight scenarios, whereas in IG’s case, the pilot had a plan to aim for South Pole and then perhaps died earlier and his body slumped over the steering column at Arc7 to cause an otherwise technically unexplainable steep dive at Arc7. The 38-40s crowd likes a Hybrid Ghost Flight with active pilot: they feel this makes sense because a straight edge ruler goes there nicely, it is max distance from OZ on Arc7, the later sunrise gives max cover of darkness until crash, and the DSTG Bayesian model confirms for them that that solution has merit.
I would say we have plenty of evidence of active pilot. The BTO/BFO really tells at least two main stories: either (1) a straight edge (great circle) LNAV no-maneuver fight path or (2) a flight with maneuvers: a slow down/descent and “curve” into Arc7. The body politic has been adamant that curved path (2) should be ruled out or require super-proof. But there is an obvious possible explanation that the “curved” path can be considered merging onto exactly where the home simulator cases went, along a 180s CMH operating line. The loss of SATCOMs can be either (1) fuel exhaustion or (2) pilot turning off SATCOM to run past Arc7 clandestinely, keeping in mind the “curve” into Arc7 scenario still has fuel to run. The descent at Arc7 is either (1) technically unexplainable autonomous descent, or (2) pilot deliberate either (a) to crash right on Arc7 or (b) strategic descent, I say probably into the thick cloud layer for cover. Not to mention debris drift makes sense from say 32.5s/100e eg on Broken Ridge. Not to mention lack of success of extensive close-to-Arc7 searches.
Part of the logic problem was searching before debris was found (in ATSB case) and before the “complete” sim data was available for analysis in both cases. “Definitive” answers were developed before that evidence with little appetite for update revisions in the face of actual evidence.
Overall post-mortem I would say China abdicated, for a while Malaysia let Australia handle it. Although Australia secretly knew from Malaysia it was likely pilot deliberate, Australia advocated a conservative approach, I would say respectful of aviation industry input and public concerns, to require definitive proof of the active-pilot-end scenario. Whereas although the active-pilot-to-end seems obvious from the simulator data, it was considered “fair game” to question the active-pilot-to-end scenario if there could be a hypothetical ghost flight explanation invented to explain the data in a less nefarious way. Simply put the crowd-source energy has largely been trying to absolve pilot, with no success, but there is not the inclination in human nature to address the criminal scenario with that same vigor.
@TBill said:
The descent at Arc7 is either (1) technically unexplainable autonomous descent, or (2) pilot deliberate either (a) to crash right on Arc7 or (b) strategic descent, I say probably into the thick cloud layer for cover.
What evidence is there that the pilot had any knowledge that the concept of Arc7. Wouldn’t that require a predictive position for the IMARSAT satellite as well as advance knowledge of an accurate model of the flight path and fuel exhaustion position?
@TBill: I think it is unlikely we will guess how the pilot flew the plane because the possibilities are endless, as are the associated points of impact for each scenario. You think it’s possible to guess.
The steep dive could have been caused by a re-start of one of the engines after fuel exhaustion, causing a high bank angle leading to a rapid descent. That scenario was observed in one the simulations that Mike shared. There might be other scenarios also resulting in a steep descent soon after fuel exhaustion with no pilot inputs.
I believe there are two categories of flight paths:
1. Automated flight with no pilot inputs after 19:41.
2. Semo-automated flight with pilot inputs after 19:41.
We don’t know how to assign the relative probability of (1) versus (2). However, if (1) is true, and if the fuel model that @DrB developed for UGIB 2020 is accurate, the air packs must have been off, and the plane crossed the 7th arc around 34S latitude with no pilot alive. On the other hand, if (2) is correct, there are many possible scenarios (not just the one you advocate), so a search strategy should prioritize (1), as it has the minimum number of assumptions.
@Victor
I think there is a clear and distinct difference between a ‘guess’ and a ‘hunch’ (that is based on deductive reasoning of a ‘plausible criminal intent’ of a technically proficient actor).
The ATSB championed the “ghost flight” for three reasons.
(a) To avoid any suggestion of the possibility of deliberate ‘criminal intent’, due to ‘denial sensitivities’, and
(b) To reduce the search problem to a simple ‘unresponsive crew’ crash on or very close to the 7th arc, which
(c) Allowed the ATSB search to proceed based on public and political support at the time.
The problem (a decade plus later) is that assumptions (a) and (b) are clearly no longer credible.
Therefore, continuing with (c) – the ‘close to the 7th Arc search’ is nothing but a ‘hiding to nothing’.
What (if anything) is there to be gained, by continuing to ‘mow the 7th Arc lawn’ ?
This is no longer a ‘search’ in the conventional sense; indeed, it should never have been so. It is clearly a crime to be solved, and to do that, you have to think like a detective, which means, getting into the head of the criminal actor.
Motive, Means, Method.
Different people have different “hunches”. You point out that TBill and I are in wildly different (but closely defined) positions, which have been derived from wildly different assumptions and deductive reasoning. Other people also have closely defined positions derived from different assumptions and reasoning. All reasonable proposals should be prioritized to be checked out individually, rather than simply ‘continuing to mow the lawn’ and praying for a miracle.
@ventus45: The DSTG study did not assume a “ghost flight”. Rather, it prioritized flights with fewer maneuvers. The formulation of the problem did not include pilot’s intent because there was no way to guess, i.e., assign mathematical probabilities to all the potential scenarios.
It seems like your recommendation is to make a list of possible 7th arc crossings based on everybody’s hunches, somehow decide which are reasonable, and then search out to 120 NM from each crossing based on a long glide after reaching the 7th arc. Now that would be quite a search area!
In fact, this is the only way to conduct a successful search for MH370 with almost 100% certainty.
If we trust the CSIRO drift data and satellite images of possible debris (PHR1-PHR4), then the most reasonable approach would be to conduct a broad search of 33S-36S, out to 120 nautical miles.
At this point, this seems the most reasonable to me.
Although I don’t rule out the possibility that it could be further north and also east of the 7th arc.
DrB. Responses to your stats critique posted 22 November.
You correctly represent the formula for log likelihood – which is what I have been using. You said “So, you are using Z-statistics even though you don’t realize it.” I very much DO realise it.
You have arrived at a different LL number and ratio of probabilities for paths ending 34.3S and 39.5S. The path models you are testing evidently have different residuals. You know what my numbers are because I shared them with you previously and subsequently by email. Here we are again:-
My 180 track, reaching arc 7 at 34.25S has BTO residuals: 0.0, 39.5, 17.8, -59.3, 21.8. Mean 4.0, RMS 34.3, STDEV 38.0. Log likelihood (including the first term) -24.92.
My 190 track, reaching arc 7 at 39.35S has BTO residuals: 0.0, 16.9, 17.2, -39.0,13.8. Mean 1.8, RMS 21.4, STDEV 23.8. LL -22.79
The ratio of likelihoods EXP(LL) is 8.39, as previously stated – much higher than the 3.4 ratio that you have computed. If you feel I have somehow fitted a sub-optimal model of the LNAV path to 34.25S (or nearby), by all means supply your own path model BTORs.
You said: “You seem to have an error in your calculations of log likelihood. It may be that you also used STDEV incorrectly.” Wrong on both counts. My formulae and calculations are correct and I used reference population sigma, not sample SD as you wrongly guessed.
You go on to explain you fondness of matching RMS – but, for reasons that escape me – seem allergic to using the accepted log-likelihood methodology for quantifying relative probability of comparable models. Without such a method, it is meaningless to say that an RMS of 33.7 is “acceptable”, because we haven’t quantified how much better/worse it is than another number.
You said: “If you are getting a different ratio, I would suggest checking your formulas against the equations I listed above, because you may have an error.” I have, and I don’t.
You said: “Your plot of BTOR probability versus Arc 7 latitude appears to be infected with this error. I have demonstrated three methods of calculating the relative BTOR probability of 39.5S versus 34.3 S, all being about a factor of three. Your plot shows a factor of 8.2X. Your plot is wrong.” It is your unfounded assertion that is wrong, not my calculation or plot.
I stand by my previous statement “I don’t use z-test difference of the means as a test of BTO fit because it is a poor discriminator of goodness of fit.” This test does not take consideration of the magnitude of constituent datapoint errors. Across the track angle range 180-195, the mean BTOR of my optimised LNAV paths ranges between 1.8 and 4.0 and the associated 2-tail p-value ranges between 0.76 and 0.89 – in other words, by this test, they all look great! Your z-test of means is only of use to throw out models with extremely poor fit and does not help to discriminate among paths models across the region of interest. In other words, it doesn’t add PRACTICAL informational value for the problem at hand.
I am entirely unsurprised that your idiosyncratic set of statistical tests for BTO led you to the conclusion that it didn’t help much in narrowing down the field. In contrast, I have demonstrated – just as DSTG have done previously – that BTO fit does produce objectively better fit for paths ending around 39-40. The fact that it does so equally well – and with the same peak latitude – from 1840 and from Arc2 tells you that a manoeuvre after FMT is superfluous to fitting a BTO-optimal path.
That is a very good reason to take the southern end seriously and to re-examine the assumptions that have led you and others to start your models from an Arc 2 latitude that -without inserting a 40 minute delay – is incompatible with the aircraft’s last known position. If the southern path solutions can be shown to be fuel compatible, then the principal basis for preferring 34S falls away.
@Paul Smithson
Thank you Paul, I am aware that you have been developing these southern routes for a while, we discussed it briefly about a year ago around the same time of the year. Fuel amount was also an issue back then, but I believe if there is even a minor chance for a plane to reach that area OI should search it along the 7th arc rather than moving away from the 7th arc for northern routes.
@Niels
ATSB did the analysis of debris fields for previous accidents. The ATSB final report on page 83 states that:
“A historical study was conducted of available previously sonar mapped aircraft debris fields. Measurements of each field were made and aircraft size compared with any impact information available for each crash site. It was found that the area of a typical debris field for an aircraft the size of a B777, with any foreseeable crash scenario, in similar water depths similar to the MH370 search area, would be at least 100 m x 100 m and very likely to be greater than 200 m x 200 m.”
After the ATSB led search there were about 600km² of data gaps and 2500km² of low probability detection areas which are 100x100m and larger. The ones larger than 200x200m comprise about 140-150km² for data gaps and 170-180km² for LPDs, but I am not sure if my interpretation of data in the final report is correct.
For what it’s worth, we have no reason to believe that OI won’t continue the search where it left off, completing the area it proposed in March 2024 with perhaps some additional area at the edges.
I am convinced that MH370 will only be found using forensics, not guesses about the intent of the pilot. The official investigation followed scientific methods despite the many speculative claims that have distracted from the start and dominate the media today.
A scientific approach requires not only staying within the reality of known physics, but examining conflicting factual evidence rather than dismissing it. If the hard evidence doesn’t seem to match, then something is wrong with the assumptions or the hypothesis, and our assumptions can be clouded by unconscious bias.
My candidate site is based mainly on hydrophone and seismic recordings that can be validated by others. However, I didn’t make the connection to the Java Anomaly and the 7th Arc until 2018 because I was caught up in looking for events where a consensus of experts were guiding the searches. Although the Java event was very prominent on the H08 hydrophone data as soon as I began analysis, I accepted the expert dismissal of it as a low level geological event. Even though I was using seismometers in 2017 to substitute for infrasound detectors at Cocos Island, I still hadn’t connected the dots to analyze the 7th Arc near Java.
I had been caught up by attachment bias, namely the assumption that there was only the Final Major Turn (FMT) after leaving RADAR, and no pilot after that. The basis for that assumption is clear. Without further evidence, only an unpiloted path could be computed to match the seven ping arcs.
Had I made the acoustic connection to the Java site early on, I’m sure it would have been included in previous searches. Now, the assumption of an unpiloted path appears cemented in place. Because the underwater searches haven’t matched the predictions based on that assumption, the constraint is being relaxed to allow a piloted glide at the end, but still sticking to the basic assumption. Candidate sites away from the 7th Arc require dismissal of factual evidence plus new speculation.
The Java evidence is new since it wasn’t reported until after the previous underwater searches ended in 2018. It thus meets the requirement of “new evidence for a specific location”. Others may challenge my interpretation, but nobody has disputed the factual findings of an anomalous event on the 7th Arc at that site, 56 minutes after that last ping. I believe it is consistent with a large section of sinking MH370 debris hitting the seabed (not a surface impact, and clearly not an intact plane).
It turns out there was no cataloged geological event matching the hydrophone timing. A cataloged M4.1 quake at 23:52:18 in the Java Trench that day is nearly undetectable on the same hydrophones, vs the 01:15:18 source event dominating all other noise sources. That makes it anomalous. Appendices H and I in the official report both include detection of the Java Anomaly with exact arrival matches to H08 and Scott Reef hydrophones.
My earlier supporting infrasound evidence at Cocos Island matches a flyby there with a turn at the airport waypoint towards Christmas Island, where another doppler shift was detected matching timing for steady low altitude flight path at holding speed.
Last year, I was doing photometric analysis of the flaperon barnacles to gauge their size and thus age. The largest one observed was 39mm. Zoomed in, I was surprised to notice some young adult barnacles growing on top of leading edge damage that could only have come after beaching. Barnacles don’t move after attachment, so their entire growth before discovery must have been after beaching.
I’ve presented that image here and in my 370Location.org report. The French DGA/AT and BEA have an open investigation including the flaperon. They likely have very detailed photos of its discovery.
I’d like to put out a call for any sleuths to find more detailed images of the inboard leading edge of the flaperon at Reunion Island.
There has been ongoing discussion of the drift studies. Most analysts make an assumption about the transit time being near the date of discovery. Some include debris finds other than the flaperon. The “Roy” cowling found in South Africa was the second piece found, yet it traveled the farthest of all discovered debris. That makes only those first two pieces key to drift transit speed. CSIRO initially considered that there was not enough drift time to get from the active search areas to Reunion. Only by applying windage and stokes measured from a real flaperon could it go just fast enough to arrive at the end of July 2015. The barnacle photo evidence breaks that assumption.
The the flaperon discovery minus arrival time equals the duration of barnacle growth. Preliminary reports and assumptions are that the barnacles grew for 15 months starting in cold SIO waters. New photo evidence shows that they likely were only growing for several weeks. Detailed chemical and isotope studies of the flaperon barnacles all agreed that they began growth at the warm end of their reproductive range as the surrounding waters cooled past 25C, then slowly warmed again. The evidence mismatches a cold water crash site, so has been broadly discounted. This temperature curve is a close match for sea temps around Reunion in mid May 2015, not the end of July. That fundamentally changes the assumptions that were made for most drift studies, and thus the validity of the results.
Some authors like JW have interpreted faster barnacle growth to mean that all of the debris was planted to match a conspiracy of Russians flying the plane to Kazakhstan by hacking the BFO timing (which nobody outside IMARSAT even knew of until MH370).
What is consistent with the new evidence is that the crash site was in tropical waters. The surface currents and Southern Equatorial Current carry debris more directly from there to where it was eventually discovered, and only debris that encountered cooler waters below about 25C developed barnacles. Regardless of how quickly the barnacles grew, that threshold puts the crash site north of around 21S. If the barnacles grew more slowly, then the flaperon must have arrived that much earlier.
There are numerous flyable paths with exact BTO arc matches to the Java site, and it does not depend on BFO or fuel endurance statistics, or the intent of the pilot. It requires examining new evidence rather than endless speculation.
The Java candidate has been discussed here before, and this forum may be the only place that invited rational discourse. Based on relative precision compared to other candidates leading to minimal cost for searching, some assurances were made made that the Java site would be included in any search recommendations, though after the priority areas had been searched. MH370 is either there or not, and it should at least be excluded by a quick seabed search in passing.
Because I have been openly critical of unscientific approaches like misuse of the acoustic evidence or WSPR, the GT/RG/BG influencers on social media have lumped the Java candidate among their 5 most ridiculous MH370 theories like alien abduction. They refuse to even reference the source, characterizing me as “some guy” out in “MH370 zombieland” with a “pet theory” that will say anything to appear credible. At the same time they claim they are the ones being defamed.
Meanwhile, my acoustic analysis continues. Not yet written up, using new noise reduction methods I have cleanly isolated a unique signature focused on the 01:15:18Z Java event location that has a common arrival with 40 nearby seismometers. That same pattern is clearly seen at about 1/10 the intensity with an origin around 00:26Z. This would be consistent with dense debris like a jet engine core sinking rapidly in the estimated 6 minutes to 3400m depth.
Even if OI does not decide to specifically search the Java site, they might take a pragmatic approach en route to resuming their next search. The BTO is considered accurate within 5 km or so. The IG is adamant about a high speed dive at the 7th Arc. We don’t know if there were more turns after FMT. Heading into the SIO from Singapore, it would only be 350km out of their way to capture unmapped bathymetry of the 7th Arc starting at the Java coast. I suspect they could do that at cruising speed. Towfish needed the prior bathymetry, but OI could slow down and run a single strip narrow AUV search of the 7th Arc down to their destination. They could expand that strip each time they provision in Singapore or include new ships. That data could be valuable as a hedge against a fruitless search, and they just might find it within days of resuming.
Thanksgiving Thanks for allowing me to contribute here. With hope, Ed Anderson
Happy Thanksgiving to everyone in the US
Victor, you said previously that you actually have little confidence the plane would be found in the remaining part of the search area I proposed, which is mostly based on UGIB 2020.
Do you have any insight into whether the OI is bound by contract to search the whole proposed area, or do they just need to cover a specific number of square kilometers?
If so, why continue in the area that the original authors of the paper (at least you, certainly R. Godfrey and I am not sure about the others) think is not promising anymore?
@370Location said: The IG is adamant about a high speed dive at the 7th Arc.
This is often repeated but not true. The final BFO values indicate a downward acceleration of around 0.7g, and therefore a crash near the 7th arc seems more likely than far from the 7th arc, all things beings equal. However, we also understand the possibility of a recovery from the dive and a long glide.
We haven’t seen your yet unpublished work. Please share it, and please don’t give up. As you say, searching your spot would be relatively inexpensive.
@Marijan: I’m not sure I said I have little confidence that plane in the search area that remains. What I mean is the high priority was searched, and I don’t know the probability of success of the remaining area. That said, I would still advocate that the proposed search that was previously started be completed. I think it’s still the best option at this time.
I don’t know the details of the contract between OI and Malaysia.
@Victor,
If the search box is correct and 90% has already been searched, then shouldn’t the probability of finding the plane in the yet to be search area be pretty high? That’s again assuming the box is accurate. Thanks in advance.
OK, I got the impression that you sort of “gave up hope” that the plane will be found in the remaining part of the current search area. I also think it is not good to quit what was already started, but in this case, considering that in the remaining area the probability of finding the plane is more or less the same as anywhere beyond the 25nm along the 7th arc, and sorry to say this openly, the remaining time and resources allocated for the search can be better spent.
@VictorI:
My suggestion for scanning an unmapped area en route was that nearest the 7th Arc has always had the highest probability. I realize there’s lots of nuance with the last ping BFO. I wonder how wide the swath would be with three HUGIN AUV.
For the writeup, there’s more seismology needed to explain the curious details of the different wave propagations (1.2 Hz Scholte waves). Further validation of my findings hasn’t mattered so far. Still, here’s a spectrogram of the stacked arrivals, with each seismometer aligned to the expected P wave arrival from the Java candidate site. There are faint signs of a surface impact just after 00:20Z and an unknown narrow 1.5 Hz blip at 00:25Z. The clearly visible event is at around 00:26:18Z (about 6 min after impact) and 00:15:18Z + 00:15:56Z (56 min after impact implies a large section floating for over 30 min).
https://drive.google.com/open?id=1_j8H5UDzef96SuTbVyeAeFOLBhH2mcEZ (dated Aug 18)
The 00:15:18Z epicenter is directly on the 7th Arc at shallow depth, with the epicenter accuracy probably within +/- 1 km. It was too unlike a quake to be cataloged, but just strong enough to pinpoint it. I hope that an expert seismologist might run the epicenter and depth calculation with better tools.
@370Location,
Ed, your idea of progressively scanning the as yet unscanned “top end” of the 7th Arc is actually a very good one, highly meritorious in fact.
There are many ships out there with “multi beam echo sounders” that could (if asked by Seabed 2030) scan “segments of the 7th Arc” (perhaps up to 150nm either side of it) whilst in normal transits when “crossing the arc” between “jobs”.
Over time, many such transits would gradually build up “a patchwork quilt” which, if regularly updated and published, would allow captains of suitably equipped ships, to “select” a virgin segment close to their otherwise planned track, which they could transit at “best scan speed” (depending on the type of equipment they have, the sea state etc) with minimal time delay, additional distance travelled, or cost.
It may be worth directly approaching Seabed 2030 for their assistance.
Thanks to Victor for this blog and to all the contributors here. So glad to see work from people like Ed encouraged on this blog as a reader for over a decade. Hope that the tireless efforts of all the researchers will yield a fruitful solution in this next search.
@Gregg: Welcome to the blog.
Although some of us believe the search area should have high priority, we really can’t assign a probability that the debris is somewhere in the search area.
@John
Arc 6/7 is where SATCOM was turned off (actually turned off twice perhaps a mistake). A few of us (not necessarily on this blog) believe Arc6/7 is where the actual “getaway” started to fly without radio contact. Keep in mind a SATphone call such as at 23:14 goes thru to the aircraft and tells us three things (1) pilot sees incoming call “ringing” (2) tells us aircraft still flying and (3) we get a BFO (trajectory) hint, not mention we see the SATCOM “IP Address” logon to the Inmarsat computer system. The pilot may not have known exactly what data is recorded, but if he was as smart as I think he probably was, he only had to realize that the SATCOM was a stealth vulnerability during the final flight leg after Arc7.
So it’s a little bit of a different paradigm, we all started out thinking 1825 SATCOM reboot was a “clueless” pilot who had no knowledge whatsoever of leaving SATCOM trail. But that could well be yet another wrong assumption.
@All
There is no guesswork required. We have some evidence-
The simulator case is evidence of a flight plan to the 30-32s area of Arc7. That is very consistent with BTO/BFO, if we relax the great circle, straight-edge maneuver-less flight path to Arc7 (and we should be relaxing that constraint). It is consistent with Blaine/Chari’s actual debris find based on drift calcs.
Rather we are saying opinion (of a less nefarious scenario) should take higher priority importance that actual debris finds consistent with the sim data and flight data.
@Gregg
There was at least one video on ATSB YouTube channel that showed the map of the search area (the box) and how the probability changed with time as the search progressed. As parts of the search box were cleared and those results incorporated into the probability map, probability rose in other parts of the box. Unfortunately the video(s) is not available anymore, but maybe some contributors of this blog have saved it.
@Marijan
Did you mean this one ?
https://mediaspace.esri.com/media/t/1_vijfao60/244789952
Or on YouTube:
https://youtu.be/dOcac4S_osQ
@Marijan
In reply to your Nov 26th post, 6:22 am
In the same ATSB report p95-97 the data gaps and LPD areas are described in percentages, and shown in fig. 73. The area which imo is among the most relevant (S35.0 – S36.0) has the highest LDP (7.8%). However, on average the LDP areas would still have around 70% confidence of detection, which means that even in this area the probability for a false negative is just a few percent.
I’m sure the ATSB carefully considered what a debris field could look like. It would nevertheless be interesting to have some more details from this historic study you referred to. I’m for example wondering if larger debris fields are usually easier to detect, or if this is a much more nuanced story.
@370Location
Let me offer a respectful correction of several statements you have recently posted regarding the position of the IG.
As an original member of the IG, my recollection is that the IG, per se, has never updated its consensus estimate of the last location of the aircraft at the time of the 7th arc login. It is close to the posts I have made over the years updating my study as more information has been gleaned from the available data and public documents. I have no inside sources.
The location is somewhere near E88.93 S37.68.
But this does not address the path thereafter. If you look back at the discussions on Duncan’s blog, you will find that the glide hypothesis was a contentious matter and a little sub-group of Barry, Geoff and I raised the issue numerous times.
But, consistent with the official “ghost flight” scenario this was beaten back as low probability, even as it was eventually accepted as technically possible with a piloted scenario.
The original IG position is entirely consistent with what we know and ostly agree is a plausible scenario.
1. The initial diversion and flight using waypoints to at least the final radar contact just prior to the first ping ring.
2. A high probability of a turn south thereafter (either manual or programmed) Our study suggested that the is within about 1 minute of 18:40 UT and that under the following conditions:
Fuel Management:LRC
Speed M=0.84
FL The higher the better
to be continued in a moment,,,,twitchy finger syndrome
Navigation Mode: LNAV or Constant True Track (difficult to separate since the path is essentially meridional). But since our understanding of the LNAV mode is that it would revert to a panel setting which probably was magnetic heading and a poor solution, suggests that either the azimuth after the FMT was set to 186T CTT at the turn OR at least prior to overflying ISBIX.
The question remains: “What did the pilot do between the FMT and the 7th arc?”
There are two options he was or he wasn’t alive.
The only statistical method used in the above studies was to minimize the least squares error between the measured and expected BTO and BTO along a numerically computed path that at least included the FMT.
End end of rant.
Taking note of likely search scenario, and absent any new information, I support extending the search to the south and including the IG original recommendation as well as to somewhere south of 39.5.
If there is any further search, a glide from the UGIB and the IG “hot spot” areas rather than swath areas is likely to be more productive.
@ventus45
Thank you very much for sharing this video Ventus, it is not the one I was referring to, but it is really informative and must watch! I will get back to it later.
There was a video from 2017, probably just 15-20 seconds long, without sound, that showed PDF and its evolution over time. It was actually mentioned on p.81 of the final report “Operational Search for MH370”:
“The effects of the completed sonar search on the PDF throughout the search can be displayed with a time sequenced animation, evolving trajectory probabilities and is available for viewing on the ATSB YouTube Channel.”
There were also a couple of other short videos related to the search for Malaysia Airlines 370 which are also referenced in the final report, and which are unfortunately also unavailable.
@Sid. My theory bears many similarities to the original IG model hypothesis, with some important differences. It provides – in my mind – a very strong theoretical case to justify searching 39.5S-40.0S.
The methodological approach was as follows:-
1. Try different turn times off N571 to obtain best fit. Instead of prioritising the combined optimum for BTO+BFO, I optimise for BTO and focus on LNAV paths, not CTT.
2. Generalise these simple path models to demonstrate limited sensitivity to a wide range of “priors” at Arc 1 (timing error, speed, track angle and offset from N571).
3. Use the optimum-fit solution to search for corresponding waypoints. This is how I arrive at IGEBO-RUNUT as best match for the southerly trajectory. In the absence of a viable single-waypoint FMT, demonstrate that MEKAR-SANOB-IGEBO is the next best alternative.
4. Confirm that a MEKAR-SANOB prior “predicts” turn timing and path solution that runs very close to IGEBO and RUNUT. Now “snap to route” and demonstrate the BTO-compliance of the full route – with a final waypoint hypothesised as 40S85E. Determine that the optimum speed/altitude combo for this path model is M0.84, FL360.
5. Demonstrate that the predicted 7th arc of this path lies at 39.6S – and that this prediction holds even if you remove the final waypoint assumption and run the models from RUNUT or even from IGEBO.
So that was my method. It is guided by BTO optimisation rather than a preconceived terminus or route preference. It produces excellent BTO fit, predicts plausible speed and altitude, predicts that the route employed the first airway available (P627) to turn left after Sumatra. Turn timing matches the window indicated by BFO (between 1828 and 1840). No unusual manoeuvres are needed. In fact no active pilot is “required” beyond the 1825 power restoration to fly this route. The predicted terminus at 39.6 lies on the only entirely unsearched segment of 7th arc within DSTG’s original search area.
My BTO-only peak probability zone matches that documented by DSTG – but rejected in favour of their final pdf that optimised for BTO and BFO combined. Unlike DSTG – who looked in vain for a waypoint route to match their solution, the BTO-optimised route predicts a [highly plausible] waypoint-navigable route that sits close to peak probability.
This is the same hypothesis that I have been banging on about for the last 3 years.
Predictions that follow from the theory are:-
1) The debris field will be found close to 7th arc at +/- 39.6S
2) The BFO data was distorted, most plausibly by cold-soaked SDU
3) The apparent fuel deficit will prove to be explainable by lower power offtake from bleed air, low electrical load, and a small amount of trapped usable fuel.
4) Debris drift from this terminus will be shown to be feasible once higher wind coefficient and directional offset to wind are considered (both of which are plausible).
In case there is classified radar data out there that can’t/wont’ be released, a confirmation of position at 1825-1830 consistent with MEKAR-SANOB (rather than continuation along N571) would represent especially strong supporting evidence for this hypothesis.
@ventus45:
It would be great if the unmapped sections of the 7th Arc were added to Seabed2030 over time, but I’m hoping MH370 will be found before we’d ever see that. Backscatter might show strong responses from large flat surfaces like the wings, but isn’t added to the depth charts I’ve seen. BTW, the vid you linked was an excellent presentation.
@Sid Bennett:
As I replied to Victor, I meant only to point out that the BTO 7th Arc is the best evidence we have. I should not have tried to characterize what IG or UGIB has written.
You say it’s black and white that either the pilot was alive or not, and that the statistical analysis was for “BTO along a numerically computed path”.
That path was my point. The unpiloted numerical paths persist even though expanding around it now implies a piloted glide at the end. More important is considering that in a piloted scenario, there were likely several more turns. The pilot appears to have been flying waypoints until leaving RADAR coverage. The turns incorporated into numerical-only paths seem to be for delaying the progress of high speed flight between arcs.
My summary was that it’s pragmatic to search the unmapped areas near the SIO 7th Arc since there is little extra cost en route.
@Sid & All,
Yes, my interpretation of the hard evidence for a crash in tropical waters does upend most of the realistic scenarios presented on various forums and media since Day 1. Still, it is a good fit for all the factual data, even if some experts prefer to call it a red herring without looking at the underpinning data.
I’ve done acoustic and other analysis to help sort out various theories presented here. Getting to the truth matters more to me than being right.
It seems clear to me that the flaperon barnacles began growth after beaching, because that’s when their environment cooled enough to allow it. I have yet to see any other conclusion to be drawn from looking at this photo of the leading edge of the flaperon:
https://370location.org/wp-content/uploads/2024/03/NPR-inboard-hinge-rot-crop-closeup.jpg
It’s not that the NPR photo is fuzzy or low resolution. We can see that the screws on that edge are worn to the nub, layers of fiberglass have been worn down, and fabric seal corners are tattered down to the underlying nylon mesh and soft foam. (This is not normal wear and tear prior to the crash, as suggested to me privately). It is most likely from grinding against a coral reef or hard sand in the surf. Yet, barnacles attached and then grew on top of that damage. Sure, some are missing as it beached again where it was found, but it doesn’t change the timeline.
Perhaps a French citizen could file a request for better flaperon barnacle images, or ask that officials there look into how this could change their prior conclusions.
I suspect that Dr Chari might also have some thoughts on how this would influence the upcoming search.
I’m not being dismissive of the many theories proposed so far, and don’t mean to step on any toes. I’m open to other explanations for what’s seen in that flaperon image.
@Paul Smithson,
Thank you for taking the time to respond to my previous comments and criticisms regarding MH370 routes to circa 39.5S. You are persistent on this topic, which is a good thing. That’s the way science is supposed to work. I also appreciate that we have been exchanging data and calculations on the side with the goal of improving our understanding of (1) what fuel is required to achieve a flight to that location, and (2) to determine if that is possible with 9M-MRO.
I have been working to address both issues. I am in the process of developing a new fuel model specifically for this purpose. So far, my comparisons with the previous UGIB fuel model are quite consistent. I have also re-examined the MRC fuel flows which I estimated many years ago. It seems possible now to improve the accuracy of fuel flow prediction for speeds near LRC (where we have a Boeing fuel flow table), including MRC, ECON with CI = 52, Mach 0.85, and Mach 0.87 (MMO). We have no Boeing fuel flow or Mach tables for those speed settings, but it now seems possible to empirically generate them based on the Boeing tables we do have. I plan to create new fuel flow tables for these speeds and to incorporate them into my new fuel model. I will then exercise this model for families of routes and determine the fuel savings (the “parasitic fuel flow reduction” percentage, or PFFR) needed to match the known MEFE time. That will provide a set of fuel savings values versus crash latitude for the LNAV and CTT routes from Arcs 2-6. Then one can see how much extra fuel savings are needed at the far south latitudes versus UGIB at 34.2S. By the way, the PFFR is 2.4% for the UGIB route. Using 1.56% for the air packs being off at FL30 still yielded a small fuel shortfall of about 0.8% in UGIB, and you need 2.4% to exactly match MEFE. Achieving 2.4% is not a problem, because we did not allow for reduced electrical loads in UGIB (compared to what is nominally assumed by Boeing in their fuel flow tables). Electrical load reductions are possible by turning off the exterior lights (nav, strobe, and logo), the primary cabin overhead lights, the galleys, and air recirculation fans. These can be shut off by switches on the flight deck. Possibly 60 kg/hour can be saved by these pilot actions, which is about 1 % in fuel savings. Thus, PFFR can be as large as 1.56% + 1.0% = 2.7%. Therefore, reasonable electrical load reductions during the SIO Route can explain the UGIB shortfall and allow some longer-range routes.
The question is, where is the fuel limit on Arc 7? Previously, based on routes suggested by others, that fuel limit was moved from 36S to 36.5S. Adding another percent of potentially available fuel will move it slightly past 36.5S. I think my new fuel model will provide a convenient and reliable answer, especially for the difference in PFFRs for different routes . Basically, I will be able to solve for the fuel savings needed as a function of Arc 7 latitude, for routes which are known independently to (1) be consistent with ALL satellite and radar data from 18:02 to 00:11 and (b) be reachable from the 18:29 estimated location (per UGIB this is at the end of the right lateral offset from N571). Then one can study the fuel savings opportunities in more detail to evaluate the likelihood the necessary fuel savings could be achieved by pilot actions. By being “consistent with the satellite data”, I mean satisfactorily matching both the BTOs and BFOs from 18:25 to 00:11. It is not useful now, in my opinion, to look at BTO-only or BFO-only fits of the SIO Route. They are misleading at best. The SIO Route has to match both simultaneously.
Let me make a few additional comments:
1. My condition that all possible routes must have a 19:41 location that is reachable from the 18:29 position North of N571 (at 7.179N, 95.684E from UGIB Figure 3) is based on the fact that the route shown in UGIB between 18:02 and 18:29 is, as far as I know, the ONLY proposed path that is consistent with the satellite and radar data we have from 18:22-18:28. One should not ignore the 18:22:12 radar position nor the 18:25 – 18:28 satellite data . Nor can one just match the BTO at Arc 1 and ignore the BFOs then, if you want to be credible. Now, if someone can demonstrate a different path after 18:02 that matches the 18:22 radar data and the 18:25-18:28 satellite data, then please prove that route before proposing a subsequent SIO Route whose 19:41 location is not reachable from the (assumed of fitted) 18:29 position. You should not ignore that data by allowing turns southward from N571 prior to 18:29 which are incompatible with the satellite and radar data from 18:22-18:28. There is no technical reason not to accept that data at face value and ignoring it could be perceived as an unscientific, willful bias toward routes which might be impossible to fly. My new flight/fuel model starts at 17:26 and runs until 00:19. It also computes the en route delay time needed to match the 19:41 position at 19:41. If that delay time is negative, then the route is not flyable.
2. Some or even most of the differences in probabilities we have been getting for 39.5S routes may be caused by us using different fits with different residual errors. I am in the process of reviewing my previous fits and the ones you have kindly provided so that we can agree, as much as possible, on which routes are optimal for each crash latitude of interest and are acceptable, except for fuel considerations. It is certainly possible to generate BTO-only fits with higher BTO probabilities than fits which optimize the product of the BTO and BFO probabilities. But those BTO-only fits are not worth studying from a fuel perspective, because the True SIO Route must satisfy both BTO and BFO constraints. Then I can study the fuel savings needed for those particular routes, and we won’t be disagreeing about the BTO and BFO probabilities because of us using slightly different routes.
Paul, I appreciate your willingness to share the details of your work. That is the only way we can resolve the differences in our results. Hopefully we can agree on a set of SIO Routes over a wide range of crash latitudes which we can then study in more detail regarding fuel adequacy.
@Niels
Sorry for my late reply. I took the screenshot from the video (https://youtu.be/dOcac4S_osQ) that ventus45 shared and added the map from the final report beside it. It is uploaded here:
https://drive.google.com/file/d/19va-YhnOukNbN_YrWK8ODzLwTiIyhZyo/view?usp=sharing
Both images show the locations of the data gaps and low probability areas (LPDs) which are 100mx100m in size or larger. In addition, the image from the screenshot contains not only percentages relative to the total area but also the absolute numbers.
Like you said, the area with the highest concentration of data gaps and LPDs is the Orange area, 35-36°S along the 7th arc. One more area of particular interest (even though it has the best coverage) and with a substantial number of LPDs is the Geelvinck fracture zone, which falls in the Red area.
In my post from Nov 18th I wanted to show there are still relatively large data gaps and LPDs that fall in the Yellow area and that are at least a kilometer in size. They could easily contain the main debris field. I doubt that data holidays of this size could be found in the Red area, but probably they exist in the Orange Area. Of course it is hard to say anything more without seeing the more detailed maps.
Even though ATSB and Fugro did their best to cover as much area as possible, I actually never heard that ATSB explicitly dismissed the possibility that the wreckage was missed during their search. Andy Sherrill was quoted above by Victor: “Having said that, there is always a chance it [a tagged contact] could be from MH370, but based on our assessment the time it took to investigate each of these small contacts was not worth taking vs searching new areas.” I believe that this was the logical approach during the search, but now when almost everything along the 7th arc has been searched, save for the last ATSB recommended area 39.6-40.2°S, it is the right time to review those data and search some of the locations that have been skipped earlier.
@Marijan
You Said:
“One more area of particular interest (even though it has the best coverage) and with a substantial number of LPDs is the Geelvinck fracture zone, which falls in the Red area.”
I agree that the GFZ (Geelvinck Fracture Zone) does have a substantial number of LPDs, but I disagree where you say that the GFZ has the best coverage.
I contend that the GFZ hardly has adequate coverage, let alone best coverage, because the GFZ was only ‘transversely scanned’ by the ‘long line deep tow system’, which could not (in my view) provide adequate ‘terrain following’ (diving into the trench and then climbing out of it on the other side), and that fact alone is probably a very large part of the reason why there is a substantial number of LPDs in the GFZ.
In view of the fact that Paul Smithson has made a good case for the 39S-40S region, I think that the entire GFZ should be re-scanned axially, in its entirety, by AUV’s programmed to run along the trench axis, and all the way up the steep slopes on both sides, at least to the top of the escarpments.
@DrB
I am heartened to see that you are revising / updating / enhancing your fuel model. I (and many others) await its publication. Any estimate of when ?
@ventus45
I agree with you, GFZ area has a lot of LPDs which can be seen from the maps linked above and needs to be reviewed and potentially rescanned. That, on top of the fact that it is close to the ATSB hotspot, is the reason why I added GFZ to the Orange area as an area of interest. Coverage was indeed excellent, and Fugro launched a lot of AUV operations there. However, the confidence in search results is not high due to complex geology, and that is why in the region of GFZ we have LPDs, not data gaps. It is interesting to mention that prior to AUV operations, scanning with SSS towfish had to be performed transversely to the GFZ, as you said, due to the direction of ocean currents in the area.
@ventus45,
My goal in my new fuel model is to assure a relative error of only 1-2%, and hopefully close to 1%, for the difference in fuel needed for various Arc 7 crash latitudes. Overall, in an absolute sense, the MEFE time is predicted by the UGIB fuel model within +/- 4 minutes at one sigma. Nothing has transpired since 2020 that would allow improving this accuracy. In fact, most of it is simply the calibration uncertainties in the fuel flow gauges and in the fuel quantity sensors in the B777 aircraft. So, there is nothing that can be done now to improve that.
What can be done, and what is being done currently, is to implement a fuel flow predictor which covers the full range of Machs between Holding speed and M0.84, for every combination of flight level and aircraft weight in the Boeing tables for LRC and Holding. Victor suggested a method for doing this some years ago, and we are now in the process of trying it out. Initial results are encouraging, but it is too soon to know if this approach will be successful over the entire ranges of altitude, aircraft weight, and Mach.
If things went smoothly, which they never seem to do, I might have some reliable results in January. There is still a lot of work to do to confirm the accuracy of the predictions and potentially a huge amount of time needed to perform the hundreds of cubic fits required if they must be done manually. If this fitting process can be automated, that could speed things up a bit.
I am trying first to produce more accurate fuel flow and Mach tables for MRC. I should also be able to produce the tables for ECON with Cost Index = 52, which was the choice in the MH370 flight plan and which was in use at the time of diversion. In principle, if our approach is successful, we will be able to to accurately predict the Mach and Fuel flow for ECON with any Cost Index, and any constant Mach from Holding Mach up to about M0.85.
@Paul Smithson
G’day Paul,
You’ve certainly made a solid, cogently coherent argument. A testament to your quiet (and sometimes not so quiet) persistence.
While what Victor said about your needing edge of the envelope outcomes on a range of factors is quite correct, that is not a compelling case for dismissing your conclusions. A pretty good way of describing many aviation accidents is just that, a combination of edge of the envelope outcomes, two or more events that were generally thought to be highly improbable coalescing into something entirely unexpected.
As we’ve discussed separately, “Roy” probably presents as a bit of an issue for such a southern terminus, but then “Roy” seems to be another one of those outlier outcomes anyway.
All that said, convincing any of the alternate theorists (be they “X Marks the Spot”-ers or adherents of the extraterrestrial) and their coterie of followers about anything regards MH370 is a Herculean, if not impossible task. It is very much the case that so many are so rusted on to their own theory, regardless of how bizzare, I don’t think that any of them are for turning.
For mine, you’ve made a more compelling case than many doing the rounds recently. The problem with your terminus is that it’s not exactly in the neighbourhood for the proposed OI search box.
If I was planning the search campaign, I’d start with yours and work north-east along the arc and wider out to at least ±60 nm. That should cater for the eventually worsening sea conditions, and provide as much time as possible for search activities. I don’t think that’s the way it’s going to play out though.
Separately, the recent discussions on your theory, in comparison to say UGIB, highlights once again what is often forgotten or taken for granted, viz the number of assumptions, essentially “best guesses”, that have to be made when trying to model a theorised route and terminus; navigation mode (both lateral and vertical), cost index, air pack status, piloted/unpiloted, the nature of the IGARI turn-back and transit of the Malay Peninsula with respect to fuel remaining, etc.
It would be interesting to see a comparison of the key assumptions plotted by potential impact on outcome versus certainty/likelihood that the assumption is correct. The cumulative impact of the assumptions would almost invariably be huge.
@DrB
A couple of reflections on your post of 30 November.
Your statement that “Electrical load reductions are possible by turning off…Thus, PFFR can be as large as 1.56% + 1.0% = 2.7%” is, of course, not news to me. I proposed and quantified potential electrical load-related fuel savings with a midpoint estimate of 0.7% (0.56%-0.84%) in July – see your inbox. More recently (17 November, above) I documented much larger possible fuel savings (3%+) attributable to parasitic bleed air/packs off than you had previously estimated (1.56%). Together with roughly two-minutes worth of trapped usable fuel – these factors were large enough to bridge the fuel deficit vs. model for my proposed path hypothesis. I am hopeful that our conclusions regarding putative savings will ultimately converge.
You said: “One should not ignore the 18:22:12 radar position nor the 18:25 – 18:28 satellite data . Nor can one just match the BTO at Arc 1 and ignore the BFOs then, if you want to be credible.” For reasons already explained, I choose to model from Arc 1 rather than from a suspiciously singular last radar position of unknown precision. My path models start from a known arc of positions, with known confidence limits and rather small uncertainty of latitude (unlike models starting from selected latitudes at Arc 2).
As regards BFO, I state explicitly where there is a deviation between BFO predicted vs observed. That seems to me a more transparent approach than inserting implausible ROC at 1840 and 1941 to reconcile the path hypothesis with incompatible BFO observations. My path model results predict quite large and trended BFO residuals. I don’t count that as “supportive” of my hypothesis, but I have proposed a bias shift based on documented temperature coefficient and the possibility [probability?] that a significant change in cabin temperature occurred. That isn’t “unscientific, wilful bias”, it is using the scientific method to examine the predictions that follow from a hypothesis – then exploring whether anomalous results are explicable (and empirically testable).
I accept that you have chosen to place a high degree of certainty on your derived 1829 position based on the assumptions that you have baked in there. My assumptions don’t match yours – so the geographic limits of your fuel model will not (quite) match mine, even if we ultimately agree on FF and endurance after adjusting for “savings” vs model.
I was trying to reconcile Paul’s southern location with fuel consumption and the amount of fuel onboard, so I went back to ATSB routes and their solutions. Some things are not clear and I have some rather basic questions.
1. Does reaching 39.6°S or 40°S require an very early turn south and flying across the northern tip of Sumatra? By early I mean 18:22 or even 18:02 if the last radar point is considered not reliable. Which one of those two?
2. In order to “match” the airplane speed and heading observed with primary radar data up to 18:02 (or 18:22) with the 18:25 and 18:28 BTOs, then it would be required that the airplane:
1) significantly slowed down
2) turned north
3) turned south
If the plane slowed down then how much would it need to slow down? It was flying close to 500kts. Turning north is mostly clear and it was explained in UGIB 2020. If it turned south, can the OCXO drift at 18:25 and 18:28 be high enough to “hide” the frequency shift due to plane motion south (so it appears it was turning north)?
I am aiming here for the possibility of the plane turning south across Sumatra. If, in some hypothetical situation, Malaysia did not have primary radars, probably no one would think about the possibility that the plane turned back, crossed the Malay Peninsula, and went into the Indian Ocean. The Pacific Ocean would be searched instead. Here, I am thinking that something similarly unexpected happened, such as deliberately crossing Sumatra, for whatever reason.
@Marijan
I have never accepted the Lido Slide nor the 18:22 radar hit.
My current Sumatra over flight scenario begins with a turn near Pulau Perak circa 18:03 towards WITC, going feet wet to meet the first arc just offshore, thence proceeding to the FMT at MABIX circa 18:38, thence direct 3789S, thence IGPOL.
Dear Victor,
Official statement on the resumption of the search for MH370 from the Ministry of Transport Malaysia.
The deep-sea search for the missing wreckage of Malaysia Airlines Flight MH370 is set to resume in the Southern Indian Ocean on December 30, the Ministry of Transport said today.
In a statement, the Ministry of Transport said Ocean Infinity has confirmed it will recommence seabed search operations for a total of 55 days, conducted intermittently.
The search will focus on areas assessed to have the highest probability of locating the aircraft, in line with the service agreement signed between the Government of Malaysia and Ocean Infinity on March 25, 2025.
“The latest development underscores the Government of Malaysia’s commitment to providing closure to the families affected by this tragedy,” the ministry said.
Flight MH370, carrying 227 passengers and 12 crew members, vanished on March 8, 2014, after departing from Kuala Lumpur International Airport (KLIA) en route to Beijing.
Despite the largest search effort in aviation history, which covered over 46,000 square miles of the southern Indian Ocean, only a few fragments of the aircraft have been found, scattered on beaches thousands of miles apart.
According to Mayalysia’s MOT, the search for MH370 will resume on 30 December 2025:
“The Government of Malaysia wishes to update that the deep-sea search for missing wreckage of Malaysia Airlines Flight MH370 will be resuming on 30 December 2025.
Ocean Infinity has confirmed with the Government of Malaysia that it will recommence seabed search operations for a total of 55 days, to be conducted intermittently. The search will be carried out in targeted area
assessed to have the highest probability of locating the aircraft, in accordance with the service agreement entered between the Government of Malaysia and Ocean Infinity on 25 March 2025.”
MEDIA RELEASE – MINISTRY OF TRANSPORT MALAYSIA
RESUMPTION OF MH370 SEARCH BY OCEAN INFINITY IN THE SOUTHERN INDIAN OCEAN
Thank everyone that responded to my previous question. That was an interesting/informative video.
Non-technical question(s):
I am assuming all the high priority areas have been searched.
What if OI does not find the aircraft on this upcoming search? How long is their contract with Malaysia? 2026? How many more millions will OI be willing to spend without more definite information as to the location? Would they even continue the search?
These questions have me doubting the plane will ever be found.
@Paul Smithson
I will keep my comments short as I am traveling.
About a month ago my attention was called to the bias corrections estimate used by the “official” studies. I rely on none of them as they are sketchy and provide no reasonable ability to reproduce the results.
What I call your attention to is that a bias correction of about 12Hz was used for the BFO correction for the ping during the eclipse period. Of course if you maintain that the frequency translations in the satellite have been corrected for temperature variations of the satellite oscillator this cannot be! Alternatively if you maintain that there was a temperature-dependent effect of this oscillator at the time of the eclipse, then there would necessarily be such an effect at other times when the temperature is different.
The Inmarsat data can be analyzed, as I have done and reported, and show a very, very high correlation with the measured satellite temperature and the translation oscillator frequency, with a variation closely matching the eclipse correction. It follows that the oscillator temperature and frequency is known for the 24 hour period of the incident, except for an unknown bias offset.
I maintain that the bias error is a known function of time at all times and predominantly due to the satellite oscillator temperature variation. Also, the diurnal change in the eclipse time (or non-eclipse) means that the day-to-day changes in bias offset are complicating factors in understanding data from other flights.
The other factor is that the Doppler shift on the satellite-to- aircraft link is different from that on the satellite-to-ground link since they are on different paths and frequencies.
In short, except for the re-boot transients when the on-board satcom has been reinitialized, that oscillator is not the source of BFO error.
@Sid: You conclude “I maintain that the bias error is a known function of time at all times and predominantly due to the satellite oscillator temperature variation.” What evidence have you seen that makes you think the transponder local oscillator temperature varies at all when the spacecraft is in full sun (not in eclipse)?
@Andrew:
Thank you for your post of December 2, 2025 at 11:48 pm informing us of the media release from Malaysia’s MOT.
“seabed search operations for a total of 55 days” noted.
@ventus45
Thanks Ventus, that is a very early turn and I guess there will be a plenty of fuel. What would be the groundspeed? I cannot find IGPOL (IGPOL 41°53’48.00″S 85° 0’0.00″E) waypoint on Skyvector, is it retired?
@Marijan
Most SIO waypoints were removed from Skyvector about 2 years ago.
From turn at 18:02 to Mabix at 18:38 = 36 minutes for 274NM = av457knGS
@Sid.
We had a discussion on the same topic a few weeks ago. The EAFC strips out any distortion of bias at the satellite oscillator and I don’t think there’s any room for doubt about this. So if there is any systematic departure from assumed bias it has to be something at the aircraft level, not satellite.
@Paul Smithson: Table 9 of the JON paper has a column that presents the sum of the satellite drift and EAFC correction. If those values are used, any effects of temperature-induced drift in satellite oscillator frequency are implicitly included.
@airsealandman @VictorI
The fact that Table 9 exists establishes that the temperature related bias offset exists.
It was an early analysis that certainly served to rationalize the
BFO error at the time of the eclipse. It is hardly definitive for lower value offsets apparently due to diurnal variations. These are of the order of Hz and clearly influence the path analysis.
It would surely be instructive to do the same thing with data for other days, which no doubt exists.
I published to this blog some years ago, if memory serves, an analysis of the Inmarsat paper data that I digitized from the figures presented and determined the temperature coefficient of bias offset and fitted the residual with a diurnal (sinusoidal) correction and a slight secular drift.
I will see if I can locate a copy and re-post it.
@sid Bennett: Table 9 properly accounts for any frequency shift due to satellite temperature changes. That means if the values in Table 9 are used, the temperature effect has been properly included in the BFO analysis, whether or not there is a temperature effect outside of the eclipse.
We’ve had this discussion many times now. There is no need to relive it.
@Marijan. Delayed reply to your earlier post.
1. “Does reaching 39.6°S or 40°S require an very early turn south and flying across the northern tip of Sumatra?”
No. A single turn made from N571 with turn timing optimised for BTO fit takes you to 39.0-39.5S. Add a little uncertainty for sub-optimal path fit and end-of-flight trajectory stretches the endpoint to 40.0S. However, turn no version of this “single turn” solution can be matched to a waypoint. If the turn was navigated by waypoint, then the “least-bad” waypoint-navigable turn that matches predicted turn and southbound path goes: MEKAR-SANOB-IGEBO-southward. Arc 1 crossing is made at track 276 degrees (rather than 296) and that generates a BFO error of -15Hz. You also encounter BFO errors at 1840 and 1941 (so does UGIB, but these are “corrected” by supposed descent/climb that happen to coincide with those datapoints).
2. “In order to “match” the airplane speed and heading observed with primary radar data up to 18:02 (or 18:22) with the 18:25 and 18:28 BTOs, then…”
I agree that a manoeuvre (north) is required IF both the 182212 and Arc 1 positions are correct, even allowing for some imprecision on each. The other alternative is that 182212 is wrong or that 18:28:15 is wrong. I don’t find any of these explanations entirely satisfactory and choose to start my modelling at Arc 1 rather than speculate on specific manoeuvres to reconcile uncertain data pre-arc.
3. “Can the bias drift that much?”
In theory, yes. Compliance tests of comparable equipment showed a temperature coefficient 0.4 -0.5Hz per degree centigrade of ambient temperature. A change in temp from ~25C to somewhere in the 0C to -10C range is enough to produce a shift of 15Hz.
4. “I am aiming here for the possibility of the plane turning south across Sumatra.”
You don’t need to cross Sumatra to end up at 39.5S-40.0S though you do need EITHER a turn immediately after crossing Arc 1 or a multi-stage turn that cuts some time & distance off the corner of a single-turn from N571.
@Paul Smithson
Thank you very much Paul, this explains a lot.
@Gregg
Unfortunately, I don’t think anyone here is able to answer those questions.
Most recent update on Armada 86 05 is that it’s due to depart Legazpi, Luzon Island/PH, on Tuesday 9th December with next port indicated to be an Australian facility.
It’s not clear if the next port call will be made while enroute to search area or after positioning directly to the search area and a period of AUV deployments in the search area.
While there has been no further update on the specific area where the AUVs will be deployed, I expect A86-05’s activity to focus on the areas previously defined (see earlier post at this blog).
@Marijan
Sorry for replying late; I’m pretty busy writing my concept report (I hope to share it soon).
Regarding your Nov. 30th post:
I would be in favour of filling in the data gaps in the area “near” the arc as a priority, also indeed to not leave open ends. The maps you shared show that you could define focus areas where apparently the terrain resulted in a significant fraction of data gaps.
My results indicate that the S33.0-S36.0 area should be pretty ok, with the southern half being more relevant.
@Gregg
The details of the contract between Ocean Infinity and Malaysia have not been made public, but we know there was reference to it being an 18-month contract. Conceivably OI could request extension. It could be possible that OI sees non-monetary benefits in searching. But to me, finding MH370 would look like open offer for $70-million reward to anyone wants to search.
@Niels
Your previous answer motivated me look in more detail “Color zones” from the ATSB map and what they actually mean. So far, it seems that the quality of obtained sonar data is much better in the “Red zone” (Indicative Search Area, from 36.1°S to 39.5°S, 27.5NM NW and 25NM SE from the 7th arc) compared to other zones.
I would be free to say that considering other evidence (BTO, fuel amount, surface drift…) and operational requirements at present moment, it seems that filling the gaps south to 36°S would be the best and first (but not only) place for potential extension of the current search area. I will post more about that later.
@Andrew
TWA800/MH370:
Question for you, if we look at MH370 B777 flight, the fuel tank loading is somewhat similar to TWA800 with the center fuel tank empty. Does that imply blanketing or other measures were taken to inert the center tank of MH370 and other B777’s? Does that pose a limitation on moving fuel around during flight?
B777 was probably designed before TWA800/1996, so it makes sense that the B777 fuel tank design strategy is similar to B747. Interesting to compare to Air India 787, which probably has a post-TWA800 fuel tank design, with a whole different strategy with more focus on keeping more fuel in a larger middle tank.
I attempted to find a post of my BFO study results to this blog but could not, so I do so now as links to a research note and to the digitized data from the Inmarsat paper.
I am also tired of discussing the subject.
All I will say is that looking at the data in terms of a simple physical model fit suggests that:
1. The BFO offset can be predicted for any time when the temperature of the satellite translation oscillator is known.
2. In addition to any diurnal effect (which I am sure slowly varies with time…in days), there are second order temperature-related effects that, if not modeled, cause the analysis of other flights ( at other days or times ) to exhibit a higher variability than could be achieved if the system model accounted for the oscillator temperature variation.
3. The BFO is a valid parameter for path modeling and cannot be ignored.
I dislike having to use a single example to make such generalizations, but the available data leaves me no choice.
https://www.dropbox.com/scl/fi/p83zc7twao7wbeobezich/research-note-feb20.pdf?rlkey=antji8yn1a10ef7f99g7u00p3&st=z89114uz&dl=0
https://www.dropbox.com/scl/fi/1sr22zlfduqbo0v4rm0ob/pilot-freq-analysis012920-r1.xlsx?rlkey=wcitzmgjmr2dm7oqh4ox79ahb&st=stxcukit&dl=0
@Sid Bennett: Why do you believe that using the values in Table 9 for the combined EAFC and satellite drift are not adequate for the MH370 flight we are modeling? The same would be true for other flights that Inmarsat/DSTG modeled as long as those values (EAFC + satellite drift) are properly calculated by subtracting out the other contributions to the pilot BFO (uplink (C-band) and downlink (L-band) Doppler shifts) are known, which they are. Whatever other effects you cite (diurnal, second-order temperature) are already included. This is what Inmarsat is saying in the paper.
@VictorI
My concern is the rather high BFO error range used by DSTG in assessing the BFO significance. I believe that the EAFC corrections are automatically done and use a simple model, which may lead to the larger spread for the other flights analyzed. The raw data for those flights do not appear to be public.
What my analysis shows is the underlying relationship of the satellite temperature on a portion of BFO time series and that it is a continuous function where the temperature is a significant non-random factor.
Based on my previous technical work I am somewhat focused on using analytic models to uncover deterministic factors that are often lumped as statistical inputs to a model but are in fact dependent on deterministic mechanisms.
Both you and I are confident that the BFO error is a valid discriminant in the path studies. I took pains to mention that I did not use the Inmarsat table to anticipate questions regarding the BFO adjustment that I made in Barry’s parameter set.
@Sid Bennett: We have no reason to believe that the DSTG didn’t correctly calculate the BFO for all test flights by estimating the EAFC + satellite drift contribution the same way that Inmarsat proposes in the JON paper, i.e., query the archived BFO for the pilot signal at the time of interest and for the satellite of interest and subtract the contributions of uplink and downlink Doppler based on the orbital dynamics.
@VictorI
Then why did DSTG need to propose three different BFO models in the appendix?
But each of them was sort of right….
Again, I am only concerned about the error bounds they chose…really conservative.
@TBill
You asked: ”Does that imply blanketing or other measures were taken to inert the center tank of MH370 and other B777’s?”
A nitrogen gas inerting system was introduced on the B777-300ER and later variants to reduce the flammability of the centre wing tank fuel. The B777-200ER does not have such a system; however, a number of airworthiness directives and other changes were issued after the TWA 800 investigation, to prevent potential ignition sources inside fuel tanks. On the B777, those changes include an auto shutoff feature for the centre override/jettison fuel pumps to prevent dry running, and revised logic for the EICAS messages that direct flight crew to select the fuel pumps ON or OFF.
”Does that pose a limitation on moving fuel around during flight?”
Flight crew don’t “move fuel around during flight”. If there’s a significant imbalance between the left and right wing tanks they would open the cross feed valves and select the boost pumps on the low side off, so that both engines feed from the high tank.
@Sid Bennett: Are you referring to a historical document? In the DSTG Bayesian report where changes in the BFO bias were determined, I am aware of only one BFO model.
@Andrew
Thank you for above info.
@VictorI
Appendix C: Burst Frequency Offset (BFO) — Doppler Model Development” in The Operational Search for MH370.
@Sid Bennett: What is described in Appendix C of the ATSB document is the historical evolution that led to the final BFO model that was used in the JON paper and the DSTG Bayesian analysis. The final version of the BFO model is exactly what I described, i.e., the satellite drift and EAFC correction terms are known using the BFO of the pilot signal and subtracting the known uplink and downlink Doppler shifts. That’s what we should be using in reconstructing MH370 paths and not worry about separating the EAFC compensation from the satellite frequency shifts.
@Sid: Victor is correct about the BFO compensation.
@airsealandman
He is I am sure correct as the results are similar to mine 🙂
But please remember that prior to the publication of the ATSB the history of the BFO correction was only the Inmarsat paper and there were several apparently unresolved issues. I only had access to the ATSB report in pre-print on 12/6/2015 and by that time I had completed by own analysis and used it extensively in searching paths. So it is only natural to continue to use it. My purpose in posting the data is only to serve as a supplement to the versions of Barry’s spread sheet calculator that I have posted where some of the BFO parameters differ from the Inmarsat table. (I determined the temperature coefficient of the translation oscillator, which was somehow considered proprietary…)
Nothing that I have said should be taken as denigrating the validity of the published BFO correction as a reasonable estimate for the period of the flight.
As I mentioned when posting I am truly tired of this subject….
But I do have another question where my memory fails me.
I believe that there were only 2 phone call attempts after the loss of radar data. One roughly within a minute or two of the FMT and one some hours later.
When such calls were made, (a) was the caller aware at the time that the call was received but not answered? and (b) was the pilot aware that the calls were made and chose not to answer?
Was there any relationship between the times of the call and milestone in the original flight plan to Bejing?
@Sid,
Concerning your first question, part b): Andrew descibed the annuniciation of an incoming SATVOICE call in a previous comment, that may be useful.
First question, part a): as for the party making the call on the ground, it’s my understanding that a ground call initiation is analagous to any other network voice call. Inmarsat’s voice system was an Ericsson provided 3G system. The unanswered call timeout is configurable and, apparently, that was set to 60s.
Related to the normal conduct of the MH370 service, there was no routine need to make a voice call from the Operational Control Centre to the aircraft. Messaging via ACARS would be more routine (as evidenced during the flight but messsage transmission, GES to AES, failed to be acknowledged & subsequent network retries also failed).
@Sid
The satphone calls were approx. 1840 and 2314. The 1840 SatCall is important because the BFO shows a maneuver was made, either turn south or descent or both. Whether or not that was the FMT (final turn south) is the basic assumption difference between a 38-40s flight path and a 180s flight path, whereas the 180s straight south case generally envisions some additional maneuvers before the FMT. The 2314 call BFO is less helpful re: path but I think was done just before the official announcement of lost aircraft. I don’t think we have any confirmation that anyone in real time grasped that the Satcalls going through meant the aircraft was still flying.
“On the flogging of a dead horse”
After filling in data holidays close to the 7th arc, OI aims to search wider, over an area of about 30,000 square kilometers. The chances of finding the aircraft in this zone are close to nil.
It is beyond doubt that the end of flight was caused by fuel exhaustion.
The final power-cycling of the SDU confirms that the descent was unpowered. Unpowered descent in degraded control mode ends with a crash within a 15NM radius of point of fuel exhaustion. There is zero prospect of an “un-commanded” glide.
There is no positive evidence of pilot intervention after FMT. If the aircraft was unpressurised for 4+ hours – which is an INESCAPABLE condition for the observed fuel endurance – nobody was alive at the end.
Without human intervention, there is no possibility of pilot-controlled recovery and glide. In any case, if the “intention” was to glide and ditch, you would have extended flight by ~30 mins using an idle descent BEFORE running out of fuel – which is ruled out by Arc 6 BTO and BFO. You would not have allowed the aircraft to run out of fuel before ditching – which is ruled out by the power-cycling of the SDU. You would have extended flaps – which is ruled out by the flap and flaperon.
Using final BFOs, Holland produced bounds for final rate of descent 13800fpm – 25300fpm. This is roughly 10 times a normal rate of decent and equivalent to a VERTICAL speed of 136 – 250 knots (midpoint 193 knots). With vertical acceleration of @0.68g (increase 130kt per 10s) and a horizontal speed of (say) 250kt, the aircraft would be approaching M1 within THIRTY SECONDS of 00:19:37
Trailing edge damage due to flutter is consistent with overspeed in descent, while the state of debris recovered (as well as retracted flaps) is consistent with hard impact with the water.
The prospects of finding the aircraft more than 15NM from the 7th arc were already very slim. The chances of finding it further than 25NM from the arc must surely be close to zero – no matter which segment of the arc you favour.
I recommend that OI should prioritise an area ONE TENTH of the size (3,000 square kms) between 39.45S and 39.80S, width 10NM each side. A more conservative/comprehensive search of this zone would stretch from 39.45 – 40.00, width 15NM each side, for a total area of 7000 square kilometers.
@all: Are there any visualizations or simulations of how the debrus field will look after all these years have lapsed assuming all existing analysis of high speed descent at fuel exhaustion is accurate and there was no glide?
@PaulS
I agree on your first point.
I have to admit that it is a bit strange and unsettling that Ocean Infinity still didn’t provide information about the area they intend to search, i.e. the “search box.” Even though it was said it is mostly based on search recommendations from UGIB 2020 paper, the ship movements in February and March seem to be a sort of combination of search recommendations from the UGIB 2020 in terms of width (actually it is slightly wider) and Iannello and Ulich “Improved Prediction of MH370 Crash Location Based on Drift Modeling of Floating Debris” from 2023 in terms of the length of the area along the 7th arc. I uploaded the two maps side by side so you can compare:
https://drive.google.com/file/d/18zMBizqORsEZf16tiWDScIgrMO6NuWWb/view?usp=sharing
It is a little bit puzzling why OI decided to search the area that was not searched before (Phase 2) after filling in the data holidays in Phase 1. One would expect they would continue “down” the 7th Arc first since the probability is much, much higher than away from the Arc, right?
Although the reasons for such a search order can be benign and simply operational in nature, together with the lack of the official search map, to me it brings some sense of uncertainty.
@Paul Smithson
You said: “After filling in data holidays close to the 7th arc, OI aims to search wider, over an area of about 30,000 square kilometers. The chances of finding the aircraft in this zone are close to nil.”
They didn’t search the whole area from 36-33°S close to the 7th Arc, but I hope they will. Maps that are linked in my last post (and were also previously posted on this blog) actually contain the “data holidays” marked in dark green.
@Marijan, I have no issue with searching data holidays close to the 7th arc to the extent that any areas of plausible magnitudee remain.
@Paul Smithson
Yes, I know, but I thought you are saying that OI finished with searching the holidays in the vicinity of 33-36°S. I hope that is not the case and that they will go back and continue searching the remaining holidays close to the 7th Arc
@Don Thompson
I could not find the SATVOICE post using the link you provided. Could you provide a more detailed cite? Thanks…
@Sid
This should be the direct link to a comment by Andrew, replying to Viking, with an explanation of SATVOICE annunciation.
@Sid @Don
Apologies for butting in, but the link you provided Don has a redundant ” character at the end. Sid if you copy the link and paste it to a browser, then delete the ” at the end of the string before hitting enter, I think you will find the intended comment.
@John, @Sid Bennett, @Don Thompson
The link is repaired.
Re-visiting the end-of-flight scenarios I estimated time remaining in the air from the last datapoint at 00:19:37.
Holland estimated rate of descent at that point between 13,800fpm and 25,300fpm, midpoint 19,550fpm. Using midpoint ROD at 00:19:29 and 00:19:37 he calculated a rate of acceleration of 0.67g (22.3 feet per second per second).
Now let’s look at how much time remains in the air assuming various altitudes at 00:19:37 using midpoint ROD and estimated acceleration.
From 30,000, 20,000 and 10,000 feet you have the following predicted time remaining airborne; VERTICAL speed at impact; horizontal distance max (assuming a horizontal vector of, say, 200kt):
30,000ft: 39 seconds, 708 knots; 2.2NM
20,000ft: 30 seconds, 589 knots; 1.7NM
10,000ft: 18 seconds, 431 knots; 1.0NM
Now let’s imagine that instead of continuing to accelerate at 0.67g you don’t accelerate *at all* beyond that mid-point ROD of 19,550fpm, how much extra time and distance does that buy you? Answer: surprisingly little. The corresponding (straight-line) numbers are:
30,000ft: 92 seconds, 193kt; 5.1NM
20,000ft: 61 seconds, 193kt; 3.4NM
10,000ft: 31 seconds, 193kt; 1.7NM
If vertical acceleration did continue at the rate of 0.67g, the aircraft had horizontal vector of between 100 and 200 knots, and initial ROD was in the max-min range calculated by Holland, then you hit M1.0 – and presumably structural failure – within 29 to 39 seconds of 00:19:37.
In any variant of these scenarios, the ROD estimates produced by Holland predict that impact with the water would occur roughly 20-90 seconds after 00:19:37, within a distance of 5.1NM (and more likely <<2NM), with final speed consistent with structural failure.
Even allowing +/-6NM for final BTO noise uncertainty and about +/- 1.5NM for final BTO altitude uncertainty, the implication is that the wreckage most likely lies within 10NM of the arc.
The scenarios above obviously assume downward acceleration of 0-0.67g. If the aircraft was in phugoid at the time of 00:19:37 then it could conceivably have entered a further climb-decent cycle. However, this is counter-indicated by the fact that the high ROD occurs much earlier than any of the Boeing simulations. I subscribe to Victor's theory that this may be attributable to a shortlived relight of one engine, inducing bank and then nose-down acceleration. I stand to be corrected, but in that scenario, I don't believe the aircraft would continue through a cycle of phugoids.
@Paul Smithson: If the plane was in a wings-level phugoid with no pilot inputs, it’s doubtful the downward acceleration would reach 0.67g. At that point, either there were pilot inputs, or the plane was in a steep bank, possibly induced by an engine relight attempt.
@victor – I agree. And since pilot input in UGIB scenario is impossible due it implies steep bank and, by implication, continued downard acceleration.
…never try and edit a post on phone…
What I’m attempting to say is that steep bank and continued downward acceleration is the inevitable deduction, given the premise of no pilot input.
@Don Thompson, @John
Thanks…
@Paul Smithson: If there were no pilot inputs and the plane reached a downward acceleration of 0.67g, I agree that the plane likely crashed near the 7th arc.
@Don Thompson
I am sorry for being so obtuse, but us it correct to say that in each call the pilot would have had what is considered a call indication in the cockpit? But what about the ground station.
Could the caller determine in real time that the call was connected to the airplane and that the call was not answered? as compared with the situation where the plane had already crashed? If not, why was a second call attempted so much later?
The pilot would be aware of the call attempts, but unaware of the pings.
Armada 86 05 docked in Benoa port, Denpasar, Bali at 2300UTC this evening (Sunday Dec 14th).
The vessel is now indicating its next (port) destination as Fremantle, Australia. ETA: Jan 21, 12:00
(in 37 days).
Its departure from Benoa is yet to be confirmed.
They just departed from Benoa port, Bali a few minutes ago
Looks like Armada 86 05 initial steady heading after departing Bali is ~237° indicating they would intersect 7th arc much further north than last phase priority search areas if they hold that course?
@Ben John Thanks for the interesting information. Is it possible to estimate to what point of the 7th arc the current course of A85 06 will lead?
@Edward I could easily be wrong as Armada 86 05 might simply be heading into 2-3 m waves to soften the ride until past a small low pressure system in their path seen on https://www.windy.com/-Waves-waves?waves,-11.362,116.367,6,i:pressure,m:dnXajlU
I just used a ruler to extrapolate their current course heading towards 7th arc via MH370-Caption Armada Tracking: https://www.mh370-caption.net/index.php/armada-tracking/
@Sid asked ‘is it correct to say that in each call the pilot would have had what is considered a call indication in the cockpit?‘
Yes, that is the case.
‘But what about the ground station.
By ground station I assume you imply the calling party at the MAS Operational Control Centre? The operation of the Inmarsat network extends the public switched telephone network to aircraft. If the connection for the call could be completed (as the two MH370 examples were) the caller would hear a ringing tone. If not, the caller would recieve a ‘number unobtainable’ tone from the network. A requirement for AMS(R)S voice/SATVOICE ground-to-air calling was that it should conform to normal ITU call progress indications (ringing, busy, unobtainable) to a user on the ground.
‘why was a second call attempted so much later?‘
The second call was attempted at 23:15UTC, 07:15MST (local time in Kuala Lumpur is GMT+8). The KL ARCC was invoked at 05:30MST. 9M-MRO’s planned arrival at ZBAA was 0534UTC. No entries were recorded in the ARCC log book until 06:32MST. I suspect this second call was made on the request of the ARCC whose actions included ‘Obtain information of aircraft position if necessary by [various means]‘. Cutting through the fog, with what we know now, details of the disappearance were being realised at this time on 8th March.
I’m dubious about any assertion that either of 9M-MRO’s flight crew would understand the low-level operation of the Inmarsat Classic Aero AMS(R)S system. The packet data side, ACARS and ISO8208-SSN exploits time division multiplexing over shared and arbitrated access to radio channels, the voice side exploits a single channel per (voice) carrier. There’s a lot to comprehend there.
@Ben John, all
It’s likely too early to read anything into Armada 86 05’s track at this time. The search area is approximately 1700NM ahead of the vessel, with winds and waves to consider in the route plan. It’s clear the vessel is at present northeast of a direct route.
I have been, generously, permitted access to a web platform with electronic nautical charting, position reporting, and weather forecasting. Am presently learning my way around that and will endeavour to make updates.
@Don Thompson
Thank you for the details.
So, the two phone calls were received by the pilot, but not answered. The pilot, if alive, would know.
The call center would realize that the plane was still flying in each instance.
The plane obviously deviated from its apparent path at or about the time of the first call. We know not why.
Did the plane deviate from its later path at the time of the second call? Who knows….
But if it turned towards the South (say 180T) at 90.259E 29.218S
it would have been at 90.2E 35.9S at 00:11 if the TAS was 430kt. (This is from a quick hack at one of my spread sheets and could be wrong…) Since the TAS is unknown, other deviations to an azimuth less than 186T combined with a reduced airspeed would result in a solution.
This is not a SWAG but a WAG, but interesting.
@All
You may have noticed that I was absent from the discussions for the last few weeks. It has nothing to do with missing interest or shortage of good inputs and arguments.
The reason is totally private. my 100-year old mother was very unfortunate to fall and break her hip. Therefore I had to spend a lot of time helping her. Now her condition is beginning to improve so I am going to return gradually to the discussions.
A quick look at the latest inputs indicates that the ship heading for Freemantle seems to temporarily head towards my position near the 7 arc. It may have completely different reasons, but it is a funny coincidence. I hope they are going to search there before heading for Freemantle.
@Armada Crew and Team
If your monitoring the MH370 chatter and if you are on the search all the best on your Journey. Seasons Greetings and a Happy New Year.
@Viking – sorry to hear of your mother’s fall. Wishing her a speedy recovery!
@All,
When the issue of the fuel flow reduction produced by turning the air packs off was raised again by Paul Smithson, I submitted the following question to ATSB on November 18 :
“We are having some discussions now concerning the fuel flow reductions for 9M-MRO’s engines if the air packs were turned off in high-altitude cruise.
My fuel model, published in 2020, uses a 1.6% fuel flow reduction. There are a few values floating around the internet which are higher, about 3% or perhaps even a bit more. None of these higher values are from Boeing or from Rolls Royce, so they may not apply to MH370.
The reason I am readdressing this issue is that a larger fuel savings with air packs off would allow a slightly faster and longer route to be flown, ending at Arc 7 further south than 36.5S, below which my fuel model currently predicts insufficient fuel to achieve fuel exhaustion circa 00:17:30 UTC at Arc 7.
So, I wanted to ask you if ATSB has this information or might get it from Boeing or Rolls Royce. If I could get a more accurate number, I will re-run my fuel model for tracks further South to see how far past 36.5S is fuel-feasible.
. . .
Having a more accurate number for the air packs fuel cost would allow me to improve the accuracy of my current 36.5S fuel limit and provide a useful latitude limit should the search need to be extended to more southerly latitudes.”
On December 17 I received the following response from ATSB :
“ We forwarded your request through to our Boeing contact re MH370 and got the following response that I hope informs your work:
“Thanks for the additional background. Our aero performance group took a look at the work that was done in support of the investigation originally, and based on the information that the IG provided in their question I don’t believe we would have any information that could add more refinement to their current assumptions.
For NTSB/ATSB background:
At the time of our initial performance review, our studies determined that the impact of the packs on versus packs off situations showed the differences were of negligible impact to the overall findings. Based on our cursory review for this question, I don’t believe the differences would be any greater than the 1.6% difference they already seem to be applying.” “
In summary, the Boeing aero performance group does not believe the air packs off would reduce the fuel flow by more than 1.6%. Therefore, the current fuel limit of 36.5S on Arc 7 is still valid, with potentially a small shift further south due to reduced electrical loads (primarily in the cabin and the external lighting). In particular, except for emergency lighting, the cabin electrical loads may be shut off by a switch in the overhead panel.
@All
Please find through the link a summary of the developed methodology and key results obtained in my research during the last 2-3 years. This a concept version; comments / suggestions are highly appreciated.
https://www.dropbox.com/scl/fi/logh0fng677zeictltoto/Technical_Note_v24_6.pdf?rlkey=d48btkiezxgcqu0uui1rl3lix&st=k8chteyv&dl=0
@Niels
Neils, you persistence on this matter is appreciated and admired.
@DrB
Re, “Therefore, the current fuel limit of 36.5S on Arc 7 is still valid …”
Bobby, in determining that “fuel limit”, what is the assumed fuel remaining at say 18:22 UTC and 18:40 UTC?
@George G
Thank you, George. It’s actually quite amazing (and essential) that we keep going here collectively. And that after all those years imo the collective insight still progresses, although perhaps in small steps. The diverse backgrounds, expertise and angles of view show to be essential in the process.
@all
The civil radar ATC-Caption-2025 dataset (now version 1.5) has been updated via some corrections in the merged files of the previous version 1.4 on 09/10/2025:
1) The positions of different aircraft were mistakenly mixed up because the same track number had been assigned to them in Senai’s radar data.
2) Removal of Mode A only tracks caused by garbled codes
3) The ADS-C and Langkawi data of JAI12, SIA495, and MAS176 are now combined
The dataset is available here: https://www.mh370-caption.net/index.php/malaysian-civil-atc-radar-data-atc-caption-2025/
Armada 86 05 is planned to arrive in Fremantle on 23/12/2025 at 05:00 and to depart on 24/12/2025 at 12:00.
From her current position to Fremantle, the distance is ~810Nm. Her average speed is about ~10kt, thus ~3.4 days : this is in accordance.
The progress of the ship can be followed every ~3hours at:
https://www.mh370-caption.net/index.php/armada-tracking/
@Niels
1) Interestingly, you find a maximum likelyhood of the end of flight at -35.6° (on Arc7) which is so close to -35.7° (prolongation outside Arc7) that we found and reported here in 2023 (cf https://www.mh370-caption.net/wp-content/uploads/3-known-trajectory-and-recalculated-trajectory.pdf)
2) Taking Boeing’s table and considering the Max range at FL300 (thus Mach 0.706) appears a simpler approach, isn’t it ?
3) Why considering CTT in a FIR (ATC Region) where routes are flown using magnetic tracks ? A pilot would not do this. So there must be a switch somewhere.
4) Switching from Magnetic to True, where (~ -20°S) the declination deviates too much East towards Australia, is operationally more realistic
@Mick Gilbert,
To answer your question, I estimate the fuel on board to be about 34.7 tonnes at 18:22 and 32.8 tonnes at 18:40.
My new fuel model is a work in progress. It starts at 17:26, after the turnback was completed, when the new course, speed, and flight level settings were activated (LRC circa FL385).
For simplicity I am currently modeling what was called Case VIII in UGIB (2020). It assumes that the air packs were turned off at 17:26 and were never turned back on. I use the same 1.56% fuel reduction for the air packs used in UGIB (2020) and affirmed by Boeing. It also assumes there was no post-diversion descent and ascent or holding pattern prior to 19:41. I do allow for the small fuel cost of making the turn(s) needed to achieve a southbound course. For the fuel quantities after 18:29 I assume LRC at FL390.
DIVERSION ROUTE: 17:26:00 (diversion settings completed) to 18:29:00 (right lateral offset maneuver completed)
17:26 fuel 40.79 tonnes
18:22 fuel 34.7 tonnes
18:29 fuel 33.94 tonnes
FMT ROUTE: 18:29:00 (offset maneuver completed) to 19:41:03 (Arc 2)
18:40 fuel 32.8 tonnes
19:41 fuel 26.37 tonnes
SIO ROUTE: 19:41:03 (Arc 2) to 00:11:00 (Arc 6)
00:11 fuel 0.18 tonnes
This set of route assumptions results in a small fuel shortfall (about 3 minutes) at MEFE. This is within the accuracy of the fuel model. It can also be resolved by assuming significant electrical load reductions (equivalent to 0.8% in fuel flow) occurred soon after diversion.
@DrB
G’day Bobby. Thank you for the detailed reply.
To facilitate the tracking of Armada 86 05, the best information we can obtain will be displayed in the first 3 lines of the menu (white rectangle in the upper right corner) of our web page at https://www.mh370-caption.net/index.php/armada-tracking/
The Armada 86 05 is now expected to arrive in Fremantle on 23/12/2025 at 13:00 and depart on the same day at 19:00.
Thus, the search could possibly resume somewhere between 28 and 30 Dec.
I don’t know about you guys, but for the first time since the search started in 2014, I feel pretty confident that the plane is going to be found. Various aspects seem to have matured enough in the meantime, including various calculations, underwater search technology and even the people involved, so it feels like things are finally falling into their place.
@Jean-Luc Marchand
Interesting indeed, that these different approaches indicate a similar 00:19 latitude. On the other hand: we’ve both tried to adhere to the Inmarsat data with little BFO bias frequency offset and then S35.0 – S36.0 is typically where you end up.
It’s hard for me, without an aviation background, to judge what would be a more likely scenario from a pilot perspective. From a data perspective, putting the switch to CTT at a location where the magnetic track starts to cause deviation from the measured data strikes me as slightly opportunistic, but of course by no means necessarily wrong as a scenario. On the other hand, it is interesting that you found a good fit to the data for a route that includes such a long stretch of CMT at that specific FL and Mach number.
I would like to point out that I leave open what happened before 19:41 UTC (except for estimating the fuel left) and a switch to CTT could have happened “just” before 19:41 (in the case detailed in the technical note), or “just” after 19:41 (work in progress). But if I understand you properly, you suggest that a switch to CTT in the region close to the equator would not to be expected from a pilot / ATC perspective? But standard procedures weren’t followed anyway when diverting the plane?
@Jean-Luc Marchand
Your MH370-CAPTION.net search tracking map is much appreciated. Please consider the merits of adding the Java candidate site to your map. You currently include the “Godfrey/Coetzee/Maskell Hotspot”, “Lyne Hotspot”, “Pattiaratchi/Gibson Hotspot”, along with your own “Blelly/Marchand” site, IG, and others.
The Java site is very specific. It has a solid scientific foundation unlike WSPR, and has verifiable seismic/hydrophone acoustics vs Lyne’s speculative assertions about mythical physics. Candidates with daily media coverage get headlines, but they shouldn’t be driving the forensic approach toward locating the crash site.
On previous searches, CAPTION provided .kmz files that allowed importing the OI AIS tracking data into Google Earth Pro. Your github links were most helpful but need an update. Thanks also for the revised MH370 radar tracks that help clarify the known segments of the flight between waypoints.
@Viking
Our hearts go out to you as a full time caregiver helping your centenarian mother recover after a serious fall. You have your priorities right, and we’re glad she’s on the mend.
@*
Best wishes to the entire OI team who are working over Christmas, scouring the SIO to find the truth about what happened to MH370.
Happy Holidays to All,
Ed
@Niels
If we may, some elements for consideration 🙂
A long-haul airline pilot, in particular, knows where in the hemispheres magnetic declination varies significantly and that it must either “neutralize” it (switch to TT) or “compensate” for it by continuously adjusting their heading/track based on changes in the value (based on previous flights navigation). In our case, if he had remained in “magnetic” mode, the heading would have changed significantly, and not necessarily in the likely direction he probably wanted to take, which, in our view, was to go to the middle of nowhere. ZS often flew to Australia and was well aware of this phenomenon.
On the B777, there is an automatic switch to “true” beyond a certain latitude (82° South or North), but nothing prevents it from being done manually whenever desired, especially if it is in a mode other than “LNAV”. We believe that in the context of our case, the switching was done before because of the strong deviation towards Australia and its detection capabilities.
In our hypothesis, we did not consider LNAV because it was irrelevant for the purpose of getting lost in the middle of the SIO, but rather we considered a direction based first on a “magnetic” track (since navigation routes are based on “magnetic”), then based on “True” track. In our calculation of a route intersecting the Inmarsat arcs, the initial magnetic track remained constant at 188°mag until approximately 20°south, then switched to a constant 178°True until the end because of the arcs information. This led us to believe that he changed his navigation reference at that point.
Regarding the Mach Maxi Range (MRC), it is very much of interest when one is not “in a hurry” to go and end one’s life… (hypothesis). GS at FL300 (see below) suits very well this purpose to last: the slowest IAS speed (holding speed) recommended by Boeing .
In FCOM, the FAA or ICAO recommended « maximum holding » speed at altitude > 14,000 ft = IAS 265 kts => M0.706 => MRC (TAS 427 kts Isa+12) in Boeing table appendix 1.6 E
Also, it would be very surprising to cut off anything—one pack (or two packs -> depressurized at high altitude again with a mask ? -> very uncomfortable) and other electrical consumption—when one is certain to end their life in a few hours. He surely had other things to think about….
Finally, regarding the fuel; Our fuel consumption on the “southern” route is consistent with the FCOM tables interpolated between HOLD flaps 0 and LR, corrected for overconsumption and temperature. This corresponds to the fuel mass at the Boeing arc (18:28:06 UTC) divided by the remaining flight time (5h37’30”) ISA +12 average => MRC and FL300. Taking into account that between this Boeing arc and 18h40, the speed evolved, a total average speed TAS on the route of “431kts” was obtained. Subsequently, this indicates the second engine cut-off between arcs 6 and 7, consistent with the possible tracks graph in Appendix 1.6 E (between the blue tracks 437 and 416 TAS).
@Jean-Luc
I would agree, somewhat, but differ. For the sim cases, I would also envision LNAV is “irrelevant for the purpose of getting lost in the middle of the SIO”. Sim cases appear to me to be 180s CMH, which has long been a suggested alternate interpretation. For the actual MH370 flight, I believe the BTO/BFO data follows a 180s CTH “straight” south heading until about Arc5, where a heading change possibly to CMH was dialed-in to follow the south-easterly operating line to the Magnetic South Pole as suggested by the sim cases. I see the operating line in the sim cases as the map/evidence for all ZS MAS flights, presumably with differing entry and end points based on fuel/etc.
@Jean-Luc
I’m afraid we’re working at two different sides of the spectrum from scenario based to data driven, so for me it is quite a stretch. While I understand there is knowledge and experience based “intuition” behind finding this route, my feeling is that once you allow for a “model switch” at arbitrary position mid-way the SIO, there can be numerous other routes that would fit the data. Just as an example: you could start the CMT route at a slightly different position and initial bearing and then switch to a different CTT at a different position and moment. Wouldn’t you end up at a different 00:19 latitude?
In contrary to your “forward reasoning” approach, I work the other way around: assuming fuel exhaustion and no pilot inputs I then “reason backward” and try to find the model that fits the data best and is consistent with the assumptions for as long as possible (preferably all the way back to 19:41). And then see how much fuel would be required at 19:41, and if that can be matched with available fuel estimated for that position.
@Niels said: While I understand there is knowledge and experience based “intuition” behind finding this route, my feeling is that once you allow for a “model switch” at arbitrary position mid-way the SIO, there can be numerous other routes that would fit the data.
Exactly. This approach just leads to unresolvable squabbles about the veracity of the “hunch”. That’s not to say that any hunch is wrong.
Armada 86 05 is approaching Fremantle, it’s due at the Fremantle Inner Pilot Boarding Ground around 1500hrs AWST/local time. Then anchorage for fuel.
I’ve plotted a simple, direct route for Armada 86 05 from Fremantle, West Australia, to the search area (assuming a return to the final AUV recovery position of 2025-03-28).
Port information indicates departure from Fremantle 2025-12-24T2030AWST (0430UTC).
The prevailing weather shows waves expected to exceed 2m during 25th-26th, peaking around -26T0000UTC at 3.5m following after an earlier peak of prevailing wind at 24kts. Waves and wind to diminish during remainder of voyage.
Estimated arrival: 2025-12-28T2030UTC.
So, yes, ‘Late December’ looks to have worked out well.
@Ed
Thanks for your kind words. It is a slow process to recover from a broken hip near 101, but she is making faster progress than expected by the experts.
One negative side effect is that I do not have time to write a pedagogical version of my reanalysis of the discrepancies of headings deduced from HA01 data until she becomes significantly better. I think I have now identified the fundamental reasons for the disagreements. I no longer believe reasons are predominantly in your analysis. Everybody involved in the analysis seems to have made the same errors (including myself), but to different degrees.
The original reports you quote in your work contain the explanation if you read ‘between the lines’. It is a big disappointment for me that the authors did not state things clearly, since that could have prevented most problems for everybody else.
Exclusive video of the #MH370 search ship #Armada8605 taken by Balm Commercial Photography @GeoffreyT_Air
https://youtu.be/mrCbP2f3eSc?si=ESgp1Puo8UEBMC_c
With a healthy serve of good fortune, 2026 may be the year we can finally, and conclusively close the book on MH370.
If so, after more than a decade of the collaborative (and in some instances, combative) efforts of a retinue of probably hundreds of contributors, some acknowledgements would be appropriate.
Might I suggest the following as part of the broader scheme for recognition:
The Limpet Award – for the person or persons who have stuck most fixedly to the one position/theory, in spite of all evidence to the contrary.
The Karma Sutra Medal – for the person or persons who have earnestly propounded the most different, and most widely differing positions/theories, with apparent blithe disregard for any of their own earlier work.
The William Joyce Memorial Award (aka The Lord Haw Haw Award) – for the media person or persons who have shown the least regard for facts, accuracy, integrity, and veracity in reporting on MH370.
The Phlogiston Trophy – for the person or persons who have shown the greatest indifference to proper science, logic, and reasoning.
The Jim Henson Memorial Medal – for the person who single-handedly created, maintained and “voiced” the most sock-puppet accounts (fake, social media personas).
Best wishes for the festive season to all. Here’s to a happy, safe and prosperous new year.
@Ben John
Nice imagery, shame about the commentary.
That particular source routinely features one of the least reliable commentators on the topic of Ocean Infinity’s search campaign, and the assets being deployed by OI, while also consistently promoting a methodology that has no scientific basis.
A Theory for Your Consideration:
MH370 – Pilot, 53
Air India Crash – Pilot, 56
Bangladesh 211 – Pilot, 52
EgyptAir – Reserve Co-Pilot, (demoted), 59
China Eastern Crash, First Officer, 59
What stands out to you?
There is a type of rare dementia that tends to strike people in their 50s that destroys empathy, judgment, and executive functioning called frontotemporal dementia, which is overlooked and, if unusual behaviors come to attention, is misdiagnosed.
@Viking:
I’m always open to getting educated, but you’re implying my analysis is in conflict with other findings due to a vague insight that you decline to explain.
Your comment is specific to the CTBTO H01 Cape Leeuwin hydrophone array. That is not the basis for the Java Anomaly, which stands out on H08 and was pinpointed with seismometers directly on the 7th Arc. The SOFAR channel from H01 to the Java site is mostly blocked by shallow coastal waters, but reflections off the Java coast do register.
The accuracy of the my H01 headings on various events are within a fraction of a degree when compared to the two official Curtin reports, which are technically accurate. The difference is due to additional calibrations on both arrays that I made against known events and sources on the day of the crash. The heading to the Java anomaly has recently been confirmed by operators of the seismic survey ship off Exmouth who also detected it.
The factual hydrophone array evidence doesn’t match up with high profile media reports by Kadri, Lyne, and various others, either because they miscalculated the phase arrivals, depended on those previous miscalculations, or looked only at arrival timing that matched their theory and ignored the back azimuth entirely. Timing triangulation with several other single-channel hydrophones is all consistent with the Java anomaly location, per an unpublished French university study.
There is little doubt about the location, time, and anomalous character of the acoustic event at 8.32S 107.96E. It’s prior assumptions and other interpretations that prevent it from being a prime search candidate.
Thoughts this eve are with the OI crews working through the holidays toward finding MH370.
@Ed
I am not postulating that your analysis is wrong compared to traditional methods in the field. My criticism is directed at the traditional methods themselves. When I get more time I will surely write things up as some kind of publication.
I presented my ideas for Kadri some time ago. After that he went remarkably silent. Since then he admitted that I am right about some of the most important points. He is still uncertain about the rest. I am quite sure I will convince him on those things too.
The problems go much deeper than simple calibration errors. Actually I think the calibrations are mostly correct.
The fundamental problem is that the traditional methods are designed for strong signals (particularly from nukes) which predominantly propagate as standard guided waves in the traditional guiding channel in the sea. For this type of signals the methods are perfectly fine.
However, a plane crash is different. The amount of energy it transmits in the standard guiding channel is minute. Instead most of the energy goes into other propagation modes in the ocean. Here the Indian Ocean is also special, since large parts of it has a very deep jump layer (due to extremely high surface temperatures), capable of transmitting a lot of energy over long distances.
The standard guiding channel has low dispersion. In contrast, the other transmission channels have large dispersion (with different magnitudes and signs for different channels). This means that the signals build up significant dispersion broadening over distances to e.g. HA01. If the analysis does not take that into account it is very easy to overlook signals from MH370.
These problems are further enhanced by higher propagation speeds for the other modes (particularly for seabed R-waves). The distances between the 3 detectors at HA01 (and similar facilities) are too short to accurately determine the directions, or at least marginal.
Therefore a traditional analysis (wrongly) assigns a direction dominated by simultaneous arriving (much weaker) signals at high frequency guided in the standard channel.
Similar problems occur in optics. I can provide some links to papers I published about how to deal with such problems in optics. The same methods can be used for sound waves.
An added advantage is that determination of the dispersion provides an estimate of the distance to the event.
@Viking:
It’s true that those using the simplest back azimuth methods will only see the strongest signal at any moment. LANL had ruled out the H08 array as too cluttered with seismic survey noise to be useful. Curtin used a method where the top few peaks can be matched. They got strong H08 detections right including the Java Anomaly, and down to slight deviations in azimuth as the known survey ship moved over hours compared to the ground truth of its AIS log.
I use a much slower method where a computation is made for each possible bearing. Simpler methods work with uncalibrated spacing, but also pick up noise outside the SOFAR channel. Using calibration from deriving a combo speed of sound and spacings of the hydrophone triads on that day, much more sensitive methods can be applied. Even slight variations in phase can be detected.
The ideal spacing depends on the frequency in use. Simpler methods fail below about 3 Hz with the CTBTO triads. I’ve been able to get bearings on signals below 0.2 Hz, though the instrument highpass filter design log rolloff is down to 1/100 gain at that point. Wider spacing to get bearings below that frequency in future arrays may be of some use, but certainly doesn’t help with analysis of the recordings from 2014.
Dispersion is quite visible on spectrograms, and is in fact such a common characteristic of ice events that the amount of dispersion can allow a distance estimate when only a single hydrophone is available. However, the dispersion does not disrupt the azimuth calculations, as the propagation shift over the path at any frequency is essentially identical for all three detectors.
Also true is that other phases and modes can travel outside the SOFAR channel. I am detecting what appear to be Scholte waves that travel at the crust/seabed boundary, (similar to what you call seabed R-waves), but a surface impact would not propagate that way, just as it did not get into the SOFAR channel directly.
The sound waves from a surface event over deep ocean travel downward and outward, passing through the SOFAR channel, eventually arcing back toward the surface. This cycloid path leaves large concentric rings of detectability spaced 40-80 km apart, with shadow zones between. They can only be detected if the sensor isn’t in the shadow zone (or the signal finds another path).
These arcing sound paths from a surface impact have strong dispersion, but also a curious feature. Just past the deepest part of the arc, the signals all line up again in a “caustic” and the dispersion cancels. This behavior is explained by Alec Duncan of Curtin University in a book chapter he co-authored, titled “Introduction to Sound Propagation Under Water”.
I mention this because I now believe that the “Curtin Event” described in the ATSB official report was actually a caustic from the MH370 surface impact reflecting off the 90 East Ridge, and detected with bearing 301.6 on the H01 array at 01:34:50Z. Some key info supports this: There was no cataloged seismic event in that direction on that day. That angle to the 90E Ridge is the same as the angle from 90E to the Java event site, a near perfect carom bank shot. The azimuth also shifts slightly during the signal, which would be the result of shorter to longer paths reflecting off the ridge. The depth of the ridge in that area is just below the SOFAR channel. An expanding ring of sound reaching down near the seabed could plausibly intersect the 90E ridge at some point. If that reflection exited near horizontal, it could propagate back to H01 in the SOFAR channel.
I’ve just realized there’s a way to check this! The 90E reflection path timing is about 12.5 min slower than would be expected for SOFAR propagation speeds from 7th Arc timing. Expected SOFAR timing from the Java Site to 90E ridge is 1563 sec. That means the cycloid path speed of sound would be 48% as fast as the direct SOFAR propagation speed. The true cycloid SOS could be integrated from World Ocean Atlas data, tested, measured from major reference events, or maybe even found in a Navy handbook.
Thanks for getting me thinking about that unpublished hypothesis again. I don’t know who might be convinced if the Curtin Event turns out to match reflection timing of the MH370 surface impact, but it could be enough for OI to eventually stop by the Java site.
@Ed
If two signals arrive simultaneously within the uncertainty, one would only get the heading for the strongest provided two conditions are fulfilled:
1) The strength difference is large (30 times or more).
2) The two signals are within the same frequency range.
In that case practically everything you write is correct. However, that is not the situation we are in with the HA01 signals. There the strength differences are around a factor of two, and the main intensity contributions are in different frequency regimes. In that case things like dispersion become extremely important. The problems for the HA01 signals are further exacerbated by the signals arriving pretty much from exactly opposite directions aligned with the shortest detector distance.
In that case, signals with high dispersion and relatively low frequency (from MH370) arrive from a northerly direction while ice-signals arrive from a southerly direction. Without dispersion-compensation the northern signals thereby yield a small correlation peak distributed over a relatively long time interval while the (short and mostly high-frequency) ice signals with small dispersion from travelling in the standard guiding channel yield a large correlation over a short time interval. All standard methods will yield a heading dominated by the ice signals in that case.
I am practically 100% sure that is what happened in most cases for everybody trying to find signals from MH370. However differences between the individual routines sometimes yielded other solutions. That is the reason everybody disagree so much about the headings.
@Viking:
Sorry, but you are stating as fact some ideas that don’t match my experience. You appear to be trying a method that gets incorrect headings, assume others have the same problem, and are attempting to explain that.
The CTBTO arrays have been in use for decades. They have been extensively tested and calibrated, for example by dropping depth charges around the SIO. Even earlier, a device called a “SOFAR bomb” was carried on oceanic flights in WWII, so a downed plane could be located to rescue the crew. There are many beam-forming algorithms in use for numerous applications. I’ve tried many. Changing the algorithm doesn’t alter the ground truth or the correct headings. A dispersive chirp doesn’t diminish accuracy, and in fact improves it by removing timing ambiguity.
Our hearing utilizes just two ears with some forward directivity, locating sound sources within nearly a degree of accuracy using time difference of arrival. Noise can disrupt that, but it’s robust enough that mammals evolved with it.
Curtin University uses a published calibration method for the HA01 array derived from years of known events. They essentially apply a correction table for headings calculated using the published hydrophone locations and expected local speed of sound. My method is different. I use the same expected local speed, but relocated the hydrophones slightly until SOFAR sounds from any direction matched exactly. I then applied a small rotation to align with known sources. When the expected delays are precise, the beamforming is greatly improved. More selective comparisons can also be made between the triad pairs. For each heading in my graphic plots, the three incoming sound streams are each delayed by the expected amount. SOFAR signals will appear in phase on all three channels. Sounds from other headings, off axis, or random noise do not.
My heading measurements on both arrays are consistent with those in the Curtin reports, and I believe they are accurate. The headings don’t shift when applying narrow bandpass filtering to remove dispersion, or with higher frequencies up to the 125 Hz Nyquist limit on the CTBTO hydrophones. I invite you to compare any bearing vs time graphs on my website with the detection plots by Alec Duncan in appendix H of the ATSB final report (figures 5 and 7). Ice events unambiguously come only from the direction of Antarctica.
Anyone who gets a significantly different result on those 2014 signals is likely miscalculating. Kadri didn’t read our prior reports and declined collaboration, or he would have known that the repetitive sound clutter on HA08 was seismic survey ships off coastlines near Exmouth and Burma, not military operations from the wrong heading toward Madagascar, or that a harmonic noise on H01 matches an ice tremor on the Antarctic coast, not a meteorite impact. His uncorrected misinformation is used in support of various MH370 claims by Lyne and others. Kadri describes filtering above 5 Hz, applies an entropy detection method to identify an event, then extracts correlation peak delays on the three channels to compute a heading. There is nothing wrong with that basic method, just his computation.
It is fine to start from scratch with first principles, but any new conclusions need to be compatible with known physics and tested against historical evidence.
You say, “That is the reason everybody disagree so much about the headings.”
Not everyone is in disagreement. Again, HA01 didn’t have a direct SOFAR channel path from the Java Anomaly site, but several other hydrophones did. That allows triangulation, which matches with the site.
Triangulation was checked using HA01 and Rottnest hydrophone arrival timing of the Curtin event against the estimated 90E ridge reflection point for a surface impact at the Java candidate site. The SOFAR timing error is just 3.2 seconds (less than the duration of the signal). Triangulation is more accurate than the H01 bearings, but the Rottnest clock drift correction, or my estimated 90E intersect could also be slightly off.
@Ed
Your answer focuses mostly on detector calibration. For clean cases with only one signal present at a given time and propagation in the SOFAR channel I agree with all calibrations made by you, and the other examples you mention. However, that is not at all my point.
What I am talking about is all the ‘dirty’ cases. Practically all signals I have assigned to MH370 are in this category. Perhaps a single one is not, but let us postpone details on that one until later. All I would like to say on it now is that Kadri actually got exactly the same result as me on that one, and it agrees extremely well with my proposed crash position. Just a few hundred meters off.
For all the other ‘dirty’ cases there are some kind of disturbing signals arriving from a particular area at the coast of Antarctica. Your method systematically get the heading towards that point for all those peaks. Some signals are indeed purely from there, so there is nothing wrong about your calculations, but many of them are mixed peaks with simultaneous arrival of signals from MH370 and Antarctica. For all those cases your method fails. It yields only the Antarctic heading.
A complete explanation of that is a lengthy and delicate story which I do not have time for writing up right now. In brief, most other methods share your problem. The reason is that they all share the same fundamental assumptions.
I was fighting similar problems in optics a couple of times. The first time was in 1993. For the solution I developed a particularly good fitting program which instead of stepwise detection (as Kadri) or various methods similar to yours actually fitted a complete theoretical model to the whole problem in one procedure. That worked better, but it was not flawless. The main problem was actually the same as for you. The program favored high-frequency solutions, and sometimes with a fixed phase-shift corresponding to 0.5 effective period for the signal.
Another case (not the only one) was in 2001 where I developed a generalized method for design of grating-based optical filters together with a Ph.D. student. Before our work the best results came from a group of Russian mathematicians using methods similar to yours (in optics). Most people in the field believed it was fundamentally impossible to do it better.
However, our 2002 paper in Optics Letters showed complete optimization of very complicated grating problems in just 6 minutes on a standard PC (today it would only be seconds). The Russian group used 24 hours on a supercomputer to achieve similar results (and with clearly worse quality).
More importantly for ‘dirty’ problems (e.g. including dispersion as for HA01) their method failed in practically all cases. Our method handled 99% correctly and with practically no additional CPU time needed. The core of the method was routines from my work in 1993 combined with a smart method to handle the dispersion invented by my student.
After our publication the Russian group claimed our method was fake news and worse things. A few months later they were totally defeated and gave up completely.
Last but most relevant for MH370 I recently helped some scientists in Australia to disentangle signals recorded by a completely different type of detectors at the time MH370 disappeared. That greatly enhanced the quality of their results. Their problems were also related to ‘dirty’ signals, and failure to include propagation above the jump layer in the analysis.
@Viking:
Martin, I think “the proof is in the pudding”. Here’s a graphical plot from HA01 that matches the first LANL HA01 report timeframe:
https://370location.org/wp-content/uploads/2020/08/200720-0739-H01@003730-005721-Win4096-BearingOverTime-fig.pdf
Nearly continuous noise sources from ice events are prominent, punctuated by strong ones. Basic algorithms would set a threshold for detection, and only register the strongest peaks. Simple correlation can be used in this type of plot, but the strong signals tend to splatter with aliasing and higher noise at all bearings, contaminating the detection of weaker signals. This plot was done using covariance of STFT, which is essentially looking at phase matches at every frequency using a 16 second FFT window.
You will see some weaker arrivals around bearing 275 from HA01. These are surely ice events reflecting off of Broken Ridge crests as shallow as 500 meters, intersecting the SOFAR channel.
Because even the strongest Antarctic events aren’t obscuring weaker events on other headings in that plot, I don’t believe that hidden MH370 events are masked by ice events. My reduction of the LANL report was that they looked at spectrograms, visually saw peaks in a signal preamble, and manually calculated the TDOA timing of the peaks to infer an event hiding from heading 246.8 at 00:52:00Z. Kadri claimed in 2017 and 2019 papers that he detected a signal above 5 Hz coming from bearing 234.6 at 00:50:00Z. Not even a weak signal appears on the above plot for either claim.
If you have advanced methods for extracting signal from noise in “dirty” conditions with one signal masking another from a different direction, I’d very much like to test it. Nothing in your 2002 Optical Letters paper about applying a sinc function in gratings seems to apply here. I appreciate that you have solid academic credentials and numerous scientific publications. I have only my experience in testing untold different signal comparison algorithms, not unlike Edison’s brute force search for the ideal light bulb filament. (Where is a Nikola Tesla when we needed him?) If you could educate me on how to further improve signal to noise in my graphical plots, especially by example, we might get some new insights on the acoustics. I firmly believe this is the additional evidence beyond SATCOM needed to locate MH370.
We can take this to private email again, as I know this is not a popular topic and not consistent with where OI will soon be searching.
@victor (who was replying to @niels)
“Exactly. This approach just leads to unresolvable squabbles about the veracity of the “hunch”. That’s not to say that any hunch is wrong.”
This is the entire problem in a nutshell. Very few of the current suggestions are provably wrong (radio wave reflection not included)… But the most likely spots for many theories have already been 90+% checked.
There are so many possible options that are ruled out once the unpiloted flight seabed is fully searched that it will be impossible to know which hunch is right until, hopefully, one day the wreckage is found. Before that, how to discriminate between hunches? Almost impossible.
I do have a soft spot for the Java anomaly, not least because it is very quickly searched and ruled in or out. I also understand that the IG has focussed on unpiloted paths because of the multitude of options if you go the other route.
I know nothing, but it really feels to me like this OI search is our last, best chance so to speak. After that we have nothing but hunches, and the hope of someone funding a search based on that hunch. One of the hunches is probably right but with no real way to differentiate between them.
For the families I hope OI are successful. I’m not super optimistic.
@vodkaferret said: I do have a soft spot for the Java anomaly, not least because it is very quickly searched and ruled in or out.
Me, too. It would be a great triumph of science and perseverance. I do hope eventually it gets searched, although I do believe the current search area should have priority.
@Victor, @vodkaferret, and @all,
Thank you for your words of support! Opinions on these pages, especially from IG members, have clearly shaped past searches. A thumbs up here might get the Java candidate a spot on search tracking maps for others to consider. I am confident in the analysis, and so believe that the site will be searched, eventually.
@370Location
We found the resources to add the Java Hotspot to our map 🙂
In addition, the Armada-tracking page has been updated with the possibility to download a kmz file tracking lively the ship on GoogleEarth.
https://www.mh370-caption.net/index.php/armada-tracking/
@all
Happy New year to all !
@Jean-Luc Marchand,
Many thanks for adding the Java site to your map. Much appreciated!
Your live .kmz download now works nicely with Google Earth Pro. Excellent!
Not to clutter your map but as an FYI, I submitted one earlier 7th Arc candidate privately to the ATSB in 2017 (revealed to the public in 2019):
https://370location.org/2017/03/mh370-acoustics-a-7th-arc-candidate-near-gulden-draak-seamount/
It was searched out to at least 35 km by OI in 2018, around the time I was focusing on the Java anomaly, so I didn’t pursue it further. My 2017 report says,
“Clearly evident in the map is a detection on the 7th Arc at S28.75 E99.25 (H01 bearing 292.0 degrees).”
The inverse great circle track back to HA01 is 119.2 degrees for a delayed glide after the 7th Arc. In 2023, the WSPR hotspot was moved to within 5 km of that heading track.
Thanks again for providing the Armada tracking!
Our friend Simon Maskell was recently interviewed for an article in The Guardian:
https://www.theguardian.com/environment/2025/dec/29/malaysian-airlines-flight-mh370-mystery-search-indian-ocean-infinity-robots
We learn a couple of things:
1. He is described as being “a FORMER scientific adviser to Ocean Infinity.”
2. No mention of WSPR. Not even a whisper.
@sk999
That could make sense. A few weeks ago on the YouTube/podcast RGodfrey disclosed that he felt WSPR area was not included in the current search. In another podcast RGodfrey disclosed that he was not currently in contact with OI.
It is unlikely that OI will search the WSPR area during this campaign.
I’m waiting for Maskell to publish his results that it is not possible to use WSPR data to track MH370. That would be the honorable thing to do, considering the past media attention he generated.
@Victor, @sk999, et
on the 16th April 2025 an article in “The Diplomat” provided quotes from Professor Simon Maskell. A link is attached below.
The relevant section starts on Page 4/9.
https://thediplomat.com/2025/04/mh370-the-devil-is-in-the-data/
@all
Armada 86 05 has now ‘hove to’ and begun manoeuvres to launch its first AUV. The location indicates that the AUVs will be picking up the search progress where it was left on March 28th, on the ‘saddle’ to the south-east of the 7th arc and progressing north-east along that block.
@George,
An interesting array of thoughts attributed to WSPR from the professor.
I imagine he’s been quite busy of late with the AUKUS Innovation Challenge award and taking up the Dstl/Royal Academy of Engineering Research Chair.
@George G, @all
Thank you for the link.
So far, all of the statistical studies on WSPR have been done by Godfrey as the primary author. I, and I think others, were waiting to see the study by Maskell et al from the Liverpool University (which would have been mostly independent from Godfrey’s influence).
I asked this on Godfrey’s website on 13 September 2024:
“A possible 6 month time frame was mentioned by Maskell in the BBC documentary. That has now passed. Should we still look forward to this being published before the end of the year or has it been abandoned etc.?”
He replied: “It is up to Simon to publicise his results, when he is ready. I know that he and his team have completed their processing chain and are making a few refinements. I know Simon wants a peer review and has already engaged with a number of other academics. Any talk of the research being abandoned is fake news.”
To my recollection, there were later comments about the study being in a finalization stage and being sent to a peer review process (late 2024/early 2025).
Someone who hasn’t been banned from Godfrey’s site could ask what the status of that study is today, as it seems that he has just quietly stopped talking about it (and so has Maskell with respect to the WSPR in general it seems).
Happy New Year to All,
I’m back after multiple years absence, and haven’t managed to read everything that has been posted since… That said, I’ve also had a lot of trouble using the Site Search to find items I’ve already read – I wonder if it works properly.
Regardless, it’s now been raised that many (the majority?) of the identified parts that have been found washed up on beaches etc give indications of tension failures. Various commentators seem to have been puzzled by this, so I’d like to add my 2c worth.
First, in reality, very few pieces of wreckage have been found. And, in order to be found at all (so far), they had to be able to float. I believe this solves that riddle. Parts that suffered high speed impact compression failures are very likely to have “had the stuffing knocked out of them”. Any honeycomb sections are likely to have been crushed or compressed, any hollow items the same, or at least punctured in various ways and places, and any remaining buoyancy was insufficient to float the object, or the damage resulted in a slow loss of buoyancy, with it sinking, possibly far from the crash site before it came ashore.
This brings me to the forward LG bay door, which apparently also shows a tension failure. Again, some are puzzled. However, if the plane impacted the water inverted as suggested by Victor’s simulation, the cockpit windows would be about the first place to cave in. Water pressure could easily have blown the then upward-facing FLG door off without destroying its buoyancy.
More to follow…
@eukaryote
@George G,
@et al
If their approach was to channel Napoleon – promise everything, deliver nothing – they’re pretty much bang on target.
Note that the first “papers” on the topic were self-published half a decade ago. The first application of the purported methodology to tracking MH370 was just over four years ago. The involvement of the University of Liverpool, through the offices of Professor Simon Maskell, started some three years ago.
If there was any substance whatsoever to applying historical WSPR records to detecting and tracking aircraft at trans-global distances, a couple of competent physics or engineering graduates would have been able to conclusively demonstrate the bona fides of that within a week, perhaps even over a weekend.
Three years of purported university collaboration, and not one independent paper from that university (or any other, for that matter) attesting to the bona fides of the methodology.
The silence is deafening!
@Don Thompson. Yes interesting.
Unlikely but possible, lack of public comment about WSPR does not mean necessarily that there is no ongoing interest in it.
If, unexpectedly, there should be, then that would not necessarily be disclosed.
Even using that to locate the sunken MH370 wreckage might be a no-no.
So lack of comment to imply, of itself, that Maskell has come up with nothing might not be warranted.
@David F: We can be sure that Maskell has “come up with nothing” useful regarding tracking of MH370 with WSPR data because the idea has no merit. If he concludes otherwise, he’s either incompetent or a fraud.
@VictorI. I did not mean to imply that his research would be limited to WSPR.
[Comments here are closed. Continue the discussion under the new article.]