[ August 15, 2015 (revised proposal May 25 2016) — by Ed Anderson ]
Searches in the Indian Ocean have continued for Malaysian flight MH370 since its disappearance on March 8, 2014. Satellite radio communications with the plane in flight have been analyzed in detail, giving a search path tens of kilometers wide on a long arc with a radius of over 5,000 km. The arc is defined by the timing of the seventh and last INMARSAT handshake with the plane. The narrowed search area along this “7th arc” remains large, with ship soundings since mid-2014 focusing on about 1,000 km of the arc, centered some 2,000 km WSW of Perth. This area is based on the expected fuel range of the aircraft and estimates of an autopilot flight bearing based on radar, satellite, and cellphone contact signals at the time the flight went missing.
The Australian Transport Safety Bureau (ATSB) is coordinating search efforts and early on requested analysis of acoustic data for either an impact or implosion that might narrow the search region. The possibility of an underwater implosion after the plane was down has been acknowledged by experts, but there does not appear to be any analysis of available data for that scenario.
Shortly after the loss of the plane, a media release from the UN’s CTBTO nuclear test ban monitoring organization suggested that acoustic location might be possible using their network of sensors, with atmospheric infrasound more likely than underwater hydrophones to detect any surface disturbance. They put out a call for scientific experts in their member states to analyze data, but nothing was initially found due to the difficulty of picking out an event from the environmental noise.
This is a proposal to investigate for underwater implosion sounds and airborne infrasound recording that could narrow the very large search area. It depends on gaining access to the CTBTO recording database. Access depends on a no-fee contract with the CTBTO. This proposal was made in August 2015, but disallowed by the CTBTO because their offices could not establish a primary contract with a private individual, only institutions. A workaround was offered where a qualified institution could establish the CTBTO contract and include a private researcher. The project cannot proceed until someone at a governmental or educational institution is willing to act as a project sponsor, or the awaited change in CTBTO policy comes through.
The CTBTO operates multiple hydrophone listening stations in the Indian Ocean. Each station has at least one triad of microphones spaced 2 km apart, using the different sound arrival times to get a bearing toward an event. These hydrophones are placed at a depth to detect sounds that can propagate long distances within the SOFAR (SOund Fixing And Ranging) channel.
The SOFAR or Deep Sound Channel (DSC) is a layer of the ocean where sound propagates at a local minimum speed because it increases with both temperature and pressure. The effect is like a lens for sound that refracts sounds toward the center of the channel. Similar to gradient index fiber optics that allow light signals to travel long distances, the SOFAR channel allows whales to dive down to the channel depth and communicate across oceans.
About two months after the call for acoustic location, the Australian ATSB search organizers reported that hydrophone data was being analyzed by the CTBTO and IMOS.org.au (Integrated Marine Observing System) as a factor in defining the search area. In early June, Curtin.edu.au put out a press release describing how they physically retrieved an IMOS recorder RCS (PAPCA) near Perth and found a candidate event.
Unlike the CTBTO hydrophone arrays that have cables some 100km long to shore and stream live to a data center, the IMOS devices are remote hydrophones placed to record marine life. They are normally retrieved annually to minimize cost, and to make the battery last they record only about 1/3 of the time on a 15 minute cycle. The Rottnest Continental Shelf (RCS) hydrophone is located on the edge of Perth Canyon. Its depth of about 400 meters near the sea floor is shallower than SOFAR depth, but the slope of the shelf helps to reflect sounds upward from the channel.
The report from the Curtin researchers explains that the timing of the IMOS event allowed a more careful check of the CTBTO hydrophone data from the HA01 array near Cape Leeuwin and a matching event was found. Analysis of arrival time differences among the three HA01 detectors provided a compass bearing to the event, and they report that the Rottnest hydrophone was used to get an approximate range. These hydrophone analysis results were graphed on a map of the Indian Ocean, but the proposed search area was very large, stretching over the Arabic Sea and centered West of India around the Maldives.
That search area does not include the 7th Arc, being assumed as too distant to match the timing of last satellite handshake with MH370. A June 1 letter to the ATSB reported that the acoustic data was incompatible with the generally accepted satellite data. In October, Curtin researchers reported on the analysis of an additional event signal obtained from another IMOS hydrophone retrieved from Scott Reef in September. The projected location was near the Arabian Sea, and researchers concluded that because of the conflict with satellite timing and different wave envelope, it was most likely a natural event. This resulted in news stories that the hydrophones were not useful for locating MH370.
Dismissal of the existing acoustic data appears to be based on both an assumption that the event would have been at the time of an impact, and the way the analysis was presented on the map. The experts addressed the possibility of an implosion, but perhaps in the context of the longer distances. The Executive Summary released as Appendix B of the June 24 ATSB Definition of Underwater Search Areas (p 47) concludes:
“… however, should the arc defined by the handshake data be called into question, the various timing and acoustic considerations discussed here would suggest that a reasonable place to look for the aircraft would be near where the position line defined by a bearing of 301.6° from HA01 crosses the Chagos-Laccadive Ridge, at approximately 2.3°S, 73.7°E. If the source of the detected signals was the aircraft impacting the sea surface then this would most likely have occurred in water depths less than 2000m and where the seabed slopes downwards towards the east or southeast. These considerations could be used to further refine the search area. If, instead, the received sounds were due to debris imploding at depth it is much less certain where along the position line from HA01 this would have occurred.“
In followup comments to the June 4 report from Curtin, a note on July 21 says:
“With regard to the suggestion of using the arrival time and (reasonably well) known speed of sound to estimate the range, we have done this based on the earliest and latest times the plane is likely to have hit the water, and the results are consistent with the uncertainty box derived from the arrival time differences at Rottnest and Cape Leeuwin. Unfortunately the speed of sound is large (about 1480 m/s) and the uncertainty in the impact time is also large (more than half an hour), so this doesn’t reduce the size of the uncertainty box by much. It also assumes that the sound we heard was the actual impact. It could also have been the implosion of sinking wreckage, but that would have been expected to occur some time later, which would confound this calculation.”
The confounding factors and the probable depth are apparently references to the efficient propagation of sounds only after they enter the SOFAR channel. Surface sounds, like from an impact, propagate mostly downward passing through the channel while portions of the sound traveling more horizontally are refracted back upward toward the surface by the speed gradient. This has a complex effect of causing convergence zones of strong and weak nodes along the surface with a larger attenuation of sound and only a small portion entering the SOFAR channel, plus multi-path ghosting of the signal. A similar effect occurs from an event on the sea floor below the SOFAR channel. Nearly horizontal signals are refracted back downward and attenuated.
The difficulty of sounds entering the SOFAR channel from above or below may be why atmospheric infrasound was initially suggested as a more likely avenue of investigation than hydroacoustics.
SOFAR channel propagation can also be disturbed by variations in the sea floor. It can be completely blocked by a seamount or ridge that nears the surface. Temperature variations along the path can also change the wave speed. Such variables can also cause horizontal refraction of the signal, causing it to take a different path to the hydrophone, or show a false bearing.
These effects may partly explain why the event was not detected at the other CTBTO hydrophone array in the Indian Ocean, HA08 in the British Indian Ocean Territory, South of the Maldives. That hydrophone array was experiencing local noise from a sounding source, but there may still have been windows of detection.
More importantly, the known signal analysis was done for a significant surface impact event. The one candidate event analyzed was initially missed by CTBTO detection methods but later confirmed as a strong signal in their dataset. Only a range of time that might coincide with the 7th Arc was considered. Los Alamos National Labs (LANL) did an excellent 2D plot of hydrophone noise levels along the 7th Arc but restricted to the time range of impact plus only about 1 hour, just short of the event detected by Curtin.
The time window for implosion events is much longer, as it is dependent on how fast any components might sink to crush depth. Any noises from buckling or crushed metal could propagate directly into the SOFAR sound channel at crush depth. A key ocean hydroacoustics researcher and author of CTBTO monitoring handbooks has confirmed that, “Certainly, signals from equipment crushed at depth would be far more likely to generate long-range signals than the initial impact.”
Many modern signal analysis methods are available for deciphering signals from noise, and multiple recordings improve that ability. It should be apparent that acoustic methods for narrowing the search have been prematurely dismissed by the news media based on a single event being classified as seismic.
Still, a single candidate event was detected by five hydrophones at three widely separated locations. Using the fourth Rottnest hydrophone not only provides a range, but another triangulation method. This shows the capability of improving accuracy beyond that available from a single array, but the CTBTO holds the only known continuous recordings of MH370 environmental acoustics.
The much longer baseline of 343 km between the RCS hydrophone and the center of HA01 provide a more accurate triangulation than the shorter 2 km spacing of the HA01 hydrophone array. With the event time unknown, the near range acoustic search path defined is not a straight bearing, but a shallow arc that bends toward the closest hydrophone. This is not indicated in the earlier analysis, which clearly shows a straight great circle path ending directly at HA01, the furthest hydrophone. No explanation was given for the shape of the Curtin search area, but the Southern limit does appear to start on the Southern error limit of the HA01 bearing and arc toward RCS before being truncated where it intersects the Northern error limit of the HA01 bearing. Had that full arc been drawn, it would be seen that this particular triangulation path does cross the “7th arc” search area.
The time difference between signal arrivals is the key element. Without access to the original acoustic data, it can be visually cross correlated from the graphic display of the two sound envelopes provided by the researchers. An optimistic estimate is 66.6 seconds, with an accuracy of about a second. The researchers say the uncertainty is about four seconds, which could be taking into account long term timing accuracy of the clock on the underwater Rottnest recorder. [8/26 update – IMOS PAPCA is checked for clock drift against GPS timing at deployment and recovery.]
This time difference can be turned into a path length difference given the local speed of sound in water along each path. The speed of sound in the SOFAR channel varies mainly by locality and seasonal water temperature change, but not over a very large range. Different researchers have estimated at 1460 m/s (used recently by LANL), around 1472 m/s on a bearing of 301.6 to the Arabic Sea, 1480 m/s in a comment from Curtin, and past papers with maps showing computed SOFAR sound speeds for the Indian Ocean.
Using an estimated 1488 meters/second, the path length difference from this particular event to each sensor is approximately 99.1 km. [Dr Alec Duncan at Curtin later confirmed and revised the SOFAR sound speed to more accurate values of 1487.2 for HA01 and 1487.4 for RCS paths.]
An October 2015 version of this proposal previously put a stronger emphasis on the accuracy of the broader baseline triangulation, and had a calculation error using a fifth hydrophone at Scott Reef. That caused a false premise for a very narrow search area crossing the 7th Arc 60 nm East of the Batavia Sea Mount.A careful recalculation using the current method finds that the Scott Reef timing is very consistent with the HA01 triad bearing of 301.6 and the HA01+RCS triangulation. The event appears as the Curtin researchers deduced on the Calsberg Ridge West of the Maldives. The new accuracy places the event within 50 km of 4.65N 66.62E, and thus probably seismic. The .kml file linked below shows the event area.
Improvements in accuracy were found by obtaining revised source locations for the hydrophones. If other candidate signals are found using a detailed analysis of more CTBTO data, further improvements could narrow the acoustic search path to within 5-10 km. New candidate events have been found using additional sparse recordings provided by IMOS, but they need confirmation using the CTBTO hydrophone data.
The width of an acoustic search path depends on an accumulation of errors in the variables. An error in the estimated 1488 m/s does not create a large deviation from the path because the seafloor topology is very similar for both paths. Substituting the slowest 1460 m/s speed causes an error of about 5 nm SW perpendicular to the path. Is is unlikely that the SOFAR speed is much higher than the chosen estimate. The commonly published locations of the hydrophones are not accurate. The HA01 hydrophones are located about 4 km ENE of the 34.9S 114.1E listing. The center of the triad at 34.890S 114.143E is used, assuming that the three separate signals were correlated and summed for noise reduction. The 31.885S 15.0195E location used for the Rottnest RCS hydrophone were fairly accurate, based on a map showing placement of the hydrophone after retrieval from 2009-2013. Revised locations for three hydrophones along with their recordings have been provided by IMOS. (The Scott Reef IMOS hydrophone was located 171 km SSW of the Scott Reef report used previously.) Hydrophone coordinates are probably within 1 km.
The width also is defined by the error limit of the calculations, with the highest probability toward the center. That error range can be computed by figuring the shortest and longest likely path differences. Since the path difference is a product of wave speed and time, errors in both are easily summed. Sound speed values of 1460 and 1514 m/s are used for the minima and maxima (though the value of 1488 is probably within 2 m/s). A one second uncertainty has been estimated for the signal timing. It is hoped that the hydrophone signals could be matched to within milliseconds (as within the HA01 triad). An additional 4 km of error has been added to account for errors in hydrophone placement and sound speed variation between the paths. The result is a search path that is about 23 nm wide (42 km or 26 miles). Even with conservative estimates of the errors, this search area is rather narrow because it is much closer to the hydrophones than previously assumed, and the baseline is 160 times longer. The acoustic search path error is still relatively small compared to the 200 nm breadth of the 7th Arc search zone, and tiny compared to the length along it being searched since October 2014.
The errors described can potentially be reduced to within the width of a single sonar pass by experimental post-calibration of the data. Sound speed over the path could be verified by detonation of more easily detectable test charges at a calibrated location and time anywhere near the 7th Arc search path intersection. A pattern of strong sound impulses at SOFAR depth could calibrate all hydrophones in the area to narrow the signal observation window. The accurate bearing to the source could be used for filtering of various acoustic signals to reduce noise. For the RCS and Scott Reef hydrophones to record the test signal, it would need to be timed for reception when they are both active, then they could be retrieved once more. A hydrophone could be lowered near the Rottnest location during the test. Surface and seafloor detonations could be useful in characterizing the signal so that other MH370 acoustic events might be found. A surface explosion would also be detectable by atmospheric infrasound.
Given the lack of results from the current search area, some speculative possibilities arise. The Northern portion of the 7th Arc has not been fully explored. Early ULB ping detections were discounted. Part of the reason for dismissal of those signals was that they were heard beyond the likely range of the ULBs on the recorders. If the resting place of the plane were on the NE skirt of the Batavia seamount, that location might have a reflective focusing effect on the sound waves toward the NE. While a location NW of the 7th Arc intersect seems slightly early, it is possible that the plane was already down by the time of the partial handshakes, or that it spiraled back along its path after a loss of power.It is possible that the disabled aircraft was directed by the pilots toward potential landing sites. The plane may have missed Langkawi, passing directly over it. A later attempt may have been to set a bearing for Cocos Island airport which has a 10,000 ft runway.
Cocos Island Infrasound Recordings
Cocos Island also hosts an active CTBTO infrasound recording array. Even if the initial bearing was accurate, variations in the great circle path vs compass bearing may have caused the plane to miss the island without correction. This may be significant because although the distance from the island may have been too great for any late night sightings, the CTBTO infrasound listening station IS06 would have recorded the passing of MH370. The assumption so far has been that the signal was lost in the noise, but the event detection threshold may have been arbitrarily high looking for an impact impulse rather than a subtle doppler shift in a broadband noise source.
The location of Cocos Island between the 4th and 5th arcs (21:41:27 and 22:41:22) gives an interpolated flight path time there of 22:19:49 UTC. If the path were 80 km East of Cocos, to coincide with the earliest Northern search area, the closest approach for doppler drop would be at 22:27:29 UTC. Flight paths within 300-500km West of Cocos place it in the Southern search region. This creates a broad infrasound listening window from 21:30 to 23:00 UTC. Speeding up the recorded signals to the audible range and listening binaurally to the multiple recorders in the array, especially with additional beamforming techniques, could reveal a distance, general flight bearing, and approximate speed if any doppler shift is detectable.
It is hoped that any researchers or organizations involved will take this analytical critique in the constructive way it is intended, using it to refine the available information.
Above all, heartfelt condolences go to the families and friends of those presumed lost on flight MH370. Please excuse any insensitive wording.
Beware that the embedded map above and mobile devices may show rhumb lines rather than great circle paths. Please use Google Earth on a desktop computer to load the .kml file, which contains additional notes.