[Revised Aug 10, 2016]
Four candidate MH370 search locations are presented, based on the detection of unidentified hydroacoustic events consistent with the expected flight path. These are preliminary results found by searching for impact and late underwater implosion events using multiple hydrophones and new analysis methods.
The focus of this investigation has been for implosion events in the SOFAR Deep Sound Channel where sounds can propagate across oceans with little attenuation. Surface impact sounds tend to stay near the surface and dissipate. Only the portion of the signal that entered the SOFAR channel at a depth of about 1000m is likely to be detected. Finding an implosion event can point to the weaker impact signal, which would further narrow the event search region.
A .kml map file linked at the end of this report helps to clarify the locations and bearings described here, with additional notes.
A Prominent Signal
The first detection was for a candidate implosion event 5 hours after the last contact with the plane on Mar 8 2014 at 00:19:37 GMT. It is plainly visible on spectrograms from hydrophones at three separate locations, the strongest signal for several hours. The event had previously been dismissed because its directional bearing from the CTBTO Cape Leeuwin H01 hydrophone array was typical of ice shelf noise from the Antarctic coast. The later signal arrival at the H08 array in Diego Garcia also points toward Antarctica. Only by crossing the different hydrophone bearings and comparing arrival times did the event stand out as different from typical ice events. It is over 1000 km north of Antarctica, but some 2000 km south of the current search area.
The flight path to this location is consistent with satellite timing and passes over the center of the sonar search area along the 7th arc. The flight bearing would have been nearly due south along the 90th meridian. (Saved locations from the pilot’s home flight simulator separately indicate a 90th meridian waypoint and a heading due south from where the plane deviated from its flight plan. This is not meant to imply malicious intent and my simply be coincidence – each of the saved locations was due south of some key turn along the path.)
At 1000 km/h (Mach 0.81) and 2320 km from the 7th arc, the estimated impact time would have been 02:38:49. The additional fuel required for 2.5 hours of flight does not match fueling documents, but is not entirely inconsistent with further analysis of ACARS reported fuel burn rate during initial climb. This location also does not match modeling of surface drift patterns, where flotsam from this far Southern location travels eastward.
The best alternate explanation for this event would be an iceberg. It is well north of usual iceberg range, but infrared satellite images provided tracking of a large 20km iceberg that was 350 km SSW of the candidate site. It’s possible that a fragment of that large iceberg may have been the source, but its signature is different from earlier (pre-impact) sounds coming from the bearing of the known iceberg. No iceberg was seen near the location, but might have been obscured by cloud cover. Ice tracking databases should have more info on icebergs as small as 15 meters.
The details of the strong signal are an H01 arrival at 04:59:20 GMT from bearing 210.3. The matched H08 bearing 166.8 arrival was at 05:28:53 GMT.
The event was detected on a IMOS Rottnest single hydrophone 3315_PerthCanyon at 05:03:02 GMT. The arrival at three different hydrophone locations allows a long baseline triangulation that is independent of the triad bearings. The bearings and triangulation converge on a location currently estimated at S 55.6° E 90.65° within about 50 km in an ellipse along a narrower search path. The location error is mainly dependent on hydrophone placement accuracy and sound velocity estimates, which can both be improved with further calibration.
Update: Additional detections have since been found for this first candidate location. An earlier strong H01 4:38:42 event has a matching H08 arrival at 05:08:30, both on the correct bearings. An H01 arrival at 4:42:08 also matches the 210.3 bearing. An 03:35:33 matching impact candidate is detailed below that would end the flight at 03:02:54 GMT.
Searching for Implosions
A standard approach to measuring the delay between two acoustic signals is to use cross correlation with a time window wide enough to encompass both signals. The peak of the correlation result typically matches the delay between the strongest signals within the window. This works well at low frequencies with strong signals but weaker signals can get masked.
Curtin.edu.au carefully analyzed 8 hours of H01 hydrophone signals in 2015 looking for implosion events with no significant candidates. Their algorithm improved on the standard cross correlation by checking not just the strongest correlation peaks but a selectable number of lesser peaks within a broad sliding window. They also analyzed 3 hours of H08 recordings, with results masked by two different ongoing seismic surveys during that time.
Some signals visibly clear enough on spectrograms for manual alignment were not detected, while solid correlations over several seconds were found for nearly invisible ice signals. The undetected signals were mostly short but dispersed signals that lacked low frequency components. Courtesy of Curtin researcher Alec Duncan, their algorithm was tried with various settings in an attempt to get more accurate bearings on the visible signals. By relaxing the arrival wavespeed restriction slightly, a signal towards the 7th arc became visible from H01 at 03:30:00 on bearing 249.5 degrees – a low frequency signal barely visible on the spectrogram. Applying the same filter to H08 revealed another late signal at 05:20:30 from bearing 162.65 towards the 7th arc search area.
A Second Detection
The surprising result of plotting both new bearings on a map is that they converged precisely at the 7th arc. The candidate location is S 39.725° E 85.00° within 3 km of the 7th Arc. That accuracy is beyond the precision of the supporting data. The error ellipse would be slightly broader along the 7th arc, with a typical half degree of accumulated bearing error shifting the location 30km, which could again be improved by calibration.
Finding a matching impact signal for the separate but matched implosion candidates would narrow the search region by estimation of the propagation time at a predictable speed of sound.
With some clear signals evading classic cross-correlation, attempts were made to plot the signals using a coherence measure. A previously successful technique of presenting the three channel pair spectrograms as Red Green Blue color intensity components showed matches intuitively as white. The RGB display of coherence was partly successful, revealing that alignment of the signal timing was key to refining the displayed result. In a manual two pass approach, the arrival delays that define the bearing of a signal are measured, then applied to align the signals for cross correlation and display against time.
The correlation results are also offset to correct for their known timing differences in the vertical axis. With the timing aligned, a very narrow correlation window can be used to plot the signal matches across all three channels. The bare correlation results would appear as dots containing alternating correlation-anticorrelation peaks. Passing them through a hilbert transform filter to fold sin+cos components into a measure of envelope energy, a smooth plot of the signal energy over time is presented, with best matches devoid of color.
This method is equivalent to a narrow beamformed cross correlation. An example is shown for the strong Southern signal first described. Precise timing details can be discerned, and fine structure is visible in portions of the signal with minimal lowpass filtering. The signal also shows a dispersion in apparent direction (or velocity) at peaks. The high frequency components are clearly arriving in unison across the widely spaced hydrophones. The peak spreading may be why standard correlation is not catching high frequency signals among lower frequency masking events.
With accurate delay timing for a bearing, the comparison window can be as narrow as a single sample. The hydrophone placement is not that exact, but the beamformed correlation concept can be extended to plot bearing vs time for signal analysis. The delays for each plot bearing become sine waves defined by the hydrophone spacing and orientation.
By aligning the timing at successive bearings and summing the product of two narrow signal windows, a useful scan for signal coherence can be made. Each hydrophone pair will independently match a signal, but with a matching peak in the opposite quadrant. When they all align at one bearing to make a monochrome result, the incoming signal has good directional coherence. Ambiguity can visually be resolved by following along the component color traces to check for mixing from a stronger signal.
These two signal visualization tools have been used to tentatively verify the candidate implosion events that cross bearings on the 7th arc.
Event signals matching the H08 bearing 162.65 also arrived there at 4:43:27 and 5:40:00 GMT.
A Third Candidate Location
Scanning the arrival time range for impacts near the current search area revealed another candidate location, an H01 arrival at 00:39:18 from bearing 271.73 on the 7th Arc at S 32.97° E 95.40°. The sound propagation time over 1748 km at a profiled wave speed of 1.484 km/s is gives an estimated impact time of 00:19:40 GMT. That is 3 seconds after the last partial ping, which would place the plane with 5 km of the 7th arc – approximately within the error range of the geodesic calculations.
A Fourth Candidate Location
Closer examination of the impact timing range has now revealed a fourth impact candidate near the 7th arc. The H01 arrival at 00:41:27 is a clear signal on bearing 269.5 that crosses the 7th arc at S 33.48 E 94.75 but 103 seconds later than expected for 7th arc timing. While that timing broadens the search range along bearing 269.5, it is consistent with some end-of-flight scenarios.
Another Match for the First Candidate
A matching impact candidate for the far Southern event on bearing 210.3 source has been detected arriving H01 at 03:35:33 GMT. A distance to H01 of 2920 km and a profiled wave speed of 1.4754 km/s gives a propagation time of 32:29 for an impact estimate of 3:02:54 GMT. This is just nine minutes before the M.81 flight duration estimate to the location described for the first candidate. It not characteristic of ice events. It has an early component that spreads the energy toward the south, but the bulk of the signal is centered at 10 Hz around bearing 210.3 degrees from H01. The correlation window in the following figure has been widened to 2 seconds to show the central bearing more clearly.
This correlation plot shows the signal arriving with the same bearing alignment used for the 04:59:20 signal correlation plot above. The bearing is the same within the sample resolution of 1/250 of a second.
Comparison to earlier analysis
The October 2014 LANL LA-UR-14-28179 report on 7th Arc noise sources uses a similar method for aligning the hydrophone signals and measuring energy content, with a mathematical description. The scan method used here treats the pairs separately, presenting the signal energy in RGB color channels to graphically check alignment and intensity. On a second pass the candidate signal is then examined more closely by plotting the cross-correlation using a narrow window.
The LANL report also identifies three main clusters of signals along the arc. They mapped their results filtered by propagation time, but a minor note is that a value of 1.466 km/s was used for ranging. That is the perceived arrival speed at H01, which is more an artifact of the hydrophone placement. A derivation of wavespeed by profiling along the path as described in an earlier report here provides a more accurate group velocity value of 1.484 km/s. This would shift their search map about 30 km, though they do describe analyzing beyond that range.
Of the three LANL report clusters, the largest is their first candidate signal, which has been carefully examined using the new plotting tools, finding subcomponent signals from different bearings but not at the reported 247 degrees. Their second and third candidate signal clusters appear on the LANL figure 2 & 3 map near the 7th arc candidate locations presented here, but the timing is different.
A Hydrophone Data and MH370 meta analysis by Richard Godfrey was published on Aug 1. It is based on the LANL noise report described above, and essentially assigns bearings with lat/lon ranges to the arrival times of LANL cluster 3.
Taking a closer look at the LANL Figure 4 timing as referred to by Godfrey, the three signal clusters have been carefully examined and excluded. The main cluster 1 at 00:52:00 is consistent with an ice event on bearing 190, cluster 2 at 00:49:41 is on the iceberg bearing 206, and cluster 3 at 00:39:40 is at bearing 196.
Since impact events are not expected to couple well into the SOFAR channel, the new tools are being used to more carefully scan several hours of recordings for candidate implosion events. A scan of earlier data could also check for pre-impact iceberg events on the 210.3 bearing. Several early events at 205-209 have already been detected, which would be from the Antarctic coast or the large iceberg at bearing 207.
Other hydrophones were active in March 2014. Requests have been submitted for data from the French OHA-SIS-BIO hydrophones which are much closer to the search area. Their SWAMS Amsterdam location is a hydrophone triad array right on the 7th Arc, about 900 km SW of the implosion signal candidate. The NEAMS is a single hydrophone about 800 km NW and perpendicular to the 7th Arc active search area. Their S-SEIR is a single hydrophone further West of the search area and could be used for triangulation. Another OHASISBIO-WKER2 hydrophone triad lies 1100 km beyond the SWAMS triad but could provide additional confirmation. The hydrophone data was retrieved in early 2015. Any assistance in access to the recordings would be appreciated.
The CTBTO Cocos Island infrasound has yet to be fully explored, and any detection could help narrow the search area.
Many thanks to Alec Duncan at Curtin.edu.au for his patience, help, and sharing of his expert knowledge on hydroacoustics. Thanks also to Tom Kunkle of LANL for his helpful advice.
Sincere condolences go to the friends and families of those lost on flight MH370. This effort is for you. Please forgive any insensitive language in delivering the analysis. The acoustic search will continue until the plane is found.
— Ed Anderson in San Francisco