A Strong Anomalous Acoustic Event on the Seventh Arc near Java

A very strong MH370 candidate signal was reported here a year ago, arriving at the Diego Garcia H08 hydrophone array from the direction of Java. The T-wave arrival is far stronger than a later M4.4 quake near Java and four times stronger on local seismometers than a nearby M4.1 quake. Despite the stronger signals, this event was not included in earthquake catalogs like the others. Analysis of nearby seismometers places the origin as 8.36S 107.92E directly on the 7th Arc at 1:15:18 UTC, almost an hour after the expected impact time. This may be evidence of an implosion near SOFAR channel depth as the plane sank after impact.

Overview Map of the Indian Ocean

This map is a one minute frame out of a sequence of 19 hours. It is designed to be a cartographic overlay for use in Google Earth. (Diego Garcia at top left, Cape Leeuwin middle right. Edges are on the degree boundaries)
01:16:00 Overlay map of an acoustic event near Java arriving at the Diego Garcia H08 hydrophone array.

Spectrogram of the Java Anomaly arrival at H08 amid repeating airgun noise.

Analysis

This event first stands out prominently on analysis of the H08 hydrophone data at Diego Garcia, nearly 4000 km distant.

The signal arriving on bearing 92.1 at H08 at 01:59:20 UTC despite the bearing 113 airgun blasts.

A matching weak arrival and bearing on the H01 hydrophone array at Cape Leeuwin gave an approximate origin time of 1:16:00 and an initial location estimate farther west. It was assumed to be a quake because it has a stronger T-Wave arrival than a reference M4.4 quake at 08:50:55 later along the Sumatra coast. Another nearby reference M4.1 quake off Java earlier at 23:52:18 is barely detectable at H08 due to survey ship sounding noise.

Regional seismometers were checked for arrivals of the event.  The AU.XMIS station on Christmas Island is about the same distance from the candidate event as the 23:52:18 reference M4.1 quake, but the candidate is four times stronger. An estimate of the event scale based on reference quakes would be close to 4.5 magnitude.

Seismic data networks catalog most events magnitude 2.5 or greater.  This event has more impulsive high frequency components than a typical quake, and high frequencies do not propagate as well. The limited number of close-range seismometers might not have been sufficient for a robust catalog record. A seismic analysis plot of the arrival at nearest seismometer GE.CISI (Cisindang, Java) reveals some details of the event.

Seismic arrival at the nearest seismometer GE.CISI on the central Java coast.

Several seismometers that registered the signal were marked for P and S-wave arrivals. By using a Wadati plot of the phase time differences, the origin time is estimated graphically, with local variations in propagation speed evident across the Java Trench, and possibly visual phase picking errors.

Wadati plot of P-wave and S-wave arrival time differences to estimate the origin time.

In a Wadatai plot, the event origin time is at the zero time difference crossing on the Y axis origin, and the slope shows the Vp/Vs velocity ratio that is characteristic of the region (Vp/Vs = 2.182).

The location estimate was determined by running an iterative linear optimization to minimize the least squares Tau-P time estimates for each seismometer distance, with weighting toward the closer stations. A 1.7 second delay was added to allow for the slower propagation through 2650 m of water to the seafloor. S-waves do not propagate in fluids, so it is assumed that the seabed directly below the event is the origin of the seismic S-waves.

StationGE.CISIGE.UGMGE.SMRIAU.XMISGE.MNAIII.COCO
Distance (km)89.752290.848311.080342.319705.3281286.065
error (km)0.619-1.671-0.1188.054-31.241-64.543

Geodesic distances were computed using GeographicLib in MATLAB for good precision. Standard software used by seismologists that takes into account known local variations in wave speed could give a better estimate than using global TauP values. Using more seismometers would also improve accuracy. Stacking of the 50 RESIF YR.ME array stations in central Java might be possible.

Candidate event detail map

Discussion

This late signal arrival fits the hypothesis that components of the plane may have imploded or made buckling sounds as they sank. Even weak sounds emitted near the 1000 m depth SOFAR deep sound channel could be detectable.  The tires would pop from the rims at a shallower depth. The aircraft had eight titanium fire suppression spheres and other sealed containers on board that may implode under intense pressure. Whatever the source, it must have made its way into the SOFAR channel to be far stronger at Diego Garcia than the nearby catalogued quakes.

Further evidence for an event near the SOFAR channel can be seen in the details of the signal arrival at H08. An updated “eigen” algorithm plot (mid-2018) uses the product of inter-channel covariance and the eigenvalue ratios from PCA analysis. The improvement in resolution over the simpler covariance method above is significant.

Java Anomaly arrival at H08, with background airgun noise.

Note that the background airgun blasts do compete with the display of the signal coherence, taking notches out of it. Also note that the airgun bursts every 8 seconds are consistently from the same bearing 113.1, while the bearing of the signal appears to increase past 92 degrees over a brief time. While the absolute bearing accuracy is not precise (off by about 0.7 degrees here) and may vary for different targets due to horizontal refraction, the relative accuracy for a given source should have little error.

By oversampling the signal, an increase in the bearing discrimination is possible for analyzing the shift using the following figure.

Java Anomaly arrival at H08, bearing shift detail.

The variation over 45 seconds of signal shown is from 91.75 to 92.50 degrees, with a peak around 92.25 (actual geodesic 92.95) degrees. This shift of 0.75 degrees corresponds to a north-south distance of roughly 50 km near the origin. It is speculated that the spreading over time of the signal and its bearing shift is caused by the progressive reflection of the acoustic energy along the coastal shelf of Java, where the signal is conducted into the SOFAR channel.

The previous report on detection of lightning strikes shows that impulse surface events over deep water do not propagate well into the SOFAR channel and are difficult to detect at even short distances. Sounds that get reflected into the SOFAR channel, such as the airgun blasts over shallower water, propagate very well. It is therefore unlikely that the sound entered the SOFAR channel on a direct path to H08 from the origin.

The sloping canyons of the coastline may have had a focusing effect. The T-Wave conversion mechanism was discussed previously where SOFAR sounds are concentrated by upslope reflections of increasing angles between seabed and surface until crescendo at a depth of around 200m. Seismic events are also converted downslope into the SOFAR channel. Proposed here is an upslope conversion reflected back downslope. As the origin is only 50 km from the shelf, it is likely that an implosion event outside the SOFAR channel was able to be focused upslope along the shore to 200m depth, and those reflecting focal points then became secondary sources for downslope propagation into the 1000m SOFAR channel. The time period of H08 arrival and bearing shift are a good match for reflections off the Java coastline.

The map elements shown here may be downloaded as a Google Earth map overlay file containing detailed locations and notes:
Java-Anomaly-Map.kmz


2019 Improved Plot

By filtering the covariances using SVD derived eigenvalues, and oversampling at 8x, a higher resolution plot is obtained for the H08 arrival of the Java Anomaly signal. Incorporating very low frequencies broadens the peaks, but shows weaker signals. The earliest direct path to the event would be on geodesic bearing 93.677 degrees.

Revised bearing plot of the Java Anomaly arrival at H08 showing enhanced detail.

The Java anomaly is presumably from a late implosion. Knowing the precise bearing from H08 and the propagation time, it is possible to look more closely for signs of an impact at the same site with 7th arc timing at 00:19:37 UTC.

Matching H08 scan for a Java site impact time of 00:19:37 UTC.

The scan above covers the same time span and bearing range. The strongest peak was expected at 01:03:39 at H08. There is a significant arrival at approximately the right bearing just three seconds later. This would likely be a reflection from the coastline.

The first arrivals of the stronger Java anomaly are about 16 seconds before the peak. Just 16 seconds before the 1:03:42 peak there is a signal on the expected bearing of 92 degrees. Buried in the noise are hints of a weak unexplained signal that spreads linearly at bearings both higher and lower than the center. A shock wave spreading from an impact and creating secondary wave sources with the very sensitive SOFAR channel might explain the faint pattern.

April 2019 Update

In evaluating the error bounds for the Java anomaly location, new details became evident that might significantly shift the site. The early 2017 location estimate of the Java Anomaly used an iterative method based on an assumption that the P and S wave arrivals came from directly below the origin. Later work with the Wadati plot applied T-wave conversion point timing estimates at the seismic receiver sites, but still on the assumption of the seismic epicenter being at the origin. As the hypothesis evolved of secondary T-waves as a source for the strong detection at Diego Garcia, it is now realized that the strongest seismic epicenter may be at the secondary T-wave conversion points, not at the origin. The conversion points may therefore be particular to the direction of each seismometer from the origin, and the S-waves may come from further downslope at SOFAR depth rather than the 200m conversion point depth. It should be noted that steeper slopes generate stronger P-waves, while a long slope may only show S-waves.

Polarization plot of the Java Anomaly arrival at seismometer GE.CISI

Evidence for this is can be seen by examining the nearest seismometer GE.CISI which has the strongest arrival for the Java Anomaly. That strong P and S wave signal obscured weaker arrivals in a 30 second preamble before the main impulse. This is revealed by comparing a spectrogram of the analytic signal phase (ignoring amplitude, similar to full compression) with the timing of the strong wave arrivals.

It is now speculated that the sound from an implosion was not necessarily near the SOFAR channel depth, but close enough to the coastal slope that the acoustic waves were still concentrated by upslope conversion. The weak early arrivals would then be P-waves and S-waves from interaction with the seabed between the origin and the coast. This puts the actual time of the implosion up to a minute earlier than the Wadati plot estimate based on T-wave conversions.

An autocorrelation plot is helpful for sorting out S-wave and P-wave phases. In the case of this signal, the P-waves do not autocorrelate well with following portions of the signal, while S-waves do match up.

AutoCorrelation plot of the Java Anomaly arrival at seismometer GE.CISI

The polarization analysis portion of the plot is also useful in sorting out the S and P waves by the arrival azimuth. The S-wave azimuths would generally follow the angle of the coastline, while the P-wave would be the azimuth from the seismometer. A quick transition in the azimuth indicates a change in the dominant wave phase. From the autocorrelation and azimuth details, it can be seen that the GE.CISI arrivals are:

Ptime  = 2014-03-08 01:14:50.0 UTC   earliest P-waves
Stime  = 2014-03-08 01:15:05.0 UTC   earliest S-waves
TPtime = 2014-03-08 01:15:35.6 UTC   strong T-wave converted to P-wave
TStime = 2014-03-08 01:15:43.3 UTC   strong T-wave converted to S-wave

Reading from the polarization plot, the GE.CISI converted T-P-wave arrival would be from about bearing 140, which intersects with a 200m coastal depth at a conversion point. On a coast with canyons, the conversion would generally be strongest where the slope is square to the arriving wavefront. The strongest S-wave polarization angle of about 20 degrees from north should point toward the origin from the conversion point. One estimate is S8.0 E108.2 for a T-wave to P-wave conversion point. S-waves would be converted further down slope.

The early wave arrivals should allow an estimate of the distance from GE.CISI to the origin. Using known timing to the additional conversion point, it would seem that triangulation becomes a basic algebra problem. To allow for fine tuning such as using geodesics, depth delay and different weighting, finding the origin lat-lon was coded as an iterative linear optimization. The result turned out to be very sensitive to the conversion point location, where tiny changes would move the origin west by many kilometers.

A standard distance calculation using the S-P wave arrival difference gives a much more stable result, though it is still sensitive to the estimated local wave speeds:

Pspeed = 6.343;  % local estimate. The global average TauP P-wave velocity for under 150 km is 5.800 km/s
Sspeed = 2.907;  % local estimate. The global average TauP S-wave velocity for under 150 km is 3.360 km/s
Tspeed = 1.4954; % profiled from WOA data

The local P-wave speed of 6.343 km/s was taken from an iris.edu GE.UGM station analysis as nearest currently active station, since GE.CISI is no longer operational. That analysis also gives a local Vp/Vs velocity ratio of 2.10 to derive the local S-wave speed. The Wadati plot above gives a measured local ratio of 2.182 on this signal, which give an S-wave speed of 2.907 km/s. The resulting distance from (Ts-Tp)/(1/Vp-1/Vs) = 80.5 km from GE.CISI. The intersection of that radius and the 7th arc would move the Java Anomaly origin about 9.7 km north along the arc. To test the sensitivity to local speeds, the same distance formula was run with the global average P and S wave speeds. This gave a distance of 119.8 km which would instead move the Java Anomaly origin south 31.0 km. Using local speed estimates is more consistent with the result from a dozen seismometers, but it shows that small shifts in the timing estimates create a large error circle.

The Java Anomaly as a Quake vs Implosion

An expert has suggested that the progressive shift in the H08 bearing on the Java Anomaly signal is common for a SOFAR detection of a long earthquake fault slip, where the propagation rate of the fault rupture might match 850 m/s at this distance. Since the magnitude is related to the volume of mass movement, a long fault slip implies a large magnitude quake, and a prolonged seismic event at multiple seismometers. Conversely, this event was not cataloged, which implies that it had less than 2.5 magnitude, while nearby magnitude 4+ were either undetected or barely detectable in the SOFAR channel. The event also had a very sharp onset on both hydrophones and seismometers, unlike most quakes. Lastly, the intensity dropped off rapidly with distance, while even moderate M3.0 quakes are typically picked up by many seismometers.

Characterizing an Underwater Implosion

A reference pattern has been sought that might help distinguish an implosion in the SOFAR channel from various other possibilities. Creating reference events by imploding glass spheres for comparison has been discussed, but requires considerable resources. A rare known implosion comes by way of another tragic loss at sea. The Argentinian submarine vessel ARA San Juan went missing on November 15, 2017. A coordinated search over a wide area found no trace of the submarine. On November 23, news came that the CTBTO detected an event off the Argentinian coast in the right time frame using two distant hydrophone arrays at mid-Atlantic Ascension Island and Crozet Island between Madagascar and Antarctica. The bearings and timing of the signals gave an estimate of a November 15 13:51 UTC event at 46.12S 59.69W in Almirante Brown Canyon with a depth there of about 900 meters. By November 30th, the search area was reduced to a radius of 40km with eight survey ships, and 68% of that area already scanned.

It was not until a year after the event that the wreck was found on November 16, 2018. It was found by Seabed Constructor of the firm Ocean Infinity, using the same AUV technology they used to scan the 7th Arc for MH370. They pinpointed the location 907 meters deep at 45.949722S -59.772778W, just 20 km from the CTBTO estimated location.

The CTBTO press release has a spectrogram of the 14:50:00 signal arrival at the Ascension H10S array, which shows three signal peaks over several minutes. Fortunately, that hydrophone data has been released to the public IRIS network. This allows a close look:

First signal arrival at H10S Ascension Island from ARA San Juan implosion.

The strongest arrival is the the first signal peak, shown here as an energy spectrogram, with hydrophones 1,2,3 displayed as color component R,G,B. The first thing to notice is that hydrophone 1 on the red channel behaves differently than the others, giving a cyan color cast to the signals. Instrument profile corrections might correct the frequency response difference. In this plot, the energy at each frequency is scaled to the median over time, giving a corrected flat background noise level. This corrects for the large amount of low frequency noise below 3 Hz, and should be equivalent to a frequency response correction. Inspection of the raw data indicates a possible sampling threshold problem on hydrophone 1.

The signal does have some characteristics that might be a signature pattern for an impulse in or near the SOFAR channel. There is no visible frequency dispersion as seen in Antarctic ice events. Those upswept chirps may be caused by extra surface reflections at higher frequencies when the SOFAR channel is at a very shallow depth. Nearby impulsive signals or seismic events outside the SOFAR channel have sharp onset and slow taper. This signal is reversed, where a long preamble crescendos to a peak, with little trailing coda. Another event that matches this signature is the 04:59:20 Southern Event candidate that was the strongest signal on March 8, 2014. It probably came from a large ice floe cracking, but far enough north where the SOFAR channel is hundreds of meters in depth.

An autocorrelation plot better reveals which portions of the signal are similar, which may be part of a SOFAR implosion signature:

Autocorrelation plot of the first signal arrival at H10S.

By comparing the timing marks, the preamble has the strongest self-similarity. The peak matches a weak signal that follows 3 seconds later, which may be a reflection. The weak signal at 14:58:15 matches the preamble alone, and not the peak, as does an even weaker earliest arrival at 14:57:53 UTC.

Bearing Shift from Reflections

The Java Anomaly signal does not arrive as a single burst, but is speculated to be a series of reflections off the coastal shelf of Java. It may be possible to isolate a single reflection to see of it matches the crescendo signature. More importantly may be verification that reflections from a coastal shelf are a plausible interpretation of the Java Anomaly, versus a quake. The ARA San Juan implosion arrived as three separate peaks.

A broader time interval showing three signal arrivals from the implosion.

To measure the bearings accurately, the H10S hydrophone array was calibrated on the known bearing 218.0 direct path origin of the implosion. This allowed these H10S plots to be aligned with the proper delays between channels. The revised hydrophone locations are:

    % hydrophone locations from the Iris datafile mytrace header:
    % lat = [  -8.941177,  -8.959100,  -8.9561918 ];
    % lon = [ -14.648430, -14.645310, -14.6629000 ];

    % Please excuse the convergence precision being greater than the input accuracy:
      lat = [  -8.941213253,  -8.959031597,  -8.952851643 ]; % H10S calibrated on 4/5/2019
      lon = [ -14.648363660, -14.645313650, -14.662735730 ]; % aligned to ARA San Juan implosion bearing: 218.00331
      wavespeed = 1.48183; 

With the recalibrated hydrophone locations it is now possible to get an accurate relative bearing to the second and third implosion arrivals. Various algorithms can emphasize different aspects of the signals. Also, using a larger time window with oversampling gives better bearing resolution while reducing noise, versus better time resolution with a shorter window. Here is a sample of an eigenvector filtered plot for estimating the narrowed bearings.

Three H10S implosion arrivals with a measurable shift in back azimuth.

The second and third pulses appear to be shifted from 218.0 to 221.4 and 225.5 degrees. They are delayed by 86 and 146 seconds. This indicates that these are reflections from the Argentinian coastal shelf north of the implosion site. Projecting the bearings from the Ascension Island array back to the coast allows picking two specific reflection points based on the presumed depth of maximum SOFAR interaction. The accuracy of those points can be gauged by the expected propagation times over the longer paths. When the reflection points were first picked at the SOFAR depth of 900m, the propagation time errors were about 15 and 20 seconds too fast. Picking instead for a 200m T-wave reflection gave surprisingly good matches within a second of the profiled prediction.

Map of T-wave reflection paths from the ARA San Juan implosion to H10S.
(Click for .kmz file)

This seems to confirm that the hypothesis of an upslope/downslope T-wave reflection fits the observed implosion signals. It provides additional validation that the Java Anomaly arrivals at the Diego Garcia H08 array would follow the same physics and could well be from an implosion within the plane rather than a natural quake event.