MH370 – Locating the Curtin Event Origin with Seismometers

The first acoustic candidate reported by Curtin University in 2014 still stands out as an anomaly. Its location is not precisely known for lack of good matches on other hydrophones for triangulation. This analysis provides an improved estimate of the Curtin event origin by detecting the conversion of hydroacoustic sound waves as seismic waves received by seismometers on islands around the Maldives.  

Previous Estimates of the Event Origin

The first Curtin report analyzed data from the CTBTO Cape Leeuwin H01 triad array which can reveal the direction of origin, finding only one event arriving from the N-W sectors on bearing 301.6 that fit with any possible impact timing for an unknown range. Additional Australian IMOS hydrophones were retrieved and analyzed from Perth Canyon and later Scott Reef locations. Two of the intermittent recordings provided possible matching signals that gave approximate locations first by Carlsberg Ridge south of the Arabian Sea, then SE of  the Maldives in June 2015.

Work here in 2015 appeared to point to a late implosion near Batavia sea mount where the 301.6 bearing crosses the 7th Arc. Corrected long baseline reanalysis using revised Scott Reef hydrophone timing then pointed to an origin near 4.65N 66.62E, again west of the Maldives. This triangulated location also implied an origin time about one minute after the last ping handshake with the plane. The location closely matched 7th arc timing but was too far from the consensus search area for serious consideration. It was noticed in 2016 that the direct path from this location to Scott Reef was blocked by islands in the Maldives.

[2018 Update:  On October 3, 2017,  the ATSB released a 430 page summary titled, The Operational Search for MH370. Appendix H and I (p.369-401 and p.402-413) of that document are the first public access to previously unseen acoustic analysis reports submitted by Curtin University dated June 23 and Sept 4, 2014. The later report includes their refined location of 2.11N 69.31S west of the Maldives. Note that the direct path from that location to Scott Reef hydrophone (15.483S 121.251E) is also completely blocked by Kandudu atoll. This significantly disrupts the premise of the estimate. ]

The Curtin signal was not detected at CTBTO hydrophone array H08 offshore SE of Diego Garcia, which is blocked from signals to the NW by the topography.  Unpublished research from the University of Washington [also referenced in the Oct 3, 2017 report appendix I ] attempted to use H08 low frequencies and possibly reflected signals to estimate a NW origin, but a verification attempt here using a Bearing vs Time visualization showed only a general origin through a gap at bearing 306 from H08.

The IMOS hydrophone locations were corrected here after receiving the actual recording data. It was later realized that the direct signal path from the triangulated location to Scott Reef was blocked by an atoll. The path from bearing 301.6 to H01 also passes just south of the Maldives through a gap in the island chain north of Diego Garcia. Since the signal was not strong, there was a possibility that the H01 bearing was slightly off and the  direct signal path was blocked, taking a diffracted path around the southernmost Maldives atoll.

H01 signal intensity by Bearing vs Time for the Curtin Event
Spectral plot for the Curtin event arrival at H01

The phase spectrogram of the signal arrival appears to show a linear dispersion in frequency over time. This is unlike other common signals from ice events and quakes. If reference surface events were to have the the same linear signature, it might help distinguish this event from a quake.

Experimental autocorrelation methods appeared to reveal weaker variants of the Curtin signal arriving through other gaps between islands at both H01 and Scott Reef. Months of effort were made in 2016 and early 2017 analyzing these faint multipath arrivals. This offered the potential of triangulations using virtual hydrophone locations at the diffraction points. Unfortunately, different methods gave various estimates for the origin, so none were submitted as candidates.

Examining satellite imagery near those non-candidate locations did reveal what might have been smoke plumes at different wavelengths, but surrounded by clouds that had similar shapes to the “plumes”. The intriguing NASA EarthData AQUA/Modis satellite view at 06:13 UTC can be seen here:

Five other hydrophones operated by the French OHA-SIS-BIO survey in the Southern Indian Ocean on that day were retrieved in 2015 and analyzed in their unpublished report. These detections were combined with the CTBTO array arrivals and the report assigned two different possible origins for the Curtin signal, both east of the Maldives. There is still hope that the French recordings (and report) will someday be released for independent study.

Background on Seismometer T-Waves

In discussions of whether a surface impact over deep ocean might be detectable at all, it was suggested that hydrophone recordings from the 2009 crash of Air France flight 447 in the Atlantic Ocean could be useful. The CTBTO hydrophone data from the Ascension Island array had been examined during the 2009 search with no signals detected, but now knowing the exact location and time might reveal some weak trace.

A search for other hydrophones with a direct path to the AF447 site brought up a short list, including locations in the Caribbean. It was pointed out by Alec Duncan at Curtin that these were not actually hydrophones but seismic T-wave stations. These are seismometers on small islands that can detect the conversion of SOFAR channel sound to seismic signals. Investigating the T-Wave method inspired the possibility of utilizing the rich public network of active seismometers around the Indian Ocean to help locate MH370.

In truth, an expert LANL team had already examined the closest seismometer to the 7th Arc, station G.AIS on Amsterdam Island and others, but dismissed them along with H08 as too cluttered by various noise sources to be useful.

T-waves are a rarely used seismic classification, where standard earthquake origin analysis involves the timing characteristics of the fastest P (Primary/pressure) phase waves and slower S (Secondary/shear/surface) phase waves. There are other slower surface waves through the Earth, but the T or Tertiary phase waves travel at the much slower speed of sound in water (typically in the SOFAR channel).

The timing of T-wave arrival at a coastal interface is much more dependent on local geography than P or S waves. T-Waves are a form of pressure wave, since a liquid medium does not support shear waves. The arriving T waves first become S waves as they interface with the sloping seafloor (and faster crust) at SOFAR depth. This minimum sound speed depth ranges from near the sea surface in the Antarctic to about 1600m near the equator, but can be predicted from temperature and salinity.

Across physics, when a wave reaches an interface between two media of different travel speeds, it can be both refracted and reflected.  As the T-waves arrive upslope toward the coast, they continue to generate S waves by refraction but the pressure waves are also reflected and focused between seafloor and surface until they crescendo at a critical depth of about 200m. At that conversion point the wave energy is conducted as an impulsive P wave into the crust. Because the focusing is caused by a halving of the slope angle on each bottom bounce, steep slopes generate sharper P waves.

The timing at the seismometer is now dependent on the locations of the S and P wave conversion points, with S waves atypically arriving before P waves at the shore. With a station farther from the shore, the P waves will outrace the S waves. At a certain distance, the P waves will arrive in the middle of the slower S wave arrivals. More inland stations would detect the P wave first, but care must be taken to compute the combined propagation times if it is realized that the signal is to be interpreted as a T-wave. Also consider that there will be stronger and weaker conversion points along a coastline giving a mix of arrivals for any particular event.

Another aspect of using geophones is that seismologists typically analyze only very low frequencies, filtering out higher frequency events as anthropogenic noise, like explosions. Many broadband seismometers are sampled at only 20Hz, giving a useful range only below the Nyquist cutoff at half that frequency. The fastest seismic sampling rates available are about 100Hz, as compared to the CTBTO hydrophone sample rate of 250 Hz and IMOS marine observatory hydrophones sampled at 6,000 Hz. Algorithms for teleseismic quake location generally filter below 1 Hz. The frequency range for SOFAR event detection is  above 3-4 Hz.

Tools for visualizing the higher frequency seismometer data are still under development here. They currently show correlation between the three axes, spectral details, and the results of polarization analysis. That method attempts to extract azimuth and inclination of the arriving signal by finding eigenvectors, similar to the PCA (Principal Component Analysis) approach.

While the CTBTO hydrophone data for the 2009 AF447 event remains restricted to qualified researchers, recordings from multiple Atlantic seismometers were downloaded via the IRIS, FSDN, and GFZ public seismic data networks. No obvious trace of direct T, S, or P-wave seismic arrivals match the 2009 event. AF447 crashed due to a continued low speed stall which may not have generated a large acoustic wave, but the lack of a detection makes it less likely that an MH370 impact over deep water would generate a detectable acoustic wave in the SOFAR channel.

Analyzing Seismometers near the Maldives

There are limited seismometers in the Maldives region that have an unimpeded acoustic path to the west. The clearest event arrival was at station IN.MNC on the small Indian island of Minicoy, just north of the Maldives. It has a clear path to the west and recorded a strong event at 00:28:40 UTC. The signal has high frequency components that increase up to the 45 Hz limit of the 50Hz Nyquist cutoff filter. This is characteristic of a sharp impulse source. Continuous 14 Hz resonant noise (local pumps?) plus various narrow harmonics are masking weaker seismic signals and disrupting the polarization results.

Curtin event T-wave 00:28:45 arrival at Minicoy Island

Maldives station GE.KAAM at Kandudu airport is on the western limb of the atoll. It has a moderate background noise level but detected an event at 00:30:00 UTC. 

Maldives seismometer GE.KAAM arrivals from 00:19:37 Carlsberg Ridge candidate

Using distances derived from the T-wave arrivals gives a location that varies somewhat with time, but errors are minimized near the consensus (satellite data) impact time. Using 7th Arc timing of 00:19:37 gives a Carlsberg Ridge origin at 5.431N 66.158E west of the Maldives. The geodesic bearing from the H01 array to here is 301.64 which matches the original 301.6 +/- 0.5 Curtin bearing.

The largest unknowns are along an east-west search path. The candidate origin moves eastward at 1.5 km per second of event time after 00:19:37 UTC. A shift of the origin to the west is also likely because the calculations were made for predicted deep SOFAR channel sound speeds, while waves from a surface impact would travel sightly faster outside the channel. A mismatch of about 5 km between distant H01 and the nearer seismic stations is already incorporated into the origin estimate.

Having another seismometer for triangulation verification could pin the origin time and narrow the search area. Unfortunately, the GE.HMDM station was offline at the time. The II.DGAR station at Diego Garcia is blocked by the large Chagos Bank, like the H08 hydrophones. The II.MSEY station in the Seychelles is in the center of a populated island with 100km of shallows between the island and the SOFAR channel path. GE.SOCY on Yemen’s Socotra Island did not have any arrivals matching the Carlsberg Ridge location. Its only match for the H01 Curtin event would have put it perpendicular to the 301.6 bearing path, at a much earlier time off the coast of Oman.

However, standard seismology could confirm the origin timing. Liquids conduct only pressure waves. A surface impact would create a strong pressure wave toward the seabed. From the seabed conversion point, seismic waves, including shear waves, would propagate normally. A valuable tool for computing seismic travel times is the TauP toolkit. It estimates wave arrivals using average values for density and discontinuities at different depths in the earth. This does not take into account local variations, but this deep ocean location is similar to the global average.

Expected TauP wave arrivals at KAAM from the Carlsberg candidate location at 00:19:37 (accounting for the 1.3 second surface to seafloor delay) would be 00:21:42 for the the first P-wave and 00:23:16 for a cluster of S-waves. A close examination of the KAAM spectral plot does show weak arrivals at those times. A stronger arrival from 00:31:30 to 00:32:30 may correspond to a PcS wave reflection off the outer core, which could be indicative of the stronger downward component of the pressure wave. Another plot (not shown) displaying signal phase only (independent of amplitude) shows the best signal coherence at the S-wave and T-wave arrival times.

The T-wave conversion points for KAAM and MNCY were chosen based on the expected arrival direction and conversion depths. Details and calculation notes are contained in the Google Earth .KMZ file.

370location Maldives triangulation area snapshot.


This is not a prime candidate location, but it does match up wit the H01 hydrophone detection plus T-waves on two seismometers.

An event originating at the same time as the truncated last ping from the plane seems beyond coincidence, even if it does not correspond to the satellite BTO/BFO timing placement on the 7th Arc. Although far from the active search area, early drift analysis showed the area as more likely to be an origin for the recovered debris.

Putting out the revised location and its new basis may help with future searches or aid in other analysis, even if it does not meet the current threshold for inclusion in the MH370 search.

The Plume1 site noted above happens to be on the intersection of the seismometer T-wave arrivals at around 00:21:32 = 1:55 after 7th arc final ping timing.  However, the H01 arrival at 01:34:50 would need to have originated from Plume1 before 00:20:56 to arrive by even the most direct path. The triangulated location for a 00:19:37 origin is the best match with the signals.

This location does not coincide with historic or recent seismic activity along fault zones. Two closely spaced events were weakly detected at H01 several hours earlier, but from around bearing 299 that could match with a cluster of seismic activity near 4.1N 62.6E or an active quake area on Mar 7 near 2.35N 66.84E. They could also hint at distant origins off Yemen or Oman.

[ Feb 2018 note: This post in draft form was started in August 2017 but work on new seismic findings took precedence. The plots were rerun with corrected polarization code, and the map data has been updated for a release in chronological order of development. ]

Mar 2019: Improved Bearing plot using SVD eigenvalue filtering, 8x oversampling, and H01 hydrophone recalibration. The averaged bearing from H01 is 300.95 degrees. That precision does not support the accuracy, so 301.0 is a better estimate.

Mar 2019 improved Curtin event plot, average bearing 300.93 degrees.