A possible Flyover of Christmas Island by MH370

Christmas Island in the Indian Ocean has a runway that handles daily jet traffic. Its location is between the 5th and 6th ping arcs, making it a logical next destination after Cocos Island, which is between the 4th and 5th arcs. Near the expected travel time between the arcs, a lone seismic signal arrival after dawn may have been the plane on approach to the Christmas Island airport. Detection of jets by seismometers has been shown. A slow Doppler shifting seismic signal does appear earlier at Cocos at about the right time. Detecting the plane at Christmas Island would be further evidence that it was navigating by waypoints, and that a pilot could have sought out a landing site.

There is no infrasound array on Christmas Island, but two seismometers have data available from March, 2014. Station AU.XMIS is in a quiet area at the center of the island, and AU.XMI is located next to a large hangar at the airport, with regular disturbances by a compressor pump during airport hours. The expected flyby time can be found by interpolating time and distance on any straight line path between the 5th and 6th arcs that crosses the island. This would be around 23:58 UTC. Adding +7 hours gives 6:58 AM local time, when the airport was open.

Coincidentally, at 23:52:18 UTC there was a nearby M4.1 quake, strong enough to be cataloged. There is only about one cataloged quake per week in this region over the Java Trench, and it unfortunately arrived at XMIS right in our flyby time window. It dominates the signal scape on both seismometers from its P-wave arrival at 23:53:06 through to 23:57:00 UTC. For 15 minutes after the quake, all is quiet on XMIS with no sign of a flyby. There is only background noise for 30 minutes before the quake arrival, until there is a unique low frequency sound that fades in for about 45 seconds, then fades away for 15 seconds, just before the quake hits.

Spectrogram of quiet period before a possible XMIS flyby signal.

The signal peaks at 23:51:50 which may be its closest point to XMIS. There may be a match to a faint trend that has been dropping from 10Hz to 9Hz over the previous five minutes. Zooming in for best detail shows the prominent signal sweeping upward in frequency from about 14 Hz to 15.3 Hz. The frequency plot is a close match for A380 engine noise seen by the COCO seismometer. This is unlike a Doppler shift, which is always descending for a constant frequency source, as described with the COCO flyby.

A zoom into the possible flyby signal spectrogram for XMIS.

This upward sweep might be possible if the plane was slowly throttling up to hold altitude as it came out of a glide approaching the airport. It may have flown past the island, or been on approach to the YPXM airport for runway 36. The autopilot approach would have been from the west on heading 80, then a left turn to heading 355 to line up with the runway. This may have happened at a higher altitude where it went unnoticed. It is unlikely that the terminal operates radar or glide slope with such low traffic. Taking an audible minute to fly past is about right for a jet a low altitude. There are hints of lower frequency Doppler-like shifts that might be related to the more constant airspeed of the plane rather than engine speed.

If the plane was at a constant altitude and flying between waypoints, it would likely have been on heading 79.66 degrees from COCO, passing north of seismometer XMIS by about 2.5 km toward the YPXM VOR. If the plane was executing a turn to the next waypoint, that might also account for a slow throttle up.

About 30 seconds after passing XMIS, the plane would have been passing the airport seismometer XMI. Unfortunately, the compressor noise is occluding any signature.

Airport compressor noise pattern at XMI just before the quake.

To bring out weaker signal components between the louder pattern noise, it is possible to compress the signal amplitude to the point where it looks like an FM carrier wave, representing the phase components of the signal.

The same XMI timeframe with phase-only compression.

In the phase-only plot, the noise appears to be multiple overlapping frequency descents. The source is unknown, but might be related to mining or construction. At 23:51:00 and 23:52:00 there are faint cyan colored traces that have a slight upsweep around 16 Hz. At 23:52:15 there is a brief upswept 15 Hz signal that would match the longer 23:51:50 XMIS fly-by. The delay at the peaks is about 25 seconds from the XMIS arrivals. The slightly higher frequency near the same times could be due to Doppler shift. The compression/phase technique can be applied to also partially ignore the intensity of the quake, but little is gained.

The 25 second delay between a fly-by of XMIS and XMI can be used to estimate the speed of the plane. Assuming the plane was on a waypoint track from Cocos YPCC airport to Christmas YXPM airport with a possible turn, the distance between closest approaches to the two seismometers is about 4.5 kilometers, regardless of altitude. Covering that distance in 25 seconds would mean a ground speed of about 650 kph, or 350 knots. This is well above a final approach speed.

Validation of Seismometer Infrasound Detection

To verify that the XMIS seismometer can detect a plane in the air from several kilometers away, the same flight schedule used to check the Cocos Island infrasound includes the arrivals at Christmas Island CXT/YXPM airport. Flight VOZ1917 from Cocos was due to arrive CXT at 17:50 local time = 10:50 UTC. The XMI seismometer adjacent to the runway easily detected a touchdown and roll at 10:51:52 UTC. Looking at the XMIS seismometer around the same time does show the known flyby clearly.

Reference XMIS spectrogram of a flyby for an A380 jet on final approach to Christmas Island airport.

The peak of the flyby came almost three minutes before touchdown. This implies that the plane was on a longer northern approach to runway 18. A climate model for that time shows wind from 140 degrees at 7 knots. Despite the regular one-minute noise pattern (likely waves on the shore), the same slightly rising frequency shift can be seen, followed by a slight drop that could be Doppler. The prominent frequency around 10 Hz is lower than MH370 with more overtones, but we know that VOZ1917 was descending on final approach. This might indicate that the MH370 engines were at a higher throttle setting. The second spike after touchdown may be detection of the engines again as VOZ1917 turned around and taxied back up the runway to the terminal.

North or South of XMIS?

Colors on these plots have a significant meaning. Each RGB color channel is assigned to a seismometer axis. Red=E=east-west, Green=N=north-south, and Blue=Z=Vertical. Unlike pseudocolor palettes used only for contrast enhancement of single channel spectral plots, the major directional components of a signal are encoded in the color on these plots. The XMIS signal for MH370 at its strongest point goes through a quick series of color changes, faster than the movement of the plane would allow. The yellow-orange color would normally indicate a lower vertical component, but that’s unlikely for a plane at altitude. More likely is that the conduction of the infrasound into the seismometer foundation is being affected by the vault structure. A polarization analysis on the quake P-wave however is at 140 degrees, when it should be around 73. This is a sign of problems in the data channel order, but the channel tags are correct. A python DLOPy rotation check using 54 events reports a rotation error of only -4.2 degrees. Other seismometers are consistent with the data order loaded into MATLAB. Or, the polarization might simply be overwhelmed by environmental noise. This is still being investigated.

Running a seismic polarization analysis on XMIS for MH370 shows the ambient noise having a pattern with a lateral component at 140 degrees from north, and a dip of 45-50 degrees. The likely source is waves rolling along the beaches. At XMIS closest approach from 23:51:47 to 23:51:50, there is a brief flip of the inclination to 135 degrees, and the major axis ramps from 138 to 166 degrees. If that bearing is anywhere close, then it points to the plane passing north of the XMIS seismometer. To do that it would have been flying directly between waypoints, and not on a runway approach. (Until the polarization checks out, this result is not reliable.)

Self-Selecting the Signal Pattern of the Plane

Another method that may be applied is an autocorrelation map. In this case, it is used in combination with full signal compression, to become a phase-autocorrelation map. Both axes on the plot are the same time frame, and the zero-offset autocorrelation progresses diagonally from top-left to bottom-right. The time differences between autocorrelation peaks are on the opposite diagonal, and can be obtained by taking the difference between two axis coordinates.

Phase-Autocorrelation map for XMI around the proposed flyby time.

This autocorrelation plot for XMI covers ten minutes centered on the proposed flyby time. The brightest self-similar (perhaps resonant) signal starts in the center at 23:52:10 UTC. The stronger M4.1 quake P-wave does not arrive until over a minute later at 23:53:06 and its S-wave at 23:53:34 is mostly invisible. Notice that there is a faint X pattern to the overall image, with a scattering of autocorrelations along the other diagonal at delays that line up on the center. These longer autocorrelation matches might be consistent with being selective for the plane’s signal pattern being similar before and after the center flyby time. The same pattern does not show up on XMIS, possibly because the quake had a stronger resonance there, and autocorrelations on surf noise cause diagonal banding throughout the image.


While the evidence for a flyby is not as strong at Christmas Island, the timing is just about right for the placement of the island between the 5th and 6th arcs. The time is earlier than expected by a few minutes, but the expected time was based on the plane flying the same heading before and after the island. There is no waypoint on that heading towards the 6th (or 7th) arc. Assuming the plane made another turn at the YPXM airport waypoint, it would be taking a less direct path to the 6th arc, which moves the expected flyby to an earlier time. The next waypoint is not yet known, but the nearest airport on the Java coast would be Wi1A CJN Cijulang Nusawiru Airport with a 4,700 ft runway. This would fit with a 7th Arc endpoint at the previously reported Java Anomaly site.

The Christmas Island runway would make a desirable destination, especially if the pilot had any hope of landing the plane. We do not even know whether it had the capability to land on a runway. This might be consistent with the scenario of an initial cockpit fire. The fuel load was now reduced without the ability to dump the tanks, but flaps or landing gear may have been out of commission. Hypoxia from a slow depressurization may have been handled by flying at a low enough altitude. There could have been an effort underway to diagnose the problems and make repairs.

The trove of seismic data has not been fully explored. Refining the visualization algorithms to improve discrimination will continue. Putting together a waypoint path and checking it against BTO rings, speed, fuel, and BFO will be important to presenting a credible candidate location.

The families of those on MH370 are always in our thoughts. Apologies to them, for any insensitive references to the way the plane was lost, and for the years it has taken to gather this evidence before taking it public. The truth will come out, and hopefully this effort will help us all get there.