Lightning strikes are among the loudest natural sounds near the surface of the ocean. With multiple acoustic candidates for an MH370 event, it was considered that some might be eliminated as lightning. A 48 hour database of over 3 million accurately located lightning events around the Indian Ocean was provided for this purpose. The precision of the data also opened the possibility of calibrating the hydrophone array bearings against known sources. Ultimately, even the strongest ocean strikes could not be discriminated from other noise at the hydrophones. This negative result is an indication that a surface impact of MH370 over deep water may propagate poorly over long distances as sound in the deeper SOFAR channel.
Approach
Lightning strikes on the ocean can reach sound intensities of 270 db, where a typical ambient ocean noise level is around 100 db. For comparison, a large ship might generate 190 db, while the loudest mammal sound is a sperm whale click at 230 db, similar to a large seismic sounding airgun at 230 db. Repetitive airgun blasts from multiple ships in shallow water are dominating the Diego Garcia H08 hydrophone sound spectrum thousands of km away, so it might be assumed that lightning strikes would be easily detectable.
Researching the lighting strikes would not have been possible without generous assistance from Vaisala Corporation. Vaisala operates a network of radio receivers around the world that detect the static discharge of any strike within range, triangulating and gauging the intensity and type of strike. The Vaisala Global Lightning Dataset GLD360 network has a real time feed, and also offers historical data. Vaisala Tucson meteorologist Ron Holle was an advocate, and expedited the project from inception to followups. Tucson Data analyst William Brooks quickly delivered a database containing 3.1 million strikes in a large rectangular selection covering the Indian Ocean and surrounding coastlines for the 48 hours around Mar 8, 2014.
The typical intensity of a negative Cloud-Ground strike is around 30 kiloamperes (KA), dissipating around 500 megajoules of energy. The strongest strike in the database is an impressive -973.6 KA, with dozens of strikes in the range of 300-600 KA.
To visualize the extent of the dataset, a script was written in python to organize the strikes into storm cells and generate a map. (Click for a larger 4k version.)
A dozen or more large strikes selected for analysis are starred on the map. These are downloadable as a Google Earth .kmz file with details in the notes.
Results
There were no lightning storms that would interfere with the plane. None of the search areas or candidate acoustic sites had nearby lightning strikes. There were two large strikes near the expected impact time of 00:19:37. One was a -528 KA cloud-ground strike at 00:18:56 near Cunningham Island off the coast of Pilbara, Australia. It was an ocean strike with a water depth of 1100m. That area is not in view of the Cape Leeuwin H01 hydrophone. There is a direct path to the Diego Garcia H08 hydrophone array but it was not detected, despite a strong series of signals arriving across the ocean from a seismic survey ship only 575 km along the same coast in 400m water. The other strike near 7th Arc timing was -179 KA at 00:11:38 off the tip of Kulon peninsula near Jakarta, over coastal water of 750m. It was not detectable by any hydrophones or seismometers.
For reference, two large strikes about a minute apart were within 150 km of Diego Garcia, on known bearings 114.75 and 116.14 degrees. These two strikes were barely detectable in the noise, but not with enough strength to be used for calibration.
A wave traveling outside the SOFAR channel would travel through the water at higher speed, but also over a longer path that may include surface reflections. The first strike arrived about 9 seconds ahead of the expected time, and second was 14 seconds late. The bearings were also off in opposite directions, with the first at 114.1 and the second at around 116.75 degrees. This could be due to accuracy limits in the lightning data (since the strikes at sea are far from the radio detectors), or algorithmic errors due to the methods of enhancing weak signals (using the product of covariance and eigenvalues for the triad). Both arrivals seem to coincide with a broad arrival from around bearing 29 degrees. This may be faster multi-path waves arriving sooner, and reflectively focused by local undersea terrain on that bearing. It’s also possible that the weak selected signals are simply noise artifacts of some other event.
To find if any seismometers could detect lightning strikes, a python script matched strong lightning strikes around the Indian Ocean with a collection of 5235 global seismometer recordings over the same two day time period. The closest strong strike was 69 km north of the GE.SMRI Semarang station, over shallow 100m coastal water with -335 KA at 7 Mar 22:34:02 UTC. Neither this strike, or a nearby strike (103 km N of SMRI, -347 KA, @ 7 Mar 22:59:02) were convincingly detectable at SMRI, or by stacking recordings from the 50 station YR.ME RESIF-SISMOB DOMERAPI temporary array covering the width of Java just south.
The closest strikes were from a squall that passed directly over seismic station XK W11KP at Kapiri Hospital, Malawi, part of a larger African array. Though the strikes were much weaker, the signals are stronger due to proximity. At a range of about 10 km, strikes over 200m lake Malawi appear about as strong as strikes on land.
It was considered that the Curtin event located west of the Maldives might be lightning related. A storm cell was active in the region, but 325-375 km NE of the analyzed origin. None of the seismometers were able to detect known lightning strikes to -447 KA in that storm cell near their expected P-wave, S-wave, or T-wave arrival times.
Conclusions
Although lightning strikes have a high peak amplitude, they are extremely brief compared to most ocean sounds. Though local vaporization and cavitation effects on water could extend the duration, the impulse would still have a wide spectrum spread that might not correlate well for detection. Algorithms using mutual entropy or higher order statistics might get better results than covariance methods. If so, the same techniques might be more appropriate for detecting an MH370 surface impact.