Content uploaded by Unto Kalervo Laine
Author content
All content in this area was uploaded by Unto Kalervo Laine on Apr 01, 2016
Content may be subject to copyright.
ICSV19, Vilnius, Lithuania, July 8-12, 2012 1
ANALYSIS OF CLAP SOUNDS RECORDED DURING
THE SEPTEMBER 9-10 2011 GEOMAGNETIC STORM
Unto K. Laine
Department of Signal Processing and Acoustics, School of Electrical Engineering, Aalto Uni-
versity, P.O. Box 13000, FI-00076 AALTO, Finland
e-mail: unto.laine@aalto.fi
Strong geomagnetic storms accompanied with bright and lively aurora may produce audible
sounds perceivable at ground level. However, the physical origin of these emissions is not yet
known. In order to understand the phenomenon, sound recordings with simultaneous electric or
magnetic field measurements have been performed at different locations in Finland during geomag-
netic storms since 2000. One such measurement was performed on September 9, 2011 near the vil-
lage of Fiskars in southern Finland. The main goal was to test out new instruments and to demon-
strate the measurement procedure to a colleague from the University of Graz, Austria. Signals were
stored in WAV format using a Zoom H4n four-channel audio recorder. One stereo signal was re-
corded from the internal microphones of the Zoom unit while the second stereo signal consisted of
an audio signal from a separate B&K low-noise measuring microphone mounted at the focal point
of a parabolic reflector. The latter signal was recorded on one channel while the other channel con-
sisted of a VLF antenna signal. The total duration of the recording was 130.6 minutes and covered a
time interval approximately from 19 to 21 UTC. Twenty-one short sound events, clearly discernible
from the background noise, were detected afterwards in the recording. The best ten were selected
for closer analysis. These were the loudest events (claps) and exhibited relatively high mutual cross-
correlations and a low background noise. Three of these events were detected with all three micro-
phones. This is the first time during our studies when similar sound events were recorded simulta-
neously with multiple microphones during a geomagnetic storm. Knowing the separation distance
of the microphones (B&K vs. Zoom) it was possible to estimate the direction of the sound source.
The synchronously recorded VLF signal provided data to estimate the distances of the sound sour-
ces assuming that a certain magnetic field event is associated with a certain sound event. Three in-
dependent methods to estimate distance to the sound sources yielded an average value of 70 meters.
This outcome, together with the direction estimate, indicates that the average altitude of the sound
sources measured from ground was only 60-70 meters.
1. Introduction
A question that has been discussed since the early phases of modern geophysical science is
whether the strange sounds sometimes heard during bright aurora are just an illusion or could they
be created by some unknown physical processes in the lower atmosphere (or even at ground level)
by the geomagnetic storm. Observations of these sounds have been reported for hundreds of years
before science had even started to discuss the topic. A project to study aurora related sounds was
19
th
International Congress on Sound and Vibration, Vilnius, Lithuania, July 8-12, 2012
2
initiated during the spring of 2000 at the Helsinki University of Technology (TKK, which today
forms a part of Aalto University). Additionally, some researchers from the Sodankylä Geophysical
Observatory (SGO) have actively participated in the project especially during its initial stages.
In April 2000 an unannounced strong geomagnetic storm surprised many specialists dealing
with space physics. We also were quite unprepared to measure the storm and had to assemble an ad
hoc recording setup quickly. Careful analysis of the measured audio signals revealed an increase in
the noise level during the most intense phase of the storm. The analysis confirmed the perceptual
observations made during the storm as well as a week later at the place of audio recordings
1
.
After the April 2000 event the methodology was improved and a more systematic collection
of data started in the autumn of 2000. One year later the geomagnetic storm of April 11-12 2001
was a turning point for the project. Rich and exceptional sound material with synchronous VLF
signals was collected at Koli, which is situated in eastern Finland. The first comprehensive analysis
of the data was published in a 2005 master’s thesis
2
. We found statistically significant correlations
between fluctuations in geomagnetic activity and in sound power at different delays. However, it
was very difficult to associate any individual sound event with the VLF signal or with variations in
the geomagnetic field. The correlations were detected only between data sets when using large time
windows.
This paper presents an analysis of a new event that is interesting for the following reasons:
1 The sound events occurred during testing of a new four-channel recording system.
2 Multiple similar sound events were detected, some of them with all three microphones.
3 The collected data allows for the estimation of the locations of the sound sources.
The new methodology predicts with a high probability that the sound sources were in the air
almost directly above the place of measurements, and furthermore, it provides three independent
estimates for the height of the sound sources. All three methods are in agreement that the average
height of the sound sources was below 70 meters.
2. The September 9-10 2011 event
2.1 Geomagnetic storm
Based on satellite measurements (NOAA Space Environment Center) the solar wind was at its
highest value of 600 km/s around 18 UTC on Sept. 9 and slowed down to 500 km/s after two hours
and stayed around this value throughout the entire night. The solar wind density (particles/cm
3
) var-
ied between 1-10 during these hours. The interplanetary magnetic field was turned to the south
allowing for a strong interaction between the solar wind and the magnetic field of the earth (see,
e.g., ACE RTSW Estimated MAG & SWEPAM Begin: 2011-09-09 08:00 UTC). The proton flux
associated with the actual CME was decaying. The estimated Kp-index reached a value of seven
between 15-18 UTC just before the start of the measurements, see figure 1. During the
measurements the storm intensity decayed to a Kp-index level of 5. In short, this was not a storm of
the strongest category, however, around local midnight bright green aurora were visible in the
northern sky for a few minutes. The full moon made aurora observations difficult especially close to
the zenith.
2.2 Place of observation and arrangement of instruments
The village of Fiskars is located in the SW part of Finland. The measurements took place
about one kilometre north of the village centre. The instruments were positioned on both sides of a
small country road running through open fields. The measurement location was void of large trees
up to 150 meters in every direction. The distance to the closest power line (380 V) as well as a s-
mall road (with few passing cars during the entire night) was about 150 meters. The arrangement of
the different measurement instruments and the position of the car, where the event and its recording
was monitored, are shown in figure 2. The car’s windshield provided an unhibited northward view.
19
th
International Congress on Sound and Vibration, Vilnius, Lithuania, July 8-12, 2012
3
Figure 1. NOAA Satellite Environment Data Sept. 8-10, 2011. The period of measurement (19-21
UTC) is indicated with a blue box. The blue line indicates local midnight (DST).
The recording session had two main goals: to test a new Zoom H4n recorder
3
in this kind of
an outdoor measurements and to demonstrate the used methodology to professor Gernot Kubin
from the University of Graz, who was visiting Finland. The Zoom H4n was elevated 0.8 meters
above the ground with a tripod that was located 1.5 meters to the left of the driver’s seat of the car
(black dot in figure 2). The output of the Zoom unit was monitored with Sennheiser HD 580 head-
phones inside the car.
There are two reasons why the Zoom recorder was mounted in the proximity of the car. First,
it facilitated visual monitoring of the unit as well as the use of headphones without excessively long
connecting cables. Secondly, it also allowed for the monitoring of possible sounds coming from the
car (due to temperature changes or movements of the observers in the car). Also, this helped to iso-
late sounds arriving from sources at ground level from those arriving from the sky (especially when
the source is on the car side of the B&K microphone). The Zoom unit works like a near-field moni-
toring device while the B&K microphone system captures quiet sounds further away. The Zoom’s
microphones were pointing outwards away from the car.
Figure 2. Recording arrangement at a country way.
19
th
International Congress on Sound and Vibration, Vilnius, Lithuania, July 8-12, 2012
4
The Zoom H4n recorder has two condenser microphones at one end and two symmetric XLR
connectors at the other end of its case. The microphones were fitted with a windshield and finally
the whole device was shielded against humidity by folding a cloth around its body. The Zoom H4n
recorder is able to generate four-channel digital recordings at 44.1 kHz/16 bit quality lasting over
five hours long using two AA batteries. One of the XLR connectors was used (in non-symmetric
mode) to record the signal from the B&K low noise microphone-preamplifier system (4179/2660)
mounted at the focal point of a parabolic reflector and also equipped with a windshield. This con-
struction is described in more detail in the master’s thesis of Janne Hautsalo
2
. The distance between
the microphones was measured with an acoustic impulse that was given through the car’s front-left
window. The window was exactly in line with the microphones. The estimated distance of the
microphones using the speed of sound was 13.4 meters.
The other XLR input was used for a symmetrical VLF antenna coil that was designed by the
author. It consists of a wooden frame holding two square shaped coils (edge length 67 cm) that were
coiled in a bifilar manner. Both coils are made of 100 m lengths of insulated copper wire. The coils
are connected to the XLR input of the Zoom H4n recorder as if it were a microphone with a sym-
metrical output. This inexpensive “VLF receiver” (antenna coil and Zoom recorder) seems to work
better than the vertical VLF antenna used in earlier phases of the study
2
. The antenna coil was
mounted with its axes in the east-west direction in order to monitor the B
y
component of the mag-
netic field. This component is sensitive to vertical currents in the atmosphere and less sensitive to
noise caused by distant thunderstorms that are prevalent at that time of the year
4
.
2.3 Weather conditions
The sky was mainly clear with only a few thin fog clouds. Unfortunately, the full moon was
shining but the air was still. The temperature gradually dropped from +10 to +8 °C during the
measurement period. The humidity was close to 90%. These parameters indicate that conditions
were good for sensitive acoustical measurements of atmospheric phenomena.
3. Analysis of the clap sounds
3.1 Selecting the ten best candidates
The total duration of the four-channel recording is 130.6 minutes. The signal from the B&K
system was studied manually with care in 10 seconds windows by looking at its spectrogram and by
listening, often several times at the same selection. PRAAT
5
software was used for this analysis.
The twenty-one most interesting short sound events were identified and copied to separate stereo
WAV files. In this manner the Zoom’s own stereo microphone signals were copied with the
B&K/VLF stereo samples, all synchronously.
The time locations of these 21 events are depicted in figure 3. The selected ten events are em-
phasized with larger red points. The time instance of the event is given in seconds from the begin-
ning of the recording.
Figure 3. Time instances (y-axis, in seconds) of the twenty-one sound events (x-axis). The larger red
points show the selected ten cases.
19
th
International Congress on Sound and Vibration, Vilnius, Lithuania, July 8-12, 2012
5
The detected sound events occurred more often at the beginning of the recording. Their aver-
age temporal period increases towards the end of the session. They also appear in clusters separated
by pauses. The general picture fits well with the prevailing geomagnetic conditions as the recording
was performed under decaying activity. Based on the Zoom H4n’s signal, event 15 was detected to
be a noise from the car. It also deviated spectrally from the others.
The first clap was one of the strongest and also the most clearly audible within the Zoom sig-
nals. In fact, this event triggered the entire analysis study. This sound was found in the recording
only 22.7 seconds from the beginning. The observers were still outside of the car and in a discus-
sion. Luckily the event occurred during a pause in the conversation, however, the sound level was
so low that neither of us could hear it. At that moment we didn’t even concentrate on listening.
The cross-correlation matrix in the right panel of figure 4 shows how the correlation system-
atically decreases between the first event and the next five events. The same is true for event num-
ber two (#2). This might be connected to the gradually and monotonically decaying geomagnetic
storm. The first six claps form a highly correlated set. Also, a relatively high mutual correlation was
found among claps #7, #9, and #10. Clap #8 differs the most from the other claps. When #8 is left
out of consideration the minimum cross-correlation between the other ones is 0.58. The mutual
cross-correlations of the first six claps were so high that the author was almost sure that they must
be a kind of artefact, possibly created by the instruments, e.g., by the Zoom recorder. It is still diffi-
cult to believe that some physical mechanism in the atmosphere could create so many similar
sounds in a relatively short time window - one after the other.
Figure 4. Waveforms of the best ten selected sound events (left) and their cross-correlations (right). Clap #8
deviates the most from the other sounds. The six first sounds are very similar with cross-correlations varying
from 0.74 to 0.91. Clap #7 is partially corrupted by the noise of a passing car.
The left panel of figure 5 shows the average RMS values of the ten clap sounds (arbitrary
scale) with their standard deviation added and subtracted. All RMS values rise very fast similarly to
an explosive sound. The decay occurs also quickly so that the pulses are only about 20 ms in dur-
ation. The right panel shows the spectrogram of clap #1. The signal has a spectral peak at 2.6 kHz.
A secondary peak occurs around 11-13 kHz. The higher frequencies are attenuated more rapidly
than the lower ones; this is probably due to the sound attenuation of the atmosphere. Supposing that
the sound is produced by an impulsive source that has a flat spectrum, and by knowing the outdoor
temperature and humidity conditions, it is possible to estimate the distance of the sound source
based on this attenuation. A rough estimate is that the distance cannot be less than 50 meters and
not larger than 100 meters. Thus 75 meters is the first estimate.
19
th
International Congress on Sound and Vibration, Vilnius, Lithuania, July 8-12, 2012
6
Figure 5. Left panel: mean RMS values of the ten selected claps (blue), standard deviation added (red) and
subtracted (yellow) from the mean, window length is 5 ms and step-size 1 ms. Right panel: spectrogram of
clap #1. Note the resonances close to 8, 13 and 17 kHz that are approximately 3, 5 and 7 times the resonant
frequency at 2.6 kHz.
4. Estimation of sound source direction
Clap #1 produced a stronger signal to the Zoom H4n unit than any of the other events. This
case was used to solve for the trajectory of the sound source. The clap signal arrived at the B&K
system 9.1 ms earlier than it did for the Zoom. Thus the source was 3.06 meters closer to the B&K
microphone. Figure 6 shows the solution for the sound source trajectory. It starts from the ground
level from the road between the instruments and follows a square root law upwards. The solved
trajectory may still rotate around the axes connecting the two microphones. The sound source could
still be located at the ground on the left or right side from the axes. Additional information is re-
quired from the stereo microphone channels of the Zoom to solve for the missing angle of the tra-
jectory. The Zoom recorder can provide this information but only from the level difference of its
left and right microphone channels. In this case the left channel had a 2 dB stronger signal.
With a high probability we can conclude that the sound source was located in the sky a little
left from the Zoom and closer to the B&K. At the moment we do not know the exact directional
pattern of the Zoom and this would be needed for a more accurate estimation of the source location.
Figure 6. Solving the trajectory of the acoustic source location.
19
th
International Congress on Sound and Vibration, Vilnius, Lithuania, July 8-12, 2012
7
5. Estimation of the height of the sound source from ground
In section 3.1 a preliminary and rough estimate for the distance (or maximum height) of the
sound source was given based on the flat spectrum hypothesis of the source spectrum and spectral
attenuation characteristics of the atmosphere. In this section the estimation problem is solved with
two other methods both of which are based on a detailed VLF signal analysis. These methods were
evaluated to be the best candidates for this task among many different approaches and proposals.
The VLF based source height estimation task is probably the most difficult part of the entire data
analysis. The VLF signal is very complicated and rich in electromagnetic (EM) events from differ-
ent sources around the globe. It has been estimated that the signal consists of about 50 EM pulses
per second (so called spherics) generated by thunderstorms at different locations on the earth.
Our task is to find VLF events that occur before the audio events and are local, i.e., their
fingerprint in the VLF signal should exhibit near field characteristics. Ideally, a vertical current im-
pulse in the near atmosphere should produce an exponentially decaying impulse in the VLF signal
that includes a DC component. The first method is based on this low frequency feature of the near
field VLF pulses. The reflections between the ionosphere and ground makes the picture even more
complicated. However, these reflections may also help with source localization
6
. The second
method is based on an analysis of the ionosphere-ground reflections.
The first method was realized by computing spectrograms of the VLF signals starting from a
few seconds before the corresponding clap sound and ending slightly after it had died out. The
mean spectrum was subtracted from each spectrogram in order to cancel the background noise, e.g.,
signals from VLF stations and electric power lines. Then the spectrograms were averaged in order
to see common structural areas. The upper panel of figure 7 depicts a section of this averaged spec-
trogram at about 220 ms before the clap sounds. The red arrow shows the most intensive low fre-
quency part of the data. It occurs about 200 ms before the clap events.
Figure 7. Searching for VLF event that can be associated with the corresponding clap sound. Upper
panel: averaged synchronized VLF spectrograms 220 ms before the clap sound, frequency scale from DC to
16 kHz (32-point analysis, Hamming window, hop-size: 8 samples). Lower panel: histogram of the most
energetic, systematic ionospheric reflections.
19
th
International Congress on Sound and Vibration, Vilnius, Lithuania, July 8-12, 2012
8
The second method applies FIR filters with a tap coefficient of 1.0 at distances corresponding
to the EM pulse travelling from the ground to the ionosphere and back. The other coefficients are
zero except for those of the neighbouring taps of ones. Since the height of the ionosphere is not
known we have to test many hypotheses, each having its own unique FIR-filter realization. Differ-
ent ionosphere height hypotheses from 85.03 to 98.64 km were tested. The 91.8 km hypothesis pro-
duced the strongest indications of systematic reflections. The sporadic-E layer may be the cause of
these reflections.
The outcome of method two is depicted as a histogram in the lower panel of figure 7. The
largest number of strong reflections occurs slightly after the strong low frequency peak in the spec-
trogram (upper panel: method one). The result is preliminary and still needs more work. However,
the largest peak in the histogram confirms the result based on the near field hypothesis. To summa-
rize, two independent VLF analyses confirm the preliminary spectral domain hypothesis related to
the distance of the EM-acoustic source from the place of measurement. The 200 ms delay between
the EM pulse and audio suggests a distance of about 66 meters. Thus we can conclude that the
height or altitude of the source must be below 70 meters from ground level.
6. Summary and Discussion
We performed a test and demonstration recording during the September 9-10 2011 geomag-
netic storm. Closer analysis of the data revealed details not found in earlier recordings. For the first
time during our project history we recorded sounds from EM events in the sky with three different
microphones. This enabled for the estimation of the source location. We have also tried to find other
possible sources and explanations for these clap sounds. Up to now the geomagnetic storm seems to
be the most probable activator of these EM events, even though many details still need more work.
We plan to continue with both acoustic and VLF analyses in order to arrive at a full and coherent
explanation for the observed phenomena. However, all current methods are pointing in the same
direction: EM events with sound emissions may occur during a geomagnetic storm in the lower
atmosphere. These sounds may be audible at the ground level under the source.
7. Acknowledgement
Many individuals and institutions have helped with these studies during the past 11 years. Es-
pecially, Ph.D. Esa Turunen and Ph.D. Jyrki Manninen from the Sodankylä Geophysical Observa-
tory (SGO) have worked in close affiliation with this project during its early phases. Support in the
form of geomagnetic data from the Finnish Meteorological Institute (FMI) as well as their Auroras
Now service is acknowledged. Also, communication with professor Harry J. Lehto from the Uni-
versity of Turku (Tuorla Observatory) has influenced the present work positively. Finally, I would
like to thank my home institution Aalto University for their support in the form of usage of their
high quality acoustic instruments.
REFERENCES
1
Laine, U. K., Denoising and Analysis of Audio Recordings made during the April 6-7 2000
Geomagnetic Storm by using a Non-professional ad hoc Setup, Proc. of Joint Baltic-Nordic
Acoustics Meeting 2004, 8-10 June 2004, Mariehamn, Åland, (2004)
2
Hautsalo, J., Study of Aurora Related Sound and Electric Field Effects, TKK (2005).
3
http://www.zoom.co.jp/english/products/h4n/ .
4
Manninen J. personal communication, Sodankylä Geophysical Observatory SGO, (2011).
5
PRAAT: doing phonetics with computer; http://www.fon.hum.uva.nl/praat/
6
Nakamura, Y., et al., Full wave analysis and observation of spherics generated by lightning,
AGU Fall Meeting Abstracts (2003).
