Presented -

Design and Operation of Infrasound Stations
for Hazardous Weather Detection
David Pepyne, Sean Klaiber, Jerry Brotzge, and Michael Zink
Presented at the European Geosciences General Assembly
Vienna, Austria, 24 April 2012
Author Affiliations: All authors are with the NSF Engineering Research Center for
Collaborative Adaptive Sensing of the Atmosphere; Pepyne, Klaiber, and Zink are with the
University of Massachusetts, Amherst, MA, USA; Brotzge is with Oklahoma University,
Norman, OK, USA
Corresponding Author: David Pepyne ( [email protected] )
1. Introduction
 Each year tornadoes cause property damage and death, some of
which might be avoided with increased tornado warning lead time.
In 2011 there were 1691
tornadoes in the U.S. resulting in
550 deaths and property
damages approaching $25
3 April 2012 Dallas Texas: Given more
lead time, could these trucks have been
moved out of harm’s way?
 Today, tornado warnings are issued based on
observations from ground-based Doppler weather
 In the U.S., from the WSR-88D, NEXRAD, Sband radar.
 Radar beams travel essentially straight lines over a curved earth. Thus, far from the
radar, the severe weather that impacts human activities can lie below the view of a
long range radar. With respect to tornadoes, the result is that:
 >25% of tornadoes go undetected
 80% of tornado warnings are false alarms
 Average tornado warning lead time is 12 minutes.
NEXRAD coverage at 1 km AGL
 The Center for Collaborative Adaptive Sensing of the Atmosphere (CASA) is one of
15 U.S. National Science Foundation Engineering Research Centers.
 Funded for 10 years to solve the systems engineering challenges associated with
the deployment of sensor networks for improving weather hazard prediction,
detection, tracking, and response. CASA’s main research focus is related to
small, low-cost radars to fill the low-altitude coverage gaps [1].
Small, mechanically scanned
dish radars on short towers and
Small, phased-array flat-panel radars on cell
towers and other existing infrastructure.
CASA – fill the gaps by putting
radars in the white spaces: instead
of 150 large radars, 1000+ small
[1] McLaughlin et al, “Short-Wavelength Technology and the Potential for
3 Distributed Networks of Small Radar Systems,” Bulletin of the American
Meteorology Society, 90, pp. 1797-1817, 2009.
 Recently, CASA has extended its research to investigate improved weather
hazard warnings through a Networks of Networks (NoNs) approach.
 Additional sensors to include advanced barometers, LIDAR, mesonet
stations, rain gauges, GPS water vapor estimation, etc.
 Work by Bedard and others showed that
tornadoes generate infrasound in the 0.510Hz frequency range [2],[3].
 Precursor infrasounds are sometimes
detected 30min to 1hr prior to vortex
touchdown. This is 15-45min earlier
than the current average tornado
warning lead time in the U.S.
 Integrating infrasound into the
weather hazard warning
infrastructure, therefore, has the
potential to improve tornado warning
and response.
[2] Bedard et al, “The Infrasound Network (ISNET): Background, Design, Details, and Display Capabilities as an 88D
Adjunct Tornado Detection Tool,” Proc. of the 22nd AMS Conf. on Severe Local Storms, Hyannis, MA, 4-8 Oct. 2004.
4 [3] Bedard et al, “Overview of the ISNET Data Set and Conclusions and Recommendations from a March 2004 Workshop
to Review the ISNET Data,” Proc. of the 22nd AMS Conf. on Severe Local Storms, Hyannis, MA, 4-8 Oct. 2004.
 In the spring of 2011, CASA conducted an infrasound field experiment in the U.S.
state of Oklahoma (see [4], [5] for background and preliminary results).
 Two infrasound monitoring stations placed in the heart of “tornado alley” where
statistically the majority of tornadoes in the U.S. occur.
 This being CASA’s first infrasound experiment, the goals were to understand the
issues involved in the design and operation of infrasound stations for severe
weather monitoring and early warning.
2011 CASA Infrasound
Field Experiment
[4] Pepyne et al, “An Integrated Radar-Infrasound Network for Meteorological Infrasound Detection and Analysis,”
Proc. 91st AMS Annual Meeting, IOAS-AOLS, Seattle, WA, January 2011.
[5] Pepyne and Klaiber, “Highlights from the 2011 CASA Infrasound Field Experiment,” Proc. 92 nd AMS Annual
Meeting, IOAS-AOLS, New Orleans, LA, January 2012.
2. Infrasound Station Deployment
 Infrasound is low frequency sound, < 20Hz.
 At such low frequencies, sound experiences very little attenuation and
can be detected at very long distances.
 Propagation speed, however, is fairly slow, ~340m/s.
 For hazard early-warning, therefore, stations should be no more than
~100km from the emitter: 100km @ 340m/s = ~5min propagation delay.
 Moreover, because the speed of sound
varies with temperature, and
temperature first decreases then
increases with altitude, the atmosphere
acts as an acoustic “waveguide”.
 Waveguide effect can produce an
“acoustic shadow” starting ~30km from
the emitter.
Acoustic Shadow
Figure from: Skowbo, Muh, and Fanto, “Infrasound Sensor System Detection of Atlas V Rocket Launch,” Technology
Review Journal, Spring/Summer, 2009.
 The vision for comprehensive monitoring is a dense network of infrasound
stations spaced ~30km apart.
 This would ensure detection at two or more stations with a propagation
delay < 2min after initial infrasound emission.
 With detection at two or more stations, triangulation is then be used to
determine the emitter location.
30 km
Triangulation of a tornado within a unit cell of a network of
infrasound stations. 30km spacing between infrasound stations
ensures detection < 2 minutes after initial infrasound emission.
3. Station Design
 30km spacing implies that thousands of infrasound stations would be
needed to cover the region east of the U.S. Rocky Mountains, the area of
the U.S. most prone to tornadoes.
 With thousands of stations, each must
be very low cost:
 Low equipment acquisition and
Sensors +
deployment cost
wind filters
 Small land use footprint
 Easy to maintain and operate
Data cables
 An infrasound stations consists of,
 Infrasound sensors
 Wind filtering equipment
 Data logging and communication
Data logging +
 The infrasound sensors selected were Paroscientific, Inc., Model 6000-16B
 Convert the vibration frequency of a quartz crystal
into a pressure measurement.
Have a “nano-resolution/infrasound” mode giving
high-resolution measurements at sample rates
>20Hz. A built-in, user configurable anti-alias filter
removes high-frequency noise prior to sampling.
No need for expensive calibration. Factory
calibration is maintained over the lifetime of the
Internal circuitry maintains measurement accuracy
with temperature change, so the barometers do not
need to be placed inside a temperature controlled
Digital RS232/RS485 ASCII output makes data
acquisition trivial. One can do experiments with a
laptop, powering the barometers through a USB
power cable.
Being a barometer, these instruments can sense the
whole range of infrasound frequencies from DC
(absolute barometric pressure) up.
Model 6000-16B
4 Paroscientific
barometers controlled
from a laptop through a
USB-to-serial MUX
 An infrasound station needs at least 3 sensors to be able
Wind Filters
to determine the arrival direction of an infrasound signal.
 More sensors give better sensitivity and better direction-of-
arrival accuracy over a wider range of emitter frequencies.
 Each infrasound sensor needs a filter to mitigate wind
 Currently, the best solutions are analog spatial filters
consisting of pipes or hoses laid out on the ground around
each sensor. For very high wind speeds it may be
necessary to put fences around the spatial wind filter to
slow the wind and lift it off the surface.
 The lower the frequency of interest, the larger the required
diameter of the spatial filter.
Pipe Arrays
 Sensors are small and relatively inexpensive, wind filters
are physically large and must be protected from grazing
animals (e.g., cows).
 The large land area required for the spatial wind filters can
be a major impediment to deployment.
Cost of infrasound stations is driven by the number
of sensors – one must trade-off performance and
Soaker Hoses
Figures from: Hedlin, M.A.H., J. Berger, and M. Zumberge, “Evaluation of Infrasonic Spatial Filters.”
 To control costs, we used 4 sensors arranged in a square topology with 50m
spacing between sensors.
 Has redundancy in that the failure of a single sensor will not prevent
being able to estimate infrasound direction-of-arrival.
 The limitation of using a sensor array with a fixed size aperture is that
direction-of-arrival estimation accuracy will then depend on the frequency of
the infrasound signal.
 Spatial aliasing occurs for infrasound frequencies with wavelength > 2
times the sensor spacing. In our case, spatial aliasing occurs for
frequencies > ~3.4Hz.
Array gain patterns for a 4 sensor square array 50m on a side as a function of
frequency. The figure on the far right shows the spatial aliasing that occurs at
infrasound frequencies above ~3.4Hz.
 For wind filtering, we used 8-arm, 15m diameter
porous “soaker” hose wind filters.
 Such filters are inexpensive and very easy to deploy.
They may, however, require fencing to protect them
from grazing animals, such as cows.
 Such filters are also limited in that there are no
mathematical models for predicting their performance
and they may give inconsistent, time-varying results as
they get wet and as the soaker hose material
degrades with time.
Close-up of soaker hoses, summing
manifold, and barometer enclosure.
8-arm, 15m diameter
soaker hose wind filter
Wind filter at one of our Oklahoma stations surrounded by a
cow fence. The orange bucket in the center contains the
barometer. The white sensor on the pole measures wind speed
and direction.
4. Results
 Two infrasound monitoring stations were
 One in Cyril, Oklahoma (designated
KCYR) and one in Rush Springs,
Oklahoma (designated KRSP).
 The two stations were ~30km apart.
 The infrasound monitoring stations were
collocated with two of CASA’s
mechanically scanned X-band weather
 71 days of pressure data was collected at
both stations from 18 April to 27 June 2011.
 20 days of wind speed and direction data
was collected at the Cyril station from 2
June to 21 June 2011.
Rush Springs
 There were 1691 tornadoes in the U.S. in 2011.
 One tornado detection case occurred on 22 May 2011.
 The same night as the Joplin, MO EF5 tornado (159 dead; thousands injured).
 The spectrogram data was unremarkable,
 The cohereogram data showed strong coherence in the frequency band <
 Bearing angle estimation analysis showed
strong correlation, with direction-of-arrival from
the southeast relative to both stations.
 On a map, the stations point directly towards a line of tornadoes
moving north-northeast from Texas.
Radar image from:
 In addition to severe weather detections, we got hundreds of hours
of “banded” spectrogram data.
 This spectral banding was coherent between the sensors at the
arrays. In general, the signal was stronger at the Cyril station than at
the Rush Springs station.
 Our first thought was that the banding might be due to resonance in
the spatial wind filter hoses.
 Pipe arrays are prone to resonance; porous hoses less so.
 This thought was not supported by a bearing angle analysis.
 Bearing angle analysis generated solutions with a very focused
and consistent direction-of-arrival.
 Triangulation using the mean bearing estimates from the two
infrasound stations point directly to the Blue Canyon windfarm as
being the source of the signal.
 Blue Canyon is the largest windfarm in the state of Oklahoma,
containing more than 100, 1+ megawatt wind turbines.
Blue Canyon Windfarm ~30km
west of the Cyril (KCYR) station
and ~60km west of the Rush
Springs (KRSP) station.
 In Oklahoma alone, there were some 69 tornadoes during the period of
infrasound collection.
 Including an EF4 that occurred less than 30km from the two stations
(the Chickasha/Newcastle tornado outbreak on 24 May 2011).
 Of these, we recorded only 2 clear tornado cases.
 Neither from a tornado in the 30km “early-warning” range of the
infrasound stations.
 The cause was poor wind filter performance.
 Using the 21 days of wind speed and direction data recorded at the
Cyril station, we performed a wind filter analysis.
Wind speed and direction data recorded at the
Cyril (KCYR) station over the period 2 June to
22 June 2011.
 As seen from the power spectral density plots, our soaker
hose wind filter was ineffective for wind speeds > ~5m/s.
Above 7m/s can no
longer see the
windfarm signal.
Winds when tornadoes
are nearby can be
expected to be much
higher, 10-15m/s.
Need ~20dB additional
noise reduction at
5. Summary
 This poster described CASA’s initial efforts towards low-cost, but effective infrasound
monitoring stations.
 Four Paroscientific Model 6000-16B barometers arranged in a square topology
with 50 meter aperture appears suitable for the detection and geolocation of
severe weather infrasound emitters.
 Wind filtering is the biggest challenge and biggest impediment to successful
infrasound monitoring.
 Our simple soaker hose wind filter was almost completely ineffective for wind
speeds above ~5m/s.
 Bedard in his tornado detection work found a combination of soaker hose spatial
wind filters + eddy fences was required.
 The land use, permitting and construction costs of such fencing, however, not only
more than doubles the cost of a deployment, it also severely limits where the
infrasound stations can be deployed.
 Even in Oklahoma, with its wide open parries, we are having trouble getting the land
use permissions to experiment with eddy fencing!
Eddy fence used by Bedard for infrasound
tornado detection – 2 meters tall, 15 meters
diameter, covered with semi-porous mesh, and
22 capped with a plywood crown. Need one around
every sensor
 On-going work is looking into other ways, besides analog spatial filtering,
to deal with wind noise.
 These include techniques based on adaptive filtering, adaptive noise
cancellation, and techniques for detecting sinusoids and impulses in power
law noise (e.g., wavelet based techniques) to name a few.
 Other applications
 Wind energy is growing in popularity around the world, as are concerns about
the acoustic emissions from wind turbines.
 In the U.S., most acoustic emissions studies are only concerned with audible
noise (> 35Hz) at a given offset from the wind turbine.
 The windfarm detected in our experiment here had its fundamental at 0.8Hz.
 What are the infrasound emission levels at these frequencies at the noise
offsets deemed suitable for audible emissions?
 Do the infrasound emission at these levels have negative human
 For these studies too, wind filtering will be an issue.
 What is a simple wind filter design for wind turbine infrasound
measurement studies?
 Funding for this work is provided by the Jerome M. Paros fund for
Measurement and Environmental Sciences Research; the Engineering
Research Centers program of the National Science Foundation (NSF)
under NSF Cooperative Agreement EEC-0313747; and by the NSF
Research Experience for Undergraduates (REU) program.

similar documents