Transcript 5-08 Arrowsmith - Laboratory for Atmospheric Acoustics

Collaborators
 Rod Whitaker, George Randall [Los Alamos National
Laboratory]
 Relu Burlacu [University of Utah]
 Chris Hayward, Brian Stump [Southern Methodist
University]
Overview
 Motivation
 Signal Detection
 Association/Loca
tion
 Synthetic Tests
 InfraMonitor 2.0
 Application to
the Utah network
 Summary
Motivation
 Infrasound research has been largely event-driven by:
 Direct ground-truth
 Ground-truth from seismology, satellites
 There is a need for a fully-integrated technique for
automatic regional infrasound monitoring
 Infrasound Data  InfraMonitor  Event Catalogs
 Historically, techniques for processing infrasound data are
borrowed from seismology
 But, infrasound monitoring requires different strategies due
to unique challenges
 Temporal variability of medium
 Noise issues
Signal Detection
 The human eye is remarkably competent at detecting signals in noisy
data, automatic algorithms must attempt to match this level of
capability
 Requirement: Hypothesis that can be tested
 Standard hypothesis: Noise is spatially incoherent
 This is frequently violated, leading to large numbers of spurious ‘signals’
 This hypothesis does not adapt to variations in ambient noise
 We have developed coherent and incoherent detectors with the
following criteria:
 Does not require historical data
 Accounts for real ambient noise
 Can be applied operationally in near real-time
 Thus, a sensor or array can be deployed in a new region and the
automatic detector applied immediately
Signal Detection
•
Shumway et al. (1999): In
the presence of
stochastic correlated
noise, F-statistic is
distributed as:
cF2BT,2BT(N 1)
•
Where:

P 
c  1  N s 
Pn 

To estimate c (i.e., Ps/Pn),
adaptively fit F distribution
peak to Central F distribution peak while
processing data
•
•
Apply p-value detection
threshold (e.g., p = 0.01)
Signal
Detectio
Pinedale, Wyoming
n data
Adaptive window: 1 hour
Symbols: Adaptive detector (stars),
Conventional (circles), infrasound
(filled), seismic (open)
Adaptive window: 24 hours
Association/Location
 Seismic location techniques typically use an inverse
approach (Geiger’s method):
d  Gm
 This method requires a model
 Unfortunately, state-of-the-art 4D atmospheric models:
 Have notbeen validated at local or regional scales
 Do not always predict observed phases
 We have developed a new forward technique that:
 Places bounding constraints on location (producing
location polygons)
 Does not require a model
Association/Location
 The problem can be represented by the following equations:

Observations:
 t11 L t1 j
1

t  
M
 t L t
njn
 n1



o
11
  
 o
 n1
o

L
dt max
dt11max
  M
 max
dt k1
L
dt min
dt11min
  M
 min
dt k1






L 1oj1
M

L  njo n 

Predictions
:
 p(max)

 p(min)
11p(max)
  M
 p(max)
 k1
O
L
O
L
L
O
L
11p(min) L
  M O
 p(min)
L
 k1
max

dt1m
M 

max
dt km

min

dt1m
M 

min
dt km

1mp(max) 
M 

p(max)
 km

1mp(min) 
M 

p(min)
 km

 Where
 are n arrays, ji arrivals at the ith array, k grid nodes, and m pairs of
 there
arrays

 t and Φo are observed arrival times and backazimuths
at each array
 dtmin, dtmax, Φp(max), and Φp(min) are bounding constraints on
observations for a particular location (i.e., grid node)
Association/Location
 Consider a pair of arrays, Arrays 1 and 2, and
corresponding grid node, k:
 If we are searching for any phase within a specified group
velocity range (vmin – vmax), we must search for associated
arrivals where the apparent velocity (vapp) is, for all array
pairs:
d2  d1
d2  d1
 v app 
 d2   d1 
 d2   d1 
   


   
v min  v max 
v max  v min 
Synthetic Tests
•
•
•
Synthetic Tests provide
•
Test of
algorithm/code
•
assessment of
network resolution
In each panel
•
Stars show locations
of synthetic events
•
Gray regions show
localization
uncertainty
Search parameters
represent uncertainty in
propagation
Gray
regions
enclosed
by
ellipses
  6,vg  0.28  0.34km / s
  3,vg  0.32  0.34km / s
  1,vg  0.299  0.301km / s
InfraMonitor 2.0
 Features:
 GUI interface for interactive data analysis
 Command-line functions for batch data processing
 Seamless integration of detection, association, and
location methodologies
 CSS3.0 compatible
 Requirements:
 Matlab
 + Signal Processing Toolbox
 + Mapping Toolbox
 + Statistics Toolbox
InfraMonitor 2.0
Spectrogram
tool
Spectrum
tool
Detection
Processing
Main Window
Google Earth
functionality
F-K Tool
Utah Seismo-acoustic Network
 Operated by the University
of Utah Seismograph
Stations (UUSS)
 Designed to record
seismo-acoustic signals
from rocket motor
detonations in northern
Utah.
 The arrays are co-located
with UUSS seismic stations
 100 m aperture arrays
 Porous hoses for noise
reduction.
Infrasound + Seismo-acoustic
Events
 Duration of Study: 1 month
(Summer)
 Parameters optimized for
high-frequency arrivals
 287 infrasound events
 12 seismo-acoustic events
 Analyst Review of all 287
events indicates false alarms
make up <25% of the total
 4 ground-truth rocket motor
shots are all detected
seismo-acoustically
Infrasound Events
Ground-truth association of event locations with satellite imagery from
Google Earth
Event 1: Ground-truth Explosion
Event 2: Suspected Explosion
Topography blockage
At NOQ?
Event 3: Wells Earthquake
Summary
 New methods for detection and location of regional
infrasound events have been developed
 Detector: Accounts for temporally-variable correlated
noise
 Locator: Bounding approach does not require a model
 Techniques have been validated using synthetic tests and
Utah network data
 Analyst review of Utah events suggests a low false
association rate (<25 %)
 Events from earthquakes, explosions (military + mining),
and numerous other sources are detected
 InfraMonitor 2.0 integrates detection, association and
location algorithms seamlessly into a Matlab toolbox