Noise Based Detection Method for the ANSS

1
    Noise Based Detection Method for the ANSS
by Dan McNamara
With Collaborators Ray Buland, Harley Benz, Rob
Wesson Art Frankel and Dirk Erickson
2
Topics

• ANSS Probabilistic Noise Analysis
• Noise Based Detection Technique
• Detection System Applications
• ANSS Network Design Recommendations

3
ANSS Seismic Noise Monitoring
Motivation

• Establish ANSS station noise baselines
• ANSS backbone instrumentation
• ANSS backbone site criteria
• Network detection thresholds
• Station maintenance issues
• System transients
• Prioritize repairs
• Automate problem notification
• Cultural noise source modeling
• Microseism modeling

Hailey, ID 08/2001-05/2002
Cars
Approach
Local Quakes
Teleseisms
All data is included, no pre-screening for
quakes, data gaps, glitches, high noise
data. 2370 individual PSDs, binned in 1/8 octave
intervals, are used to construct a Seismic Noise
Probability Density Function for HLID BHZ.
Results
Realistic view of noise conditions at a station.
Not simply lowest levels experienced.
McNamara and Buland (2004) in press BSSA
4
Seismic Noise PDFsnoise as a function of
location and site type
Continental Interior Mine Site
Continental Interior Borehole
Idaho Springs, CO
Bozeman, MT
Western US rocks sites tend to have low noise
although the minimum is generally higher than the
NLNM.
5
Seismic Noise PDFsnoise as a function of
location and site type
Eastern US Surface vault Binghamton, NY
Island Site Borehole Big Island Hawaii
Higher noise across all bands in Highly populous
Eastern US.
Very high noise in microseism band But quiet at
long periods due to borehole.
6
Topics

• ANSS Backbone Probabilistic Noise Analysis
• Noise Based Detection Technique
• Brune source modeling method
• Comparison of Brune source modeling results with
  NEIC autopicker
• Detection System Applications
• ANSS Network Design Recommendations

7
Method to Compute Theoretical ANSS Detection
Threshold based on Brune Source Modeling.
For each 1 degree cell we model Brune sources
over a range of frequencies Brune (1970, 1971).
A detection is declared if the Brune source
P-wave amplitude exceeds our noise threshold at 5
ANSS stations.
Calculations
For each frequency (1/period) per cell.
Shear-wave moment (dyne-cm) Brune (1970, 1971).
fc10Hz Mw3.1
fc1Hz Mw5.1
Fault Dimension in cm
Mw 0.667 log(Mo) 10.7 (Kanimori,
1977) Compute shear-wave amplitude from
Mw (Brune 1970, 1971). Apply Q(f) models to
shear-wave amplitude. Convert to P-wave
amplitude. Convert velocity amplitude to dB for
noise comparison.
8
Topics

• ANSS Backbone Probabilistic Noise Analysis
• Noise Based Detection Technique
• Brune source modeling method
• Comparison of Brune source modeling results with
  NEIC autopicker
• Detection System Applications
• ANSS Network Design Recommendations

9
ANSS Detection Threshold Modeling Results
Brune minimum Mw, 80 noise threshold
Used 63 existing ANSS backbone stations with well
established noise baselines. Detection declared
if at least 5 stations in solution. Model shows
minimum Mw at regions where network is dense in
western and eastern US. Mw max occur in regions
of low station density. Model minimums 0.2
units higher than catalog. Model maximums 0.2
units lower than catalog. General pattern close
match.
Mw/mb
NEIC Autopicker Minimum mb.
10
ANSS Detection Threshold Modeling Results
Brune minimum Mw, PDF mode noise threshold
Mw/mb
PDF mode noise threshold pattern similar to 80th
with minimum Mw regions expanded. PDF mode
noise threshold demonstrates how lowering noise
can extend minimum detection threshold. Model
minimums 0.1 units higher than catalog. Model
maximums 1.0-1.2 units lower than
catalog. General pattern close match but overall
pattern better matched by 80th noise threshold.
NEIC Autopicker Minimum mb.
11
Brune source modeling not an exact match to NEIC
autopicker?

• MbMw bias?
• Simplistic application of Q models
• Noise baselines affected by system transients
• Incomplete and complicated autopicker catalog

12
mbMw Bias
UC Berkeley Northern CA Moment Tensor
Catalog 1988-2004
Mw1.46mb-2.42
For mb 5.5-7.3
Sipkin (2003)
No Magnitude bias at small mb
13
Brune source modeling not an exact match to NEIC
autopicker?

• MbMw bias?
• Simplistic application of Q models
• Noise baselines affected by system transients
• Incomplete and complicated autopicker catalog

14
New frequency Dependent Q Models
3Hz
Q
Considerable time spent modeling Lg amplitudes
for frequency dependent US Q. Erickson et al,
2004 McNamara et al 2004 Wesson and McNamara
2003. At this point Q(f) chosen by source
region. More realistic approach is to project
each path through Q model to more accurately
predict amplitudes. Should lead to better
modeling of Mw regional variations.
6Hz
15
Brune source modeling not an exact match to NEIC
autopicker?

• MbMw bias?
• Simplistic application of Q models
• Noise baselines affected by system transients
• Incomplete and complicated autopicker catalog

16
System Transients can have an effect on noise PDF
levels.
90th percentile and mode often track data
dropouts when frequent. Causing localized
detection anomalies.
Data Dropouts
17
Brune source modeling not an exact match to NEIC
autopicker?

• MbMw bias?
• Simplistic application of Q models
• Noise baselines affected by system transients
• Incomplete and complicated autopicker catalog

18
NEIC Minimum Auto Detection Catalog Issues
mb
Catalog possibly incomplete (only 20 months in
2002-2003) Possible false triggers at mb
minimums. Mine blasts that do not behave like
earthquakes at mb minimums. Multiple magnitude
types (mb, ml, mbLg) Therefore, difficult to
achieve exact match.
19
Brune source modeling not an exact match to NEIC
autopicker?

• MbMw bias?
• Simplistic application of Q models
• Noise baselines affected by system transients
• Incomplete and complicated autopicker catalog

Match good enough to play games with detections
and learn some things about the network!
20
Topics

• ANSS Backbone Probabilistic Noise Analysis
• Noise Based Detection Technique
• Detection System Applications
• Regional Network Evaluation
• Maintenance Prioritization
• ANSS Network Design
• ANSS Network Design Recommendations

21
Regional Network Simulation
6 stations from NM regional network with well
established noise baselines. Detection threshold
lowered in New Madrid region by 0.1-0.3
units with addition of NM network. Regional
Station Limitations - high noise in Cultural
noise band (1-10Hz) - PVMO instrumented with
Guralp CMG-3esp seismometer (50Hz) and Quanterra
Q-380 digitizer at 20sps. Power rolloff at
Nyquist10Hz.
Mw
PVMO
22
Topics

• ANSS Backbone Probabilistic Noise Analysis
• Noise Based Detection Technique
• Detection System Applications
• Regional Network Evaluation
• Maintenance Prioritization
• ANSS Network Design
• ANSS Network Design Recommendations

23
Detection Maps Used for Prioritization of
Maintenance Issues
Satellite GR4
ANSS backbone distributed over 2 satellites to
protect against total network outage. Maintenance
decisions could be made based on real-time
changes in detection thresholds. GR4 expected
to die within 3 years. Hughes states. There
will be a seamless transition to a new satellite
Mw
Satellite SM5
24
Topics

• ANSS Backbone Probabilistic Noise Analysis
• Noise Based Detection Technique
• Detection System Applications
• Regional Network Evaluation
• Maintenance Prioritization
• ANSS Network Design
• ANSS Network Design Recommendations

25
ANSS Site Location Planning
SNSD
PDF noise baselines used to estimate noise
characteristics in regions without existing ANSS
stations. Interpolate from nearby stations with
known noise baselines. With noise baseline
estimates we can calculate detection thresholds
for new network configurations.
SNSD
26
ANSS Site Location Planning
Mw
22 planned ANSS backbone stations added to
simulate future detection capabilities. Mw
threshold lowered in regions with sparse station
coverage such as the northern midwest and Texas.
27
Topics

• ANSS Backbone Probabilistic Noise Analysis
• Noise Based Detection Technique
• Detection System Applications
• Regional Network Evaluation
• Maintenance Prioritization
• ANSS Network Design
• ANSS Network Design Recommendations
• Lower station noise thresholds
• Supplement backbone with regional stations
• Install Planned ANSS stations
• Recording system limitations

28
ANSS Network Design Recommendations
Based on detection work, we can lower detection
thresholds across US.
80th Percentile Noise Level, Brune Mw
Mw
Decrease Station Noise Levels
Supplement with Regional Broadbands
Install Planned ANSS Stations
Minimum saturation occurs at Mw2.2-2.5 despite
network improvements.
29
Topics

• ANSS Backbone Probabilistic Noise Analysis
• Noise Based Detection Technique
• Detection System Applications
• Regional Network Evaluation
• Maintenance Prioritization
• ANSS Network Design
• ANSS Network Design Recommendations
• Lower station noise thresholds
• Supplement backbone with regional stations
• Install Planned ANSS stations
• Recording system limitations

30
NEIC Short Period Filter Limitations
Higher frequencies required to record full
amplitudes of smaller earthquakes. Recommendation
s 1. Get rid of SP filter. 2. Increase
sampling rate.
31
Detection Simulation with NEIC Filters Removed
Mw
Mw Thresholds lowered significantly across US
with the removal of NEIC Short period filter and
sampling rate increased to 200 sps. Noise levels
projected to higher frequencies. At 200sps
fny100Hz Mw2.0 fc35Hz Mw1.5 fc62Hz Mw1.0
fc111Hz
Difficulties Short period filters reduce false
triggers. New picker would need filters to deal
with false triggers while allowing high
frequencies through for small events.
32
Conclusions

• Detection System Useful for Several Applications
• Regional Network Evaluation
• Maintenance Prioritization
• ANSS Network Design
• ANSS Network Design Recommendations
• Lower station noise thresholds
• Supplement backbone with regional stations
• Install Planned ANSS stations
• Record higher frequencies