Towards a European Reference Model: The Role of Surface Waves from Earthquakes and Ambient Noise

1
Towards a European Reference Model The Role of
Surface Waves from Earthquakes and Ambient Noise
Mike Ritzwoller, Nikolai Shapiro, Yingjie Yang,
Anatoli Levshin, and Bob Engdahl University of
Colorado at Boulder IPGP, Paris
2
Outline

1. Contrasting traditional seismic models with
   Reference Models.
2. Role of surface waves in constructing a Reference
   Model.
3. Eurasian Model (CUB) from surface waves.
4. Validation of a Eurasian Model (CUB) using a
   Ground Truth (GT) data base.
5. Surface wave tomography from ambient seismic
   noise.
6. Conclusions.

3
Traditional Seismic Models
Data

• Constructed by a single scientist or a small
  group of collaborators.
• Intended predominantly for scientific purposes
• To test scientific hypotheses.
• To explore poorly understood regions.
• To express the scientific understanding or
  prejudice of its creator(s).
• Used predominantly by its creators or a small
  group of specialists.
• Constructed from a limited set of data typically
  of a single variety e.g., regional P-waves,
  receiver functions, two station phase speeds,
  single station group or phase speeds, etc.
• Intended to predict only a limited set of data
  types, usually the data types used in its
  creation.
• Tested (validated) only against data used in
  their creation.
• Typically cover only a limited region.
• Do not have uncertainty information.

Model
Quantities Predicted From the Model
4
Reference Models
Data

• Constructed by a community of researchers.
• Need to express the understanding of this
  community.
• Intended for scientific and non-scientific
  purposes (e.g., hazard assessment).
• Used by a large community, including
  non-specialists.
• Should be able to predict a wide variety of data
  types, so they should be constructed from a wide
  variety of information.
• Need to be validated by a variety of data,
  especially by data not used in their creation.
• Constructed over large regions (e.g., Europe).
• Must contain uncertainty information about
  themselves and the quantities intended to be
  predicted from them.
• Should be constructed to be the basis for
  non-scientific decisions with social and economic
  impact (e.g., hazard assessment).

Model
Quantities Predicted From the Model
Decisions
5
Outline

1. Contrasting traditional seismic models with
   Reference Models.
2. Role of surface waves in constructing a Reference
   Model.
3. Eurasian Model (CUB) from surface waves.
4. Validation of the Eurasian Model (CUB) using a
   Ground Truth (GT) data base.
5. Surface wave tomography from ambient seismic
   noise.
6. Conclusions.

6
The Use of Surface Waves in Constructing a
Reference Model
Advantages
Disadvantages

• Homogeneity of coverage over large regions.
• Sensitivity to Vs -- complementary to body
  waves.
• Constraints on crustal structure, coming from
  short period measurements.
• Good vertical resolution of the lithospheric
  mantle.
• Naturally results in a Monte- Carlo inversion
  that yields uncertainty information on the
  model.

• Relatively poor lateral resolution -- 100s of
  km.
• Little sensitivity to Vp.

• Short period measurements are hard to obtain.

• Little information below 200 km.

• Uncertainties in measurements are poorly
  known.

7
Outline

1. Contrasting traditional seismic models with
   Reference models.
2. Role of surface waves in constructing a Reference
   Model.
3. Eurasian Model (CUB) from surface waves.
4. Validation of the Eurasian Model (CUB) using a
   Ground Truth (GT) data base.
5. Surface wave tomography from ambient seismic
   noise.
6. Conclusions.

8
Global Dataset

• More than 200,000 individual paths across the
  globe resulting from about 5000 earthquakes.
• Group speed measurements from CU-Boulder, phase
  speed measurements from Utrecht Harvard.
• Emphasis on short periods to improve resolution
  of the crust from the mantle.
• Use of regional data (e.g., PASSCAL, European
  VBBN) improves resolution in some areas (e.g.,
  Europe).

9
Global Tomography Production of Group and Phase
Speed Maps
Dispersion maps result from a
linearized inversion, basis for a
following nonlinear 3D inversion, finite
frequency sensit- ivity kernels, 2 x 2 degree
grid world- wide, differ with period and
wave type, azimuthal anisotropy estimated at
the same time.
10
Continental Scale Tomography Across Eurasia
Reflects crustal thickness

• 1 x 1 degree grid.
• 16 - 150 sec period.
• Rayleigh and Love wave
• group and phase speed.
• Followed by a Monte-Carlo
• inversion for Vs structure
• of the crust and upper
• mantle.

percent perturbation
Reflects uppermost mantle Vs speeds
percent perturbation
11
Monte Carlo Inversion for a 3D Vs Model of the
Crust and Upper Mantle Process
Dispersion measurements and fit
Point in Iran (30,60)
Vs model in E. Iran Vsh Vsv in the
uppermost mantle.
12
Monte Carlo Inversion for a 3D Vs Model of the
Crust and Upper MantleHorizontal Slices
The model is most reliable in the upper mantle --
although there is a crustal part, too. Length
scales of resolved features are large.
Isotropic part of model -- perturbation to ak135
()
13
Monte Carlo Inversion for a 3D Vs Model of the
Crust and Upper MantleVertical Slices
14
Outline

1. Contrasting traditional seismic models with
   Reference Models.
2. Role of surface waves in constructing a Reference
   Model.
3. Eurasian Model (CUB) from surface waves.
4. Validation of the Eurasian Model (CUB) using a
   Ground Truth (GT) data base.
5. Surface wave tomography from ambient seismic
   noise.
6. Conclusions.

15
Validating the Eurasian Reference Model (CUB)

• Should validate with data other than those used
  to construct the model.
• Particularly good test
• Determine the fit to groomed regional body
  wave data for accurately located events
  (earthquakes, explosions).
• Difficulties
• identifying events that are located to better
  than 5 km (GT 5 events),
• constructing a regional data set of groomed Pn
  and P travel times associated with these events.

16
Ground Truth Data Base of Endgahl and Bergman
Engdahl and Bergman identified 23 event
clusters with more than 1000 well located events
16 clusters are GT5 (5 explosions 11
earthquakes) and 7 clusters are GT10 (1
explosion, 6 earthquakes). Each cluster
surrounds an exceptionally well located
Reference Event. The Hypocentroidal
Decomposition Method is used to find the
differential location of each event in the
cluster and then shift it to an absolute location
based on comparison with the Reference Event.
17
Ground Truth Data Base of Endgahl and Bergman
Data Set of Engdahl and Bergman 1000 GT5 or
GT10 events. Large groomed travel time data
set of regional Pn and P arrivals (13,601 Pn,
2396 P), summarized into empirical path
corrections (836 Pn, 178 P).
18
Validating the Eurasian Model (CUB)
To test the fit to the groomed regional P and Pn
data set and relocate the GT events first
convert the Vs model to Vp use a modified
theoretical thermo-elastic conversion
(Sobolev et al., 1996 Goes et al., 2000),
trace rays with a 2D ray tracer. (Based on
Cerveny Psencik, 1984).
19
Validating the Eurasian Model (CUB)
To relocate the GT events using the groomed
regional P and Pn data set first convert
the Vs model to Vp use a modified theoretical
thermo-elastic conversion (Sobolev et al.,
1996 Goes et al., 2000), trace rays with a
2D ray tracer. (Based on Cerveny Psencik,
1984).
CUB model
ak135
ak135
CUB model
20
Validating the Eurasian Model Fit to the
Empirical Path Corrections for an Explosion
Fit to the empirical path corrections can be
visualized with a correction surface. Color
contours are predicted travel times for Pn from
CUB relative to the 1D model ak135. Colored
symbols are the empirical path corrections of
Engdahl Bergman.
Explosion Better than GT5
21
Validating the Eurasian Model Fit to the
Empirical Path Corrections for an Explosion
Explosion Better than GT5
Summary misfit (N71) 0.7 sec wrt CUB
1.53 sec wrt ak135. Var Red 79.
22
Validating the Eurasian Model Fit to the
Empirical Path Corrections for a GT5 Earthquake
Earthquake GT5
Summary misfit (N101) 1.40 sec wrt CUB
1.88 sec wrt ak135. Var Red 45.
23
Validating the Eurasian Model Fit to the
Empirical Path Corrections Overall
Summary misfit 5 Explosion clusters CUB ak135
Var Red 1.02 s 1.73 s 65 11 GT5 Earthquake
clusters CUB ak135 Var Red 1.24 s 1.69 s 46
24
Validating the Eurasian Model Conclusions

• GT data bases of regional body wave phase data
  following exceptionally well located events are
  a valuable resource
• to validate Reference Models.
• Assuming picking error is about 0.5 sec,
  large-scale 3D models such
• as CUB consume about half the misfit to regional
  P-wave
• phases. The remainder will come largely from
  further improvements in the 3D model e.g.,
  smaller scale structures,
• better crustal model, improved inter-conversion
  between
• Vs and Vp, etc.
• Not Shown The CUB model locates GT1 events
  using regional
• (non-teleseismic) data to about 5 km, whereas
  the 1D model
• ak135 locates them to about 10 km.

25
Outline

1. Contrasting traditional seismic models with
   Reference models.
2. Role of surface waves in constructing a Reference
   model.
3. Eurasian Model from surface waves.
4. Validation of the Eurasian Model using a Ground
   Truth (GT) data base.
5. Surface wave tomography from ambient seismic
   noise.
6. Conclusions.

26
Surface Wave Tomography from Ambient Seismic Noise

• Summary of the measurement procedure.
• Examples of cross-correlograms.
• Group speed maps from 12 - 40 sec.
• Determination of reliability
• Consistency with known structures,
• (e.g., Sedimentary basins, crustal thickness.
• Coherence among the measurements.
• Repeatability (not shown here).

R. Weaver, Science, 2005
27
Station Coverage for Ambient Noise Tomography
Across Europe
Stations from the Virtual European Broad-Band
Seismic Network (VEBSN).
125 stations
28
Measurement Procedure
N. Germany To N. Italy

• For each station pair, perform a
• series of narrow band-pass filters on each day
  of data
• 5-15, 10-25, 20-40, 33-66, 50-100, 70-150 sec.
• Perform temporal and spectral
• whitening of each time series.

29
Measurement Procedure
N. Germany To N. Italy

• For each station pair, perform a
• series of narrow band-pass filters
• 5-15, 10-25, 20-40, 33-66, 50-100, 70-150 sec.
• Perform temporal and spectral
• whitening of each time series.
• Stack results in daily, monthly,
• tri-monthly, yearly increments.

Symmetric component of 1 year stack.
30
Measurement Procedure
N. Germany To N. Italy

• For each station pair, perform a
• series of narrow band-pass filters
• 5-15, 10-25, 20-40, 33-66, 50-100, 70-150 sec.
• Perform temporal and spectral
• whitening of each time series.
• Stack results in daily, monthly,
• tri-monthly, yearly increments.
• Measure surface wave dispersion in each period
  band.

Predicted curve from CUB model
31
Example of Broad-Band Cross-Correlograms
Path N. Germany to Romania
1-year stack
time (sec/100)
32
Sample Record Section
N. Italy to Stations Across Europe
33-66 sec, 1 year stack, symmetric component
33
Group Speed Maps Across Europe 12 sec
From CUB 3-D Model
34
Group Speed Maps Across Europe 12 sec
Ambient Noise Tomography
SNR gt 5 1664 paths
35
Group Speed Maps Across Europe 16 sec
From CUB 3-D Model
36
Group Speed Maps Across Europe 16 sec
Ambient Noise Tomography
16 sec
SNR gt 5 3241 paths
37
Group Speed Maps Across Europe 20 sec
From CUB 3-D Model
38
Group Speed Maps Across Europe 20 sec
Ambient Noise Tomography
SNR gt 5 3057 paths
39
Group Speed Maps Across Europe 30 sec
From CUB 3-D Model
40
Group Speed Maps Across Europe 30 sec
Ambient Noise Tomography
SNR gt 5 2450 paths
41
Group Speed Maps Across Europe 40 sec
From CUB 3-D Model
42
Group Speed Maps Across Europe 40 sec
Ambient Noise Tomography
SNR gt 5 2760 paths
43
How do we Know if These Results are an
Improvement Over Traditional Earthquake
Tomography?
Various lines of evidence

• Agreement with known structures.
• e.g., sedimentary basins, crustal thickness.
• Repeatability of measurements.
• Seasonal variability -- work in progress.
• May yield uncertainty estimates on the
  measurements.
• Not reported here.
• Coherence of measurements.
• Fit to ambient noise measurements during
  tomography, compared with fit to earthquake based
  measurements during tomography.

44
Agreement with Location of Sedimentary Basins?
Observed 16 sec
Many of the basins across Europe are reflected in
the short period dispersion maps (e.g., 16 sec
here) N. Sea Basin, Silesian Basin (N.
Germany, Poland), Panonian Basin (Hungary,
Slovakia), Po Basin (N. Italy), Rhone
Basin (S. France), Basins in Adriatic and
Mediterranean Seas.
From Crust1.0, Laske et al.
45
Agreement with Expected Crustal Thickness?
Observed 30 sec
Low speed anomalies across Europe are associated
with mountains belts, consistent with thickened
crust e.g., Alps, Balkans, Carpathians.
From Crust2.0, Laske et al.
46
Coherence Among Measurements -- 12 sec period?
As measured by the ability to fit data sets when
doing tomography..
Misfit to Earthquake Measurements From
Earthquake Tomography
Misfit to Ambient Noise Measurements From
Ambient Noise Tomography
st dev 28.9 sec
st dev 15.0 sec
misfit (sec)
misfit (sec)
47
Coherence Among Measurements -- 16 sec period?
As measured by the ability to fit data sets when
doing tomography..
Misfit to Ambient Noise Measurements From
Ambient Noise Tomography
Misfit to Earthquake Measurements From
Earthquake Tomography
st dev 12.6 sec
st dev 22.7 sec
misfit (sec)
misfit (sec)
48
Coherence Among Measurements -- 20 sec period?
As measured by the ability to fit data sets when
doing tomography..
Misfit to Earthquake Measurements From
Earthquake Tomography
Misfit to Ambient Noise Measurements From
Ambient Noise Tomography
st dev 12.0 s
st dev 21.7 s
misfit (sec)
misfit (sec)
49
Coherence Among Measurements -- 30 sec period?
As measured by the ability to fit data sets when
doing tomography..
Misfit to Ambient Noise Measurements From
Ambient Noise Tomography
Misfit to Earthquake Measurements From
Earthquake Tomography
st dev 12.2 s
st dev 18.1 s
misfit (sec)
misfit (sec)
50
Coherence Among Measurements -- 40 sec period?
As measured by the ability to fit data sets when
doing tomography..
Misfit to Ambient Noise Measurements From
Ambient Noise Tomography
Misfit to Earthquake Measurements From
Earthquake Tomography
st dev 8.2 s
st dev 12.4 s
misfit (sec)
misfit (sec)
51
Coherence Among Measurements -- Summary
As measured by the ability to fit data sets when
doing tomography..
Dispersion measurements from ambient noise are
more internally consistent than
measurements following earthquakes earthquake
measurements are difficult to obtain
below 20 sec, source processes,
mislocation, etc. are eliminated. Above 30
sec, earthquake measurements are about as
reliable as ambient noise measurements and the
data sets can be combined without degrading the
ambient noise measurements.
earthquakes
ambient noise
52
Ambient Noise Tomography Across Europe

• Below 40 sec period, ambient noise surface wave
  dispersion measurements across Europe are
  generally more reliable than earthquake
  dispersion measurements.
• This is particularly true below 20 sec period,
  where dispersion provides information about
  crustal structure.

Earth is a noisy place!
53
General Conclusions

• GT data bases (e.g., Engdahl Bergmans) will be
  exceptionally valuable to validate an emerging
  European Reference Model.
• Surface waves are very useful in the construction
  of reference models generally. Large-scale 3D
  models (e.g., CUB) of the crust and upper mantle
  constructed from surface wave data provide a good
  start toward the European Reference Model.
• Ambient noise tomography promises to improve
  information from surface waves generally,
  particularly about the crust.
• In the context of the growing VEBBN, ambient
  noise tomography will prove to be a powerful
  method to aid in the development of the European
  Reference Model.

54
Validating the Eurasian Model Relocating the
GT5 Explosions
16 nuclear explosions at the Lop Nor Test Site
were relocated using the CUB 3D Model and the 1D
model ak135.
Mislocation of Lop Nor events 6.1 km with
CUB 9.6 km with ak135
Summary mislocation of all 38 explosions 5.1
km with CUB 12.3 km with ak135
55
Validating the Eurasian Model Relocating the
GT5 Earthquakes
34 GT5 earthquakes near Racha, Georgia were
relocated using the CUB 3D Model and the 1D model
ak135.
Mislocation of Racha events 8.2 km with CUB
21.8 km with ak135
Summary mislocation of all 207 earthqks 7.2
km with CUB 10.9 km with ak135
56
Principal Conclusions

• Earths Hum (bass section normal mode studies
  250 s) possesses tenor, alto, and soprano
  sections (lt 5 sec to gt 100 sec) useful for
  surface wave tomography.
• 2. Surface wave Green functions are extracted by
  cross-correlating long noise sequences at two
  stations.
• 3. Ambient noise tomography improves the
  resolution of the crust and upper mantle.
  (USArray)
• 4. Source of this noise (our signal) appears to
  be oceanic.
• 5. This noise is not truly ambient, but
  possesses a strong directional component from
  off-shore storms.

R. Weaver, Science, 2005
57
Earth is a noisy place!
58
Measurement Method Applied to a Pair of North
American Stations
59
(No Transcript)
60
(No Transcript)
61
(No Transcript)
62
(No Transcript)
63
(No Transcript)
64
(No Transcript)
65
(No Transcript)
66
(No Transcript)
67
(No Transcript)
68
(No Transcript)