Three-dimensional velocity models and probabilistic earthquake location

1
Title
Three-dimensional velocity models and
probabilistic earthquake location
Stephan Husen
Institute of Geophysics, ETH Zurich, Switzerland,
husen_at_sed.ethz.ch
with contributions from
Anthony Lomax
Scientific Software, Mouans-Sartoux, France,
anthony_at_alomax.net
Edi Kissling
Institute of Geophysics, ETH Zurich, Switzerland
2
Linearized earthquake location
Introduction
linearized earthquake location
Traditional earthquake location

• linearized methods (HYPO71, HYPOELLIPSE,
  HYPOINVERSE,..)

3D Velocity Model

• 1-D velocity models (plus station delays)

Probabilistic Earthquake Location

• error bars or error ellipses (linear)

Location Examples

• efficient

. but linearized methods and 1-D velocity models
are only approximations!
Conclusions
3
what do we need
Introduction
improvement
How can we improve the situation?
3D Velocity Model

• 3-D velocity models (Local earthquake
  tomography, controlled-source experiment)

Probabilistic Earthquake Location

• Non-linear earthquake location (NonLinLoc)

Location Examples
Conclusions
4
Example mine blast
Introduction
relocation mine blast
Mine Blast - True location is known
3D Velocity Model
Probabilistic Earthquake Location
True location
True location
Location Examples
Non-linear solution (3D)
Linear solution (1D)
Linear solution (1D)
Conclusions
5
Data quality
Introduction
3D Velocity Model
earthquake data in Switzerland
Probabilistic Earthquake Location
Location Examples
729 earthquakes with 10,044 P-observations
Conclusions
only highest quality data (impulsive onsets)
6
Moho topography
Introduction
3D Velocity Model
Moho topography
Probabilistic Earthquake Location
Location Examples
Waldhauser et al., 1998
3-D Moho topography beneath Switzerland as
determined by controlled-source seismology data
Conclusions
7
Min. 1D model
Introduction
3D Velocity Model
Subset of 200 earthquakes
min. 1D model
Simultaneous inversion for 1D velocity models,
hypocenter locations, and station delays
Probabilistic Earthquake Location
Location Examples
Software VELEST
Initial models
Final models
Conclusions
8
LET model
Introduction
3D Velocity Model
local earthquake tomography
Probabilistic Earthquake Location
Lower crust / Moho is not well resolved
crust is well resolved
Location Examples
3-D P-wave velocity model determined by local
earthquake data
Conclusions
Software SIMULPS14
9
CSS model
Introduction
3D Velocity Model
controlled-source data
Probabilistic Earthquake Location
Location Examples
Conclusions
3-D P-wave velocity model determined by
controlled-source seismology (CSS) data
10
Final model
Introduction
3D Velocity Model
final (combined) model
Probabilistic Earthquake Location
Location Examples
Lower crust / Moho is controlled by CSS data
crust is controlled by earthquake data
Conclusions
Final 3D P-wave velocity model determined by
earthquake data and controlled-source data
11
NonLinLoc
Introduction
Tarantola and Valette (1982)
3D Velocity Model
Posteriori Probability Density Function ?(x)
(PDF)
?(x) K?(x)exp-1/2misfitL2(x)
Probabilistic Earthquake Location
relies on known a priori information ?(x) on
model parameters and on observations.
software NonLinLoc
PDF is computed using global sampling techniques
- grid search or Oct-Tree importance sampling.
Location Examples
PDF gives complete location uncertainties.
Conclusions
Software NonLinLoc www.alomax.net/nlloc
12
Global sampling methods
Introduction
Grid-Search
3D Velocity Model
Probabilistic Earthquake Location
Grid-Search
Location Examples
complete mapping
Conclusions
inefficient and slow
13
Global sampling methods
Introduction
Grid-Search
Oct-Tree sampling
3D Velocity Model
Probabilistic Earthquake Location
Grid-Search vs. Oct-Tree sampling
Location Examples
complete mapping
importance sampling
Conclusions
inefficient and slow
efficient and fast
14
Location uncertainties
Introduction
Grid-search
Oct-Tree importance sampling
3D Velocity Model
Probabilistic Earthquake Location
solution and location uncertainties
Location Examples
confidence contours
scatter clouds
Conclusions
maximum likelihood hypocenter location
68 confidence ellipsoid
15
Example 1
Introduction
1987 05 08 0959 46.146 N 8.614 E 4.1km
3D Velocity Model
Nobs 8 RMS 0.04 s GAP 193 Dmin 1.9 km
Probabilistic Earthquake Location
Difference dx 1.0 km dy 5.2 km dz 0.1 km
Location Examples
non-linear uncertainties
SED error ERRH 1.9 km ERRZ 2.6 km
Non-linear(3D)
Conclusions
Linear(1D)
16
Example 2
Introduction
1993 04 15 1357 46.921 N 9.607 E -0.9 km
3D Velocity Model
Nobs 8 RMS 0.14 s GAP 164 Dmin 16.9 km
Probabilistic Earthquake Location
Difference dx 2.5 km dy 3.0 km dz 15.7 km
Location Examples
no control on focal depth
Non-linear(3D)
SED error ERRH 2.2 km ERRZ 2.6 km
Linear(1D)
Conclusions
17
Mine Blast
Introduction
1987 09 21 1948 46.254 N 7.899 E 4.1km
3D Velocity Model
Nobs 26(8) RMS 0.26(0.03) s GAP 73 Dmin 22.8
km
Probabilistic Earthquake Location
Difference dx 1.0 km dy 5.2 km dz 0.1 km
Location Examples
true location
mine blast
SED error ERRH 0.7 km ERRZ 2.0 km
Conclusions
18
Conclusions
Introduction
Conclusions
3D Velocity Model

• combination of local earthquake data and
  controlled-source data provides reliable 3-D
  velocity models

Probabilistic Earthquake Location

• probabilistic earthquake location combined with
  global sampling algorithms is efficient and
  reliable

• location uncertainties obtained by
  probabilistic earthquake location prove to be
  much more reliable, important for planetary data
  sets with few instruments

Location Examples
Conclusions
19
Conclusions
Introduction
Outlook

• application and tuning of existing geophysical
  methods to planetary data sets (real and
  synthetic) considering their peculiarities, i.e.
  small number of receivers

3D Velocity Model
Probabilistic Earthquake Location
Location Examples
Conclusions