SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 1

Source Inversion Validation (SIV):Quantifying Uncertainties in Earthquake Source Inversions

Martin MaiMorgan Page

Danijel Schorlemmer

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 2

Finite-source inversion are done Finite-source inversion are done almost routinely today, using a almost routinely today, using a variety of inversion / modeling variety of inversion / modeling approaches, different data sets approaches, different data sets and processing stepsand processing steps

We use the slip models to infer We use the slip models to infer rupture dynamics, to devise rupture dynamics, to devise source-characterization source-characterization methods for ground-motion methods for ground-motion simulations, to perform simulations, to perform Coulomb stress modeling, to ….Coulomb stress modeling, to ….

But: how “good”, i.e. But: how “good”, i.e. reliable and robust, are reliable and robust, are these rupture models ?these rupture models ?

Earthquake rupture modelsEarthquake rupture models

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 3

In cases where multiple slip-inversion solutions exist for a single earthquake In cases where multiple slip-inversion solutions exist for a single earthquake we often find striking differences in the slip maps!we often find striking differences in the slip maps!

What drives the large differences between these slip models? What drives the large differences between these slip models?

A suite of models for the 1999 Izmit (M 7.5) earthquake

Intra-event variabilityIntra-event variability

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 4

In cases where multiple slip-inversion solutions exist for a single earthquake In cases where multiple slip-inversion solutions exist for a single earthquake we often find striking differences in the slip maps!we often find striking differences in the slip maps!

What drives the large differences between these slip models? What drives the large differences between these slip models?

A suite of preliminary investigations of the large Chile earthquake

Intra-event variabilityIntra-event variability

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 5

Initial Project:SPICE BLIND TEST

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 6

Source geometry and station distribution similar to the 2000 Tottori earthquake Synthetic seismograms for 19 (33) near-fault sites (COMPSYN, fmax ~ 3 Hz) Known: seismic moment: 1.43 x 1019 Nm, geometry (strike, dip, rake: 150°, 90°, 180°),

hypocentral location and depth (Z= 12.5 km), velocity-density structureUnknown: slip on fault plane, rupture velocity & rise time (both constant)

SPICE: Blind Test on Source InversionSPICE: Blind Test on Source Inversion

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 7

9 groups; the slip models from 5 groups are “visually” similar to the input model waveform fits in all cases implied visually a “very good fit” ….

SPICE: Blind Test on Source InversionSPICE: Blind Test on Source Inversion

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 8

Outcome of the blind test was unexpectedOutcome of the blind test was unexpected

Despite the “simplicity” of the input model, inversions could not resolve slip very well; uncertainties in rupture velocity and rise time up to 20%

Despite differences among all inversion solutions, predicted waveforms are remarkably similar (f < 1 Hz), resulting in low misfit values (generally L2-norm)

4 out of 9 inversion results are, statistically speaking, NOT better than a random model with somehow correlated slip!

Issues in the inversion method? Issues in the parameterization? Issues in the provided synthetics (“correct” solution) Issues in the “basics”: Green’s function computation

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 9

Source Inversion Validation

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 10

SPICE source inversion blindtest

March 2005 – Dec 2007; 9 participating groups

Special AGU session (Dec 2007) with invited speakers

Problems: short-lived; no funding; suggested data formats where not used, and

hence lots of manual labor to generate comparisons

SCEC workshop on earthquake source inversion (Sept. 2008)

~50 participants, 6 invited speakers, and ~3 hrs intense discussions

General consensus that SIV (Source Inversion Validation) has to continue

Collection of general ideas on how to setup the problems and how to organize

ourselves … but no formal decisions or “constitution of a core group”

Past Activities within the SIV InitiativePast Activities within the SIV Initiative

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 11

Proposal to SCEC for financial support for 2009 SCEC-SIV workshop

Dedicated webpage launched (March 2009)

Online platform to distribute the inversion problems and all relevant meta-data

General communication & exchange platform for everyone interested in SIV

Mini-workshop during SSA 2009 (April 2009)

~20 participants for general ~2 hrs discussion on future activities

Step 0: Setup of Green’s function test and initial forward-modeling exercise

Workshop during the Annual SCEC meeting (Sept 2009)

Talks on uncertainty assessment, Bayesian modeling, robustness, source dynamics

Discussions on “Expectations from forward-modeling exercise” and “Simple

inversion exercise”

Discussion on implementation (CSEP-like approach ?)

http://siv.usc.edu

Past Activities within the SIV InitiativePast Activities within the SIV Initiative

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 12

Workshop at KAUST, (March 22-24, 2010)

~20 non-KAUST & 10 KAUST-affiliated participants

3 days of talks and discussions on

Green’s function validation

New approaches to (source) inversion and validation

Addressing uncertainties and differences in source models

Using source models in subsequent work

Computational aspects & testing center

Past Activities within the SIV InitiativePast Activities within the SIV Initiative

How to enjoy snorkeling and not to get

sea-sick in unexpectedly rough waters in

the Red Sea

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 13

Saturday, September 11, 2010 (1:30-5:30pm) Session I: Review of SIV Activities and Green's Function Test Results 1:30 – 1:45 Mai, Page, Schorlemmer: SIV Introduction 1:45 – 2:05 Causse, Mai, and all participants: Results from Green's function test 2:05 – 2:35 Shao & Ji : What did the Exercise of SPICE Source Inversion Validation

BlindTest not Tell You? 2:35 – 3:45 Discussion of Green’s function Results3:45 – 4:00 break Session II: Inversion Techniques & Slip Models 4:00 – 4:20 Song: Does earthquake slip follow Gaussian statistics? 4:20 – 4:50 Lavallee, Archuleta, Schmedes: Spectral analysis of slip spatial distributions 4:50 – 5:30 Discussion of previous talks & plans for Sunday breakout groups

Program for this workshopProgram for this workshop

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 14

Program for this workshopProgram for this workshopSunday, September 12, 2010 (8:00am-12:00pm) Session III: Inversion Techniques and Seismic-Network Geometry 8:00 – 8:30 Ellsworth: Source inversion with minimal assumptions8:30 – 9:00 Meng & Ampuero: Optimal network geometries for source inversion 9:00 – 9:45 Discussion of previous talks, setting up the initial inversion test9:45 – 10:00 break Session IV: Break-Out Session and Open Discussion 10:00-10:45 Breakout groups meet

Preliminary breakout groups: Input/Output formats, station geometry, source models, forward model calculations, misfit functions web tools & development

Where are we heading? What needs to be done? Who does what?

10:45-12:00 Summaries from breakout groups, Open Discussion

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 15

Do all groups compute the Green’s function appropriately? The SIV-project thus start with a zero-order test to verify GF-computations:

“point-source” at 10 km depth, parameterized as a 1 x 1 km2 slip patch with

homogeneous slip and boxcar slip-function of duration τr = 0.2 sec

The shear-modulus at the given depth result in: Mw 4.992, M0 = 3.4992 x 1016

Step 0: Green’s Function ValidationStep 0: Green’s Function Validation

Depth[km]

VP

[km/s]VS

[km/s]Density[g/cm3]

0.0 4.8 2.6 2.3-2.0 4.8 2.6 2.3-2.0 5.5 3.1 2.5-4.8 5.5 3.1 2.5-4.8 6.2 3.6 2.7

-18.0 6.2 3.6 2.7-18.0 6.8 3.8 2.8-24.0 6.8 3.8 2.8-24.0 8.0 4.62 3.2-45.0 8.0 4.62 3.2

point-sourcedepth

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 16

Two cases are considered for the Green’s function test purely left-lateral strike-slip rupture on a vertical fault purely thrust-motion on a 40° dipping fault Stations at Y = 1 km parallel to surface projection of fault plane, and two arrays

that are 30° and 60° rotated from the fault-parallel direction

Step 0: Green’s Function ValidationStep 0: Green’s Function Validation

SCEC Annual Meeting 2010, SIV workshop Sept 11+12, 17

Details on Green’s function test in the next talk