1 / 33

Generating Green’s Functions with Pylith

Generating Green’s Functions with Pylith. Mw 6 Queshm Island, Iran, EQ. Outline. Overview/Motivation When/where use FE models? Workflow From data to mesh to Green’s functions to model Examples Bam & Qeshm Island earthquakes, Iran. Astronaut photography http://eol.jsc.nasa.gov/.

Download Presentation

Generating Green’s Functions with Pylith

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Generating Green’s Functions with Pylith Mw 6 Queshm Island, Iran, EQ

  2. Outline • Overview/Motivation • When/where use FE models? • Workflow • From data to mesh to Green’s functions to model • Examples • Bam & Qeshm Island earthquakes, Iran Astronaut photography http://eol.jsc.nasa.gov/

  3. Effects of Crustal Structure • Vertical layering (2D) • Tends to move apparent source up or down (~10% effect) • Horizontal contrasts (3D) • Map into slip features, inferred geometry • Fialko (2006) finds 2-2.5x contrasts in So. Cal • Goal: For generic settings, what is inversion sensitivity? • Generate synthetic data using cross-fault contrast • Find best-fit solution in elastic half space • Assess bias: Inferred fault dip

  4. Choosing Model Complexity • Case 1 • Lots of info • Seismicity • Velocity/rigidity structure • Mapped faults • Atmospheric water vapor content • Computationally expensive…. Community fault model So. Cal

  5. Choosing Model Complexity • Case 2: • Sparse information • Mainly teleseismic EQ locations • No continuous GPS • Sporadic remote sensing Bam EQ, courtesy E. Fielding

  6. Case 1: How do we use all this information? When do we have to include all info? When does it make sense to simplify? Case 2 What bias do we introduce by using inadequate models? How should we present this error? Which problems can we still address? Choosing Model Complexity

  7. FE Workflow • Step 1: Matlab • Define data/fault geometry • Define crustal structure • Subdivide fault • One patch at a time, build Cubit, Pylith input files Vertical, strike-slip fault, divided into patches

  8. Okada-based Green’s functions Slip on shallow fault patch

  9. Okada-based Green’s functions Slip on deep fault patch

  10. Inversion

  11. FE Workflow • Step 2: Cubit • Build mesh • Step 3: Pylith • Generate Green’s functions • Step 4: Matlab • Assemble all patches, perform inversion

  12. FE Workflow • Step 2: Cubit • Build mesh • Step 3: Pylith • Generate Green’s functions • Step 4: Matlab • Assemble all patches, perform inversion

  13. FE Workflow • Step 2: Cubit • Build mesh • Step 3: Pylith • Generate Green’s functions • Step 4: Matlab • Assemble all patches, perform inversion Mesh quality/density?

  14. Comparison w/Okada • Discrete fault patches • Readily available (Okada, Poly3D) • Easy to visualize • Historical:compare with previous work • Node, points • More natural comparison once Pylith Green’s functions mode

  15. Comparison w/Okada 3D version • Discrete fault patches • Readily available (Okada, Poly3D) • Easy to visualize • Historical:compare with previous work • Node, points • More natural comparison once Pylith Green’s functions mode

  16. Comparison w/Okada • More misfit from shallower ramp • Can fit almost exactly with different fault patch

  17. Comparison w/Okada • More misfit from shallower ramp • Can fit almost exactly with different fault patch

  18. Comparison w/Okada • More misfit from shallower ramp • Can fit almost exactly with different fault patch

  19. Comparison w/Okada • More triangular -> wider patch w/ less slip • Moments/centroid almost identical

  20. Examples: Sensitivity Tests • Generate synthetic data using cross-fault contrast (slow) • Invert using half space (fast) • Assess potential bias

  21. Examples: Sensitivity Tests • Generate synthetic data using cross-fault contrast (slow) • Invert using half space (fast) • Assess potential bias

  22. Examples: Sensitivity Tests • Generate synthetic data using cross-fault contrast (slow) • Invert using half space (fast) • Assess potential bias

  23. Vertical Fault Deformation Patterns Can’t fit asymmetric pattern with vertical fault Apparent dip: 75º

  24. Cross-Fault Contrast Results • Retrieve input geometry when contrast=0 • Up to 20 degree error for reasonable values • Sensitivity depends on noise RMS, viewing geometry

  25. Cross-Fault Contrast Results • Retrieve input geometry when contrast=0 • Up to 20 degree error for reasonable values • Sensitivity depends on noise RMS, viewing geometry

  26. 11/27/05, Mw 6 Qeshm Island EQ Zagros Mtns Qeshm Persian Gulf Astronaut photography http://eol.jsc.nasa.gov/

  27. 11/27/05, Mw 6 Queshm Island EQ Color scale = 2.8 cm

  28. 11/27/05, Mw 6 Queshm Island EQ Color scale = 40 cm

  29. Assessing Potential Bias in Inferred Dip Error from noise Error from structure Atmospheric noise > Structure error

  30. 2003 Bam, Iran, Earthquake • > 40 cm line-of-sight deformation • Not much structural/fault location info • How well do inversions for fault dip perform? Data courtesy Eric Fielding

  31. 2003 Bam, Iran, Earthquake • Pylith: • Generate Green’s functions for distributed slip inversion • Repeat for various dip angles, cross-fault contrasts

  32. 2003 Bam, Iran, Earthquake • Increased contrast = increased dip • Best fit still no-contrast solution, near-vertical dip • Geometrical irregularities = large residual • Need more complicated geometry before can assess crustal contribution

  33. Conclusions • Sensitivity tests can largely be done with analytic inversions • More time consuming FE modeling (especially inversions) can be avoided for many problems • Large atmospheric noise • Known fault plane geometry • Patch by patch Green’s function generation very time consuming • Can be a bit more efficient, use redundant dip/strike info • Internal Pylith Green’s function producer very desirable

More Related