Simulation and parameter estimation in geophysics

Lindsey Heagy, Dom Fournier, Seogi Kang, Craig Miller and the SimPEG Team

SimPEG (Simulation and Parameter Estimation in Geophysics, http://simpeg.xyz) is a collaborative effort to develop an open source framework for solving geophysical simulation and inversion problems, including electromagnetics, seismic and potential fields in a consistent manner. The goal in the development of SimPEG is to support a community of researchers with well-tested, extensible tools, and encourage transparency and reproducibility both of the SimPEG software and the geoscientific experiments.

Presented at the BC Geophysical Society meeting.

Transcript

1. simulation and parameter estimation in geophysics Lindsey, Dom, Seogi, Craig & the SimPEG Team BCGS May 18, 2017
2. eg. Airborne TDEM Oldenburg, BIRS 2016 eg. Kevitsa
3. priorities • Flexible • Multidisciplinary • Reproducible
4. priorities • Flexible • Multidisciplinary • Reproducible
5. Implemented in Python! framework (Cockett et al, 2015)
6. forward simulation Survey: Data collection and geometry Problem: Physics
7. finite volume
8. object oriented
9. discretize & solve continuous discrete Currents Data
10. framework
11. Data Misfit Regularization Inverse Problem inversion elements +
12. inversion as optimization
13. framework
14. ( does it work? )
15. • Motivated by integrated “Earth modeling” ◦ lithology + physical properties + structure • Use software that can handle ALL geophysics Discretize Case Study - Kevitsa Ni-Cu-PGE (Finland) PF.Grav(ρ) EM.Static(σ) EM.TDEM(σ) PF.Mag(К) Survey Simulation
16. Case Study - Kevitsa Ni-Cu-PGE (Finland) • Previously owned First Quantum Minerals, acquired by Boliden (2016) • Disseminated mineralization within an olivine pyroxenite intrusion (dark grey) • 240 MT @ 0.3 Ni%, 0.4 Cu% FQM, (2011). 43-101 Technical Report Koivisto et al. (2015)
17. Kevitsa Ni-Cu-PGE Bore logs (ρ, k, σ, η) EM: Conductivity Gravity: density Magnetics: Susceptibility • Multidisciplinary: ◦ 4 geophysical surveys ◦ 3 physical properties ◦ Borehole measurements
18. Case Study - Kevitsa Ni-Cu-PGE (Finland) • 2D and 3D seismic data and interpreted contacts
19. Case Study - Kevitsa Ni-Cu-PGE (Finland) • Simulating ALL geophysical responses Unit Density (ρ) Susc (k) Cond (σ) Mafic Volcanic Moderate Moderate Moderate Olivine Pyroxenite High Moderate Low Gabbro Moderate Low Low Carbonaceous Phyllite Low Low High Peridotite Moderate-low High Low Chloritic Volcanic Moderate-High Low Moderate Dunite Low High (Rem) Low Ore High? - High? Gabbro Koitelainen Moderate Low Low
20. Potential Fields Simulation • How deep can PFs see? • Is the mineralization distinguishable? SimPEG.PF Unit Density (ρ) Susc (k) Cond (σ) Mafic Volcanic Moderate Moderate Moderate Olivine Pyroxenite High Moderate Low Gabbro Moderate Low Low Carbonaceous Phyllite Low Low High Peridotite Moderate-low High Low Chloritic Volcanic Moderate-High Low Moderate Dunite Low High (Rem) Low Ore High? - High? Gabbro Koitelainen Moderate Low Low
21. Potential Fields: Density • Forward model gravity data (2 m height). Simplified model, but good enough to test the resolving power SimPEG.PF
22. Potential Fields: Density • 3D Inversion: script available: Kevitsa_Grav_Inv.ipynb SimPEG.PF Good news! We can hope to see the lower limit of the intrusion. (Iso-surface 0.2 g/cc). But no sign of the mineralization ...
23. Potential Fields: Magnetics • Simulated TMI data (VTEM draped 60 m height) SimPEG.PF
24. Potential Fields Simulation • Simulated TMI data ◦ Large negatives in the center of the intrusive ◦ Well document remanence from the Dunite unit Induced and remanent magnetism study of Kevitsa SimPEG.PF
25. Potential Fields Simulation • Simulated TMI data ◦ Quickly design a forward simulator and test remanence of the Dunite unit M Dunite unit likely larger and deeper than expected … time to invert. Montonen (2012) Msc. Thesis SimPEG.PF
26. Potential Fields Simulation • Ongoing research: Efficient 3D-MVI ◦ Development of compact and robust inversion codes that can deal with remanence ◦ Magnetization mainly focussed in the Dunite and/or contact between the intrusion and the host stratigraphy. Likely extends at depth. ◦ Max depth resolution ~1,5 km SimPEG.PF
27. Summary SimPEG.PF Unit Density (ρ) Susc (k) Cond (σ) Mafic Volcanic Moderate Moderate Moderate Olivine Pyroxenite High Moderate Low Gabbro Moderate Low Low Carbonaceous Phyllite Low Low High Peridotite Moderate-low High Low Chloritic Volcanic Moderate-High Low Moderate Dunite Low High (Rem) Low Ore High? - High? Gabbro Koitelainen Moderate Low Low • PF doesn’t see mineralization • What about conductivity?
28. Electromagnetics - conductivity • Use DC and airborne EM ◦ Conductivity contrasts ◦ DC: galvanic ◦ Airborne EM: inductive • DC ◦ Titan24 DC-IP survey (2007) ◦ Pole-dipole • Airborne EM ◦ VTEM (2009) ◦ Time domain / Co-located loop Depth at 90 m SimPEG.EM
29. VTEM data VTEM at 198 micro-s Depth at 90 m Conductive SimPEG.EM
30. VTEM data 12150N Time decays VTEM at 198 micro-s Depth at 90 m Conductive Conductive? SimPEG.EM
31. VTEM data 12150N Time decays VTEM at 198 micro-s Depth at 90 m SimPEG.EM Conductive Conductive?
32. DC resistivity data DC pseudo-section at 12150N Depth at -130 m VTEM at 198 micro-s SimPEG.EM Conductive Conductive
33. Resistivity model Question: Can we see conductive mineralized zones embedded at depth (~200 m) ? SimPEG.EM Depth at -130 m
34. Simulation: VTEM Waveform Layer (1D) Cylinder (3D) SimPEG.EM Question: What makes this difference b/w signals from 1D and 3D model?
35. Simulation: VTEM Waveform Layer (1D) Cylinder (3D) SimPEG.EM
36. Summary: VTEM 12150N Field data: time decays VTEM at 198 micro-s Synthetic data: time decays Physics: SimPEG.EM
37. DC resistivity Field DC data: Pole-dipole (Titan24) Question: - Can we see conductive mineralized zones embedded at depth (~200 m) ? SimPEG.EM
38. Simulation: DC resistivity (currents) Resistivity section SimPEG.EM Data
39. Simulation: DC resistivity (currents) Resistivity section Currents Data SimPEG.EM
40. Simulation: DC resistivity (currents) Resistivity section Currents Data SimPEG.EM
41. Simulation: DC resistivity (currents) Resistivity section Currents Data SimPEG.EM
42. Simulation: DC resistivity (pseudo-section) Resistivity section Synthetic DC pseudo-section (3D) SimPEG.EM
43. Simulation: DC resistivity (pseudo-section) Synthetic DC pseudo-section (3D) Field DC pseudo-section Resistivity section SimPEG.EM
44. Simulation: DC resistivity (pseudo-section) Synthetic DC pseudo-section (3D) Field DC pseudo-section Resistivity section SimPEG.EM Inverted
45. Kevitsa - Conclusion • What we have learned: ◦ Gravity modeling helped delineating the shape and extent of Kevitsa intrusion ◦ Magnetic modeling helped highlight the dunite and peridotite ◦ But not mineralized zones SimPEG.PF Gravity (density) Magnetic (susceptibility)
46. Kevitsa - Conclusion • What we have learned: ◦ Airborne TEM survey can delineate sediment units, and has potential to detect conductive mineralized zone at depths ◦ DC survey can delineate conductive mineralized zones at depths SimPEG.EM Airborne EM: VTEM DC:Titan 24
47. Following up: Laguna del Maule (Craig) • Gravity inversion of an active magma system • No geologic constraints, but using a mixed norm inversion to replicate a conceptual model SimPEG.PF
48. Laguna del Maule, Chile, imaging a magma reservoir Largest region of rhyolite production in Andes: • > 70 eruptions since 25 ka • > 2 m uplift since 2007 • inflating sill at 5 km depth. SimPEG.PF
49. Checkerboard test and initial L2 inversion Checkerboard test: • build model with GoCAD • simulate data • perform inversion • test how station distribution impacts ability to recover forward model Initial SimPEG inversion with L2 “smooth norm”. Not a lot of definition of edges. Next step a mixed norm inversion SimPEG.PF
50. Compact inversion using mixed norm Create inversion scheme to fit our conceptions of a magma reservoir. • finite edges and a short gradational boundary to those edges. Mixed Lp norm inversion : • 0 p 2 • amplitudes ( ) • gradients ( ) of the model • use 0 (compact) on and 1 on . • 95th percentile for and SimPEG.PF Thermodynamic models (MELTS) to interpret density model in terms of melt, crystal and volatile proportions.
51. Magma system image from gravity inversion High melt proportion, low density body (30 km3) enclosed with a larger (115 km3) high crystal content body. Cashman, Sparks and Blundy, Science 2017. SimPEG.PF
52. Magnetics Gravity DC IP FDEM TDEM NSEM Seis Flow Methods http://simpeg.xyz people http://geosci.xyz In Vancouver: Dec 2017 http://computation.geosci.xyz