A Blind Test of Traveltime and Waveform Inversion Colin A. Zelt 1, R. Gerhard Pratt 2, Andrew Brenders 2, Sara Hanson-Hedgecock 1 and John A. Hole 3 1 Rice University, Houston, TX, USA 2 Queen's University, Kingston, ON, Canada 3 Virginia Tech, Blacksburg, VT, USA
use traveltime inversion to estimate large- scale model (and starting model for waveform inversion) – resolution fresnel zone use waveform inversion to estimate high- resolution model – resolution wavelength Methodology
Comparison of traveltime and waveform inversion for crosswell synthetic data
Talk Outline 1.True model and data 2.Traveltime inversion 3.Waveform inversion
True Model
The Data 51 shots 1390 receivers 2-11 Hz 2-D viscoelastic FD code (Robertsson et al. 1994) No noise
Example shot gather
1-D starting model 1 st arrival traveltime tomography Simultaneous refl+refr traveltime inversion for lower half of model 1-D starting model Waveform tomography
Refraction and reflection traveltime tomography combination of smooth tomography and layered model to invert first arrivals and reflections
1st arrival Model 1st arrival + refl/refr layered Model
True Model Traveltime Model
Waveform tomography 2-D acoustic modeling Forward and inverse steps in frequency domain (computationally efficient) iterative least-squares minimization of data residuals (frequency components) proceed from low to high frequency gradient direction from multiplication of forward and back-propagated waveforms in frequency domain
Waveform tomography (continued) like pre-stack, reverse time migration, but formulated in terms of velocity, not reflectivity, and iterative (Born approximation) invert for source signature require “good” starting model (from traveltimes or other method of velocity analysis) reference: Pratt 1999, Geophysics, 64,
Waveform tomography (application) Invert frequencies from 0.8 to 7 Hz (wavelengths of 5-10 km to km) Window data, initially 3 s, later 6 s PC platform, Intel Pentium 4, 2.8 GHz processor, 4 GB RAM, RedHat Linux days total CPU time
True Model Starting Model
Starting Model: waveform tomography
Final Model (2 Hz): waveform tomography
Final Model (4 Hz): waveform tomography
Final Model (7 Hz): waveform tomography
True Model
Waveform Model
Traveltime Model Waveform Model
Data comparison Input data Predicted data
Conclusions Model/data available: terra.rice.edu/department/faculty/zelt/ccss Waveform tomography shows great potential Traveltime model and/or low frequency data important Future work: elastic and 3-D See EOS feature article May 3, 2005 issue Waveform references: Brenders & Pratt 2007, GJI, 168,