WS 09 Velocity Model Building

Transcription

1

2 Full wave vs Kinematic approaches Goal in mid 80 s: Full Waveform Inversion (FWI) ~ Tomography plus Migration. There seems to exist considerable confusion regarding the sort of information about the subsurface which one might expect to extract using the L2 approach to the inverse problem of reflection seismology, and also regarding the quality of the L2 results relative to the output of conventional processing methods. The relation between L2 inversion and migration of both stacked and unstacked data is now well understood, at least in principle [see Lailly (1984) or Tarantola (1984) and Beylkin (1985)]. On the other hand, the possibility of reliable subsurface parameter estimation from realistically band-limited data seems to arise various opinions. One reads several compelling arguments in the literature that extraction of velocity trends (the main out-of-passband components of interest in reflection seismology from band-limited (low-cut) data by L2 inversion is impossible, or so difficult as to be infeasible (Tarantola, 1986). One also encounters convincing arguments in various successful stories of FWI by different groups around the turn of the new century on synthetics and on real data when both reflections and diving waves (refractions) are available. F. Santosa, W.W. Symes, 1989, An analysis of least-squares velocity inversion, 1989, SEG geophyiscal Monograph, number 4.

3 Full wave vs Kinematic approaches However, capturing long-to-intermediate wavelength information from reflection data with a least-square waveform misfit is challenging (may be impossible with a local optimization techniques) Limitations of the least-square waveform inversion formulation have been identified - Cycle skipping issue requiring a decent initial model (kinematically compatible) - Capturing long-wavelength content from reflected waves is ambiguous and difficult - Lack of intermediate wavelength information when only reflected waves are recorded Reformulations of FWI for overcoming these two pitiful features - relaxing cycle skipping - capture reflection travel-time information in a waveform inversion manner (leading to different automatic migration velocity analysis). While promising, these approaches are challenging in practice. We have chosen here to focus on the second aspect, which only start now to appear as an industrial solution.

4 Program Part 1: Overview 9h05-9h40: G. Chavent: Data Space Reflectivity and the Migration based Travel Time approach to FWI 9h40-10h15: B. Symes: Extended waveform inversion 10h15-10h30 : discussion Part 2: Methods/Examples 10h30-10h50 T. Alkhalifah: Optimizing the coefficients of the leading terms of the Born Series: FWI+MVA+more 10h50-11h10 V. Cheverda: FWI for elastic media: macrovelocity reconstruction 11h10-11h30 A. Gomes: Extending the reach of FWI with reflection data: Potential and challenges 11h30-11h50 B. Biondi: Application of tomographic FWI (TFWI) to large-scale field datasets: challenges and insights. 11h50-12h10: discussion 12h10-13h15: Lunch

5 Program Part 3: Applications 13h20-13h40 U. Albertin: Advances in and Challenges for Broadband Multiparameter Full Waveform Inversion 13h40-14h00: M.Warner: Reflection FWI without a good starting model 14h00:14h20: D. Vigh: Updating velocity fields beyond the diving waves 14h20-14h40: P. Nandi: Wave-Equation Migration Velocity Analysis in the Surface Common Offset Domain: Application to Viking Graben Field Data 14h40-15h00: Discussion Part 4: Algorithms/Methods 15h00-15h20: R. Soubaras: Mitigating the gradient artefacts of Migration Velocity Analysis by Gauss-Newton update 15h20-15h40 R.Valensi: Reflection Waveform Inversion method: solutions to the reflectivity-background coupling problem and consequences on the convergence 15h40-16h00 R. Brossier: From RWI to JFWI: including diving waves in reflectionbased velocity model building 16h00:16h20: H. Chauris: Image-domain versus data-domain velocity analysis based on true-amplitude subsurface extended migration 16h20-17h00: discussion

6 Questions we may discuss Misfit function (reformulation): How should we extend the model? In model space or in data space, with or without the Born approximation? Does it really matter? Is the migration step necessary for any reflection based full wave approach? What is feasible in 3D? How to capture both reflection and refraction information? What are the crosstalks between modeled and observed events? How should we compare the data (time pixel, trace, gather, )? What can we borrow from the progresses in conventional FWI?

7 Questions we may discuss What are the main potential benefits? What is the role of phases and amplitudes? Joint or sequential inversion of phases and amplitude? How does it compare with a picking-based approach? What will we need to have solutions as efficient as ray based tomography? Should we combine picking-based approach and waveform approach? Is coherency enhancement and even phase identification crucial (at which stage of the inversion)?

8 Questions we may discuss Which physics should we consider? Is an acoustic approach sufficient? How do we handle AVO effects? What kind of model do we retrieve (background, impedances, ) Should we consider multi-parameter inversion? What data pre-processing? Do we need a data hierarchization (primary only, separation betwen reflection and transmission/refraction, ) What is the role of the low frequencies? Other items?