1. Imaging Every Bounce in a Multiple G. Schuster, J. Yu, R. He U of Utah

2. OUTLINE Why Migrate Multiples? Migrating every Bounce Numerical Results Summary.

3. Why Migrate Multiples? Wider Coverage Better Fold Better Vert. Res.

4. Motivation 1: Extend Coverage 3D Courtesy of B. Paulsson PGSI

5. Shot radius 160 level Receiver array Depth in Well: 4,000-12,000 ft 22,000 ft 20,000 ft 22,000 ft The 3D Image Volume from a Massive 3D VSP ® Survey Courtesy of B. Paulsson PGSI

6. 3D View of the image volume around the 3D VSP well

10. 0 8 km 0 (Zhang & McMechan 1997) 24 km Motivation 2: Peek Around Corners with Multiples

11. OUTLINE Why Migrate Multiples? Migrating every Bounce Numerical Results Summary.

12. Migrating Every Bounce 1. Predict Multiple Traveltimes from Data Primary Pick t(s,g’) Pick t(s,g’) sg’

13. Migrating Every Bounce 1. Predict Multiple Traveltimes from Data sg’g sg’gPrimarysg’g t(s,g’) + t(g’,g) t(s,g’,g) = min( ) Asakawa & Matsuoka, 2002

14. 0 m 180 m 0.1 s 0.0 s Time (s) X (m) 3rd-order 2nd-order 1st-order primary Free-Surface Multiples & 2-Layer Model

15. Migrating Every Bounce 1. Predict Multiple Traveltimes from Data s g 2. Migrate Multiples Sum Data along Predicted T(s,x,g) d(g, t(s,g’) + t(g’,x,g’’) + t(g’’,g) ) g m(x) = Predicted from Data xg’g’’

16. Migrating Every Bounce 1. Predict Multiple Traveltimes from Data s g 2. Migrate Multiples Sum Data along Predicted T(s,x,g) d(g, t(s,x,g’) + t(g’,g’’) + t(g’’,g) ) g m(x) = x

17. Modeling Peglegs (Jakubowicz; Reshef, Keydar, Landa; Weglein, Gasparotto, et al). sg X y t(s1y) + t(x2g) - t(xoy) = t(s1o2g) 12 o ?PrimariesPegleg A B Choose x & y so incidence angles agree

18. Summary s g Migrate Multiples d(g, t(s,x,g’) + t(g’,g’’) + t(g’’,g) ) g m(x) = x model model

19. OUTLINE Why Migrate Multiples? Migrating every Bounce Numerical Results Summary.

20. d 0 km X 5 km 0 km Z 2.5 km 0 km Z 2.5 km m0 + m1 + m2 m0 m1m2 2-Layer Model: Migration 1 CSG

21. 0 km X 5 km m0 + m1 m0 X z X z m0 vs m1: 2-Layer Migration Images Even Illumination

22. OUTLINE Why Migrate Multiples? Migrating every Bounce Numerical Results Summary.

23. One part of SMAART model Depth (ft) 0 30K Offset (ft)050.55K reflector 0 reflector 1 reflector 2 reflector 3

24. Time (s) (s) 0 9Geophone(#)1 540 540 Free-Surface Multiple

25. Time (s) (s) 0 9Geophone(#)1 540 540 Another Multiple

26. KM before de-multiple Depth (ft) 0 30K Offset (ft)1 50.55K

27. KM after de-multiple Depth (ft) 0 30K Offset (ft)1 50.55K

28. Multiple migration Depth (ft) 0 30K Offset (ft)050.55K Using multiple 02020

29. Multiple migration result Depth (ft) 0 30K Offset (ft)1 50.55K

30. Multiple migration result Offset (ft) Depth (ft) Depth (ft) 3750 26,250 6,750 9,375

31. OUTLINE Why Migrate Multiples? Migrating every Bounce Numerical Results Summary.

32. Raw Data of CRG#15Ghosts of CRG# 15 0 0.37 Time (s) 1000 ft 00

33. Raw Data of CRG#15Primary of CRG# 15 0 0.37 Time (s) 1000 ft 00

34. Primary Image1st Ghost Image 0 1 Depth (kft) 445 ft 00

35. Primary Image1st Order Ghost Image 0 1 Depth (kft) 445 ft 00

36. Well & Primary ImageWell & 1st Ghost Image 0 1 Depth (kft) Offset=105ft

37. Well & Primary ImageWell & 1st Ghost Image 0 1 Depth (kft) Offset=200 ft

38. Summary Use data to design migration kernel Benefits: Better resol. & fold kernel Use Delft method predict multiples Test on field data

39. Primary Migration x Sum Data along Predicted T(g) Predicted by Ray Tracing

40. Multiple Migration x Sum Data along Predicted T(g) Predicted from Data Predicted from Ray Tracing

41. Prediction+Subtraction u Predict or pick the traveltime of a multiple. u u NMO the multiple within a time window. u u If a significant overlaying primary is suppressed at the same time, use the same strategy to predict it and fill the gap. u u Predict the multiple by a multichannel two-way prediction filter. u u Subtract the predicted multiple.

42. OUTLINE Why Migrate Multiples? Prediction of Multiple T(g) Joint Migration Pattern Recog. Joint Migration LSM.

43. What Good are Natural T(s,g’,g)? Primary g’g s’ t(s,g’,g) = min(t(s,g’) + t(g’,g)) t(s,g’,g) = min(t(s,g’) + t(g’,g)) g’g’g’g’

44. Answer 1: Natural Decon of Multiples Primary g’g s’ Actual Multiple Predicted Multiple Adaptive Subtraction Deblurring: d = G d 10

45. Answer 2: Semi-Natural Migration of Multiples g’ s Actual Multiple Predicted Multiple x d(g, t(s,g’) + t(g’,x) + t(x,g) ) g m(x) = g

46. OUTLINE Why Migrate Multiples? Prediction of Multiple T(g) Joint Migration Pattern Recog. Joint Migration LSM.

47. 0 km 5 km 0 km 7 km Salt Model

48. Joint Migration : LS Multiple Migration PROBLEM : Multiples get Coherently Migrated Primary Multiple SOLUTION: Least Squares Joint Migration MultiplePrimary L0 m0 + L1 m1 = d (L0 + L1) m = d

49. 0 km 5 km 0 km 7 km 0 km 7 km L0 0 0L0L1L2 ++ Standard Migration L1 L2 Correlation Wt L0 L0L1L L1 L2

50. s 0 km 5 km 0 km 7 km 0 km 7 km Migration with Correlation Weights Correlation Weights

51. 0 km 5 km 0 km 7 km 0 km 7 km Ground Truth Migration

52. OUTLINE Why Migrate Multiples? Prediction of Multiple T(g) Joint Migration Pattern Recog. Joint Migration LSM.

53. Multiple Migration x Sum Data along Predicted T(g) Predicted from Data Predicted from Ray Tracing

54. Middle Bounce Migration x Predicted from Ray Tracing d(g, t(s,g’) + t(g’,x,g’’) + t(g’’,g) ) g m(x) = Predicted from Data

55. Rigorous Theory? D(g) = R f + R f + ….. 1211212dataprimary1st-order 1 2

56. Frechet Derivative D(g) = R f + R f + ….. 12112121 2 r rr r R 121 f R 121 rR121 fR121 + product rule

57. d 0 km X 5 km 0 km Z 2.5 km 0 km Z 2.5 km m0 + m1 + m2 m0 m1m2

58. Multiple 01020 Time (s) 0 9 Geophone(#)1 540

59. Multiple 0202010 Time (s) 0 9 Geophone(#)1 540

60. 0 km 2 km 0 km 2 km 2 km 0 km 2 km 2 km 0 km 2 km 2 km Graben Model Standard Migration Least Squares Migration Least Squares MigrationGhosts