Presentation is loading. Please wait.

Presentation is loading. Please wait.

Evaluating the utility of gravity gradient tensor components Mark Pilkington Geological Survey of Canada.

Similar presentations


Presentation on theme: "Evaluating the utility of gravity gradient tensor components Mark Pilkington Geological Survey of Canada."— Presentation transcript:

1 Evaluating the utility of gravity gradient tensor components Mark Pilkington Geological Survey of Canada

2 Tensor component choice Txx TxyTxz Tyz Tyy Tzz  Single components  Combinations  Concatenations Which to use?  Qualitative interpretation  Quantitative interpretation

3 Tensor component choice Quantitative interpretation [Inversions] (T xx, T xy, T xz, T yy, T yz )Li, 2001 (T uv, T xy ), T zz Zhdanov et al., 2004 (T xz, T yz, T zz, T uv )Droujinine et al., 2007 (T uv, T xy )Li, 2010 (T uv, T xy ), T zz, (T zz, T uv, T xy )Martinez & Li, 2011 T zz, (T xz, T yz, T zz), (T xz, T yz, T xz, T yy, T xx )Martinez et al., 2013 Rating the solutions:  goodness of fit  sharp/smooth  close to geology

4 Inversion versus component combinations Martinez et al., 2013 T zz T xz, T yz, T zz T xz, T yz, T xz, T yy, T xx T xz, T yz, T xz, T zz, T yy, T xx Components inverted: RMS error Txx Txy Txz Tyy Tyz Tzz 1-C 23.9 23.2 31.8 23.1 26.1 16.5 3-C 17.5 16.0 15.9 16.0 12.4 22.5 5-C 16.6 12.6 16.3 15.8 12.2 24.3 6-C 15.7 13.0 17.9 13.8 13.8 21.4

5 Outline Aim: quantitative rating of component/combinations Approach: inversion using a simple model – estimate parameter errors Method: linear inverse theory – analyse model/data relations

6 Inversion method used Inversion Parametric [underdetermined inversion problem] n data m parametersm >> nm << n Model3-D volumeSpecified shape quantity SolutionPhysical propertyParameters (density …) (depth, dip…) MethodologyRegularized inversionOverdetermined least – squares SolutionResolution, covarianceParameter errors appraisal

7 Prism model z t b w xc yc 

8 Inverse theory Forward problem: b = f (x) b = data x = parameters (linearized)db = Adx A = Jacobian [ model dependent ] a ij = db i /dx j Inverse problem : dx = A + db A = U  V T singular value decomposition

9 Inverse theory A = U  V T singular value decomposition U = data eigenvectors V = parameter eigenvectors  = singular values R = VV T Resolution matrix (=I) S = UU T Data information matrix C = C d V  -2 V T Covariance matrix

10 Model parameter errors C = C d V  -2 V T Parameter covariance matrix C d = Data covariance  =  singular values small  large C large  small C C d = e 2 I Equal data error C d = DVariable data error

11 Variable component errors Components have different error levels: e.g., e(T xx ) = e(T xz )  only relative levels required  estimate based on FFT or equivalent source method  ratio Tzz : Txz, Tyz : Txy : Txx, Tyy = 1 : 0.70 : 0.37 : 0.59 Component quantities are combined: e.g., H1 = sqrt (T xz 2 +T yz 2 )  combine errors: e(Tuv) = [0.5 (e(T xx ) 2 +e(T yy ) 2 )] 1/2

12 Component quantities tested Single components: T xx T yy T zz T xy T yz T xz T uv Invariants: I1 = T xx T yy +T yy T zz +T xx T zz -T xy 2 -T yz 2 -T xz 2 I2 = T xx (T yy T zz -T yz 2 )+T xy (T yz T xz -T xy T zz )+T xz (T xy T yz -T xz T yy ) H1 = sqrt (T xz 2 +T yz 2 ) H2 = sqrt [T xy 2 +0.25(T yy -T xx ) 2 ] Concatenations: (T uv, T xy ) (T xz, T yz, T zz) (T xy, T yz, T xz) (T xx, T yy, T xy) (T xz, T yz, T xz, T xy, T xx ) (T yy, T yz, T xz, T xy, T xx )

13 Inversion tests Procedure: Specify model and evaluate matrix A [db=Adx] Calculate covariance matrix C Get parameter standard deviations (p.s.d.) Rank p.s.d. for each parameter versus component quantity Models tested: xc yc z t w b 

14 Eigenvector matrix V

15 Invariants: I1 = TxxTyy+TyyTzz+TxxTzz-Txy 2 -Tyz 2 -Txz 2 I2 = Txx(TyyTzz-Tyz 2 )+Txy(TyzTxz-TxyTzz) +Txz(TxyTyz-TxzTyy) H1 = sqrt(Txz 2 +Tyz 2 ) H2 = sqrt[Txy 2 +0.25(Tyy-Txx) 2 ]

16 Eigenvector matrix V

17 Correlation matrix corr ij = cov ij [ cov ii cov jj ] 1/2

18 Parameter errors xc,yc = location z = depth t = thickness w = width b = breadth  = density

19 Parameter errors xc,yc = location z = depth t = thickness w = width b = breadth  = density

20 Parameter errors xc,yc = location z = depth t = thickness w = width b = breadth  = density

21 Parameter error ranking [29 models] error high low

22 Parameter errors versus averaging No averaging correction With averaging correction

23 Conclusions  Concatenated components produce smallest parameter errors  Invariants I1, I2 best performers in combined component category  Purely horizontal components poor performers  Tzz best single component

24

25 Parameter rankings I1T xz higher error

26 Width error versus coordinate rotation  coordinate axis body axis

27 Information density matrix

28 Information density versus eigenvector


Download ppt "Evaluating the utility of gravity gradient tensor components Mark Pilkington Geological Survey of Canada."

Similar presentations


Ads by Google