UBC Faculty Research and Publications

Surface related multiple prediction from incomplete data Herrmann, Felix J. 2007-12-31

You don't seem to have a PDF reader installed, try download the pdf

Item Metadata


herrmann07a.pdf [ 198.26kB ]
JSON: 1.0107400.json
JSON-LD: 1.0107400+ld.json
RDF/XML (Pretty): 1.0107400.xml
RDF/JSON: 1.0107400+rdf.json
Turtle: 1.0107400+rdf-turtle.txt
N-Triples: 1.0107400+rdf-ntriples.txt
Original Record: 1.0107400 +original-record.json
Full Text

Full Text

ABSTRACT Incomplete data, unknown source-receiver signatures and free-surface reflectivity represent challenges for a successful prediction and subsequent removal of multiples. In this paper, a new method will be represented that tackles these challenges by combining what we know about wavefield (de-)focussing, by weighted convolutions/correlations, and recently developed curvelet-based recovery by sparsity-promoting inversion (CRSI). With this combination, we are able to leverage recent insights from wave physics towards a nonlinear formulation for the multiple-prediction problem that works for incomplete data and without detailed knowledge on the surface effects.  EAGE 69th Conference & Exhibition — London, UK, 11 - 14 June 2007  Surface-related multiple prediction and seismic interferometry are examples where weighted multi-dimensional cross-convolutions and cross-correlations of seismic data volumes provide information on Green’s functions that describe the Earth response at the surface. For instance, surface-related multiples can approximately be predicted through a weighted multidimensional convolution of the data with itself, while ’daylight imaging’ techniques extract the Green’s function by cross-correlation of wavefields (see e.g. Wapenaar et al., 2006, which contains a collection of the most recent papers on this topic). Recently, new approaches have been proposed, where the Green’s functions are extracted through inversion or deconvolution (See the contributions of Snieder et.al, Schuster et.al. and Berkhout and Verschuur in Wapenaar et al., 2006). We follow a similar approach, where we are interested in finding an alternative formulation for the following two operations: • wavefield defocusing, where the wavefield is convolved with the ’primary’ wavefield. This convolution maps the primaries into first-order multiples and firstorder multiples into second-order multiples etc., i.e., we have the mapping p(m )(x, t) → p(m+1) (x, t) = G p˜0 , A pm (x, t)  (1)  with G p˜0 , ψ · := A† ∗x,t p˜0 ∗x,t ·, the symbol ∗x,t denoting multi-dimensional cross-convolution and A† the weighting; • wavefield focusing, where the wavefield is correlated with the ’primary’ wavefield. This correlation maps the primaries into first order multiples and first-order multiples into second-order multiples etc., i.e., we have the mapping p(m+1 )(x, t) → p(m) (x, t) = F p˜0 , B pm+1 (x, t)  (2)  with P p˜0 , B · := p˜0 ⊗x,t B † ∗x,t ·, the symbol ⊗x,t denoting multi-dimensional cross-correlation and B † another weighting. In these expressions, pm refers to the (m)th -order multiple in the data (p0 represents the primary wavefield) and p˜0 represents an estimate for the ’primaries’ that is assumed to be given (we used ” to indicate that in practice we only have approximate knowledge of the primaries since the sole purpose of this work is to estimate these primaries). G and F are the defocusing and focussing operators that map the mth -order component to the (m + 1)th -order component and back. The defocusing operator consists of a multidimensional weighted cross-convolution between the ’primaries’, p˜0 , and the wavefield, followed by a deconvolution by A† that contains the surface reflectivity and the pseudo inverse (denoted by the † ) of the source and receiver directivity and time signatures. After cross-convolution, the wavefield has a tendency to spread out, i.e., we added a ’travel path’, hence the name defocusing. The wavefield focusing operator, on the other hand, constitutes a weighted cross-correlation of the wavefield with the ’primaries’, removing a ’travel path’. This weighting by B † is defined by the damped pseudo inverse of the ’autocorrelation’ of the ’primary’ operator. In this paper, we seek an alternative formulation, where (i) no information is required on A; (ii) that is stable w.r.t. incomplete data and (iii) where focusing is accomplished by sparsity-promoting inversion (replacing B † ). Fig. 1 illustrates the effect of missing data on the prediction of multiples. To accomplish these goals, we combine the focusing property and the sparsity of curvelets. After discretization (lower- and upper-case bold symbols refer to discretized vectors and matrices), define the following  EAGE 69th Conference & Exhibition — London, UK, 11 - 14 June 2007  aliased total data  SRME−predicted multiples  50  50  100  100  150  150  200  200  250  250  300  300  350  350  400  400  450  450  500  500 50  100  150  200  250  (a)  300  350  400  50  100  150  200  250  300  350  400  (b)  Figure 1: SRME multiple prediction (b) from incomplete data (a) with 80 % of the traces missing. Notice the artifacts due to the missing data leading to a deterioration of the multiple prediction.  incomplete data representation y = Ax with A := RPCT  (3)  with R the restriction operator; P := p0 ∗x,t the ’primary’ operator; CT the transpose (inverse) of the curvelet transform (see e.g. Hennenfent and Herrmann, 2006b, and the references therein), x the curvelet coefficient vector and y = Rd the incomplete data. This signal representation differs from standard-CRSI (Hennenfent and Herrmann, 2006a; Herrmann and Hennenfent, 2007) by including the ’primary’ operator P.  CRSI with focusing: By inverting the defocusing, seismic data is focused, bootstrapping the sparsity obtained by the curvelet transform. Using this property, the recovery from incomplete data can be written as follows   x = arg minx x 1 subject to Ax = y F: (4) d(m−1) = CT x   T d = PC x. In words, the solution of F involves finding the sparsest set of curvelet coefficients that matches the incomplete data when convolved with the primaries. The data, d, in this case includes primaries and multiples (see Fig. 1). As such, the estimated coefficients represent an estimate for the focused data since they are converted back into data by the ’primary’ operator during the optimization. Eq. 4 corresponds to a curvelet-sparsity regularized inverse of Berkhout’s focusing matrix and of the convolution operator in interferometric imaging (Vasconcelos and Snieder, 2006). The symbol d(m−1) refers to focused data with primaries mapped to the focal point and mth -order multiples mapped to (m − 1)th -order. The result of the sparse recovery from the incomplete data using standard-CRSI (Herrmann and Hennenfent, 2007) and CRSI + focusing are summarized in Fig 2. Expectedly, the curvelet transform compounded with the primary operator improves the recovery. Defocusing with CRSI: After successful recovery of the incomplete data, multiples can be predicted using the nonlinear mapping defined in Eq. 1. This mapping through multi-dimensional convolution, however, has the disadvantage that an estimate is needed for A. By defining A := P∗ CT with P∗ = p0 ⊗x,t the adjoint of the primary  EAGE 69th Conference & Exhibition — London, UK, 11 - 14 June 2007  CSRI recovered data  recovered from primaries  50  50  100  100  150  150  200  200  250  250  300  300  350  350  400  400  450  450  500  500 50  100  150  200  250  300  350  400  50  (a)  100  150  200  250  300  350  400  (b)  Figure 2: Comparison between CSRI (a) and CSRI + focusing (b) for data with 80 % of the traces missing. Notice the significant uplift from compounding the inverse curvelet transform with the focusing ’primary’ operator.  operator, multiples can be predicted by solving G:  x = arg minx x d(m+1) = CT x  1  subject to Ax = y  (5)  with y = d (estimated above). In words, the solution of G corresponds to finding the sparsest set of curvelet coefficients that matches the data when cross-correlated with the primaries. As such the estimated coefficients represent an estimate for the multiples, since this estimate for the multiples is converted to the primaries after applying the correlation during the optimization. More precisely, this formulation corresponds to a sparse inversion of the operator that maps multiples to primaries. The advantage of this formulation is that it does not require information on A as can be observed from Fig. 3.  Examples: Without loss of generality, we considered the acoustice reflection response of a 1-D medium consisting of three layers and a free surface. For this type of medium, the multi-dimensional cross-convolutions and correlations become simple convolutions and correlations that are diagonal in the f − k domain. In practice, the primaries are not known and the data itself is used instead, as part of an iterative procedure. In this case, spurious non-physical events may occur an observation reported in the literature (Snieder et.al. in Wapenaar et al., 2006).  Discussion The methodology presented in this paper banks on two complementary aspects of wave phenomena, namely, (i) the focusing and defocusing by multidimensional cross-convolutions/correlations, reflecting certain physical relations, and (ii) the existence of a multiscale and multi-directional curvelet transform that sparsely represents high-frequency solutions of wave equations. Pairing these two aspects leads to a new formulation for the prediction of multiples from incomplete data, without knowledge on the surface effects. The focusing is found to improve the recovery because the data becomes sparser in the curvelet domain after focusing and this explains the improvement over curvelet-only CRSI. We also observed that F corresponds to the focal transform and interferometric imaging by deconvolution formalisms, opening the interesting new perspective of adding more robustness. The prediction for the multiples after the recovery also benefited from the sparsity promotion. Again the sparsity of curvelets, that is  EAGE 69th Conference & Exhibition — London, UK, 11 - 14 June 2007  SMRE predicted multiples  CSRI−predicted multiples  50  50  100  100  150  150  200  200  250  250  300  300  350  350  400  400  450  450  500  500 50  100  150  200  250  300  350  400  50  100  150  (a)  200  250  300  350  400  (b) 4.5  0.6 true multiples SRME predicted multiples CSRI−predicted multiples  0.5  true multiples SRME predicted multiples CSRI−predicted multiples  4  0.4  3.5  0.3  3  0.2  2.5 0.1  2 0  1.5 −0.1  1  −0.2  0.5  −0.3  −0.4 0.6  0.8  1  1.2  (c)  1.4  1.6  1.8  2  0  0  50  100  150  200  250  (d)  Figure 3: Comparison between convolution-based multiple prediction (a) and sparsity-based multiple prediction (b). Aside from the acausal artifact, the sparsity-promoting multiple prediction according to Eq.5 improves the frequency content and makes it closer to the spectrum of the true multiples. Tracewise comparisons in (c-d) between the true (including internal) mutiples (blue), the multiples predicted with conventional (green) and sparsity promoted predicted multiples (red) confirm this observation. The difference in the spectrum are partially due to the fact that we only predicted the surface-related multiples.  related to the invariance of curvelets under wave propagation, leads to an improved prediction. This improvement can be understood because the method inverts the adjoint of the primary operator that contains the surface effects. The improved predictions will in turn improve curvelet-based primary-multiple separation (Herrmann et al., 2006). Interferometric prediction of ground roll will be discussed elsewhere in these proceedings. Acknowledgments: The authors would like to thank the authors of CurveLab for making their codes available. This work was in part financially supported by the Natural Sciences and Engineering Research Council of Canada Discovery Grant (22R81254) and Collaborative Research and Development Grant DNOISE (334810-05) of Felix J. Herrmann and was carried out as part of the SINBAD project with support, secured through ITF (the Industry Technology Facilitator), from the following organizations: BG Group, BP, Chevron, ExxonMobil and Shell.  REFERENCES Hennenfent, G. and F. Herrmann, 2006a, Application of stable signal recovery to seismic interpolation: Presented at the SEG International Exposition and 76th Annual Meeting. Hennenfent, G. and F. J. Herrmann, 2006b, Seismic denoising with non-uniformly sampled curvelets: IEEE Comp. in Sci. and Eng., 8, 16–25. Herrmann, F. J., U. Boeniger, and D.-J. E. Verschuur, 2006, Nonlinear primary-multiple separation with directional curvelet frames: Geoph. J. Int. To appear. Herrmann, F. J. and G. Hennenfent, 2007, Non-parametric seismic data recovery with curvelet frames. Submitted for publication. Vasconcelos, I. and R. Snieder, 2006, Interferometric imaging by deconvolution: theory and numerical examples: Presented at the SEG International Exposition and 76th Annual Meeting. Wapenaar, C., D. Draganov, and J. Robertsson, eds. 2006, Supplement Seismic Interferometry. SEG.  EAGE 69th Conference & Exhibition — London, UK, 11 - 14 June 2007  


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics

Country Views Downloads
China 25 28
United States 10 0
Russia 9 0
Brazil 7 7
Germany 4 0
France 1 0
Egypt 1 0
Canada 1 0
Japan 1 0
City Views Downloads
Shenzhen 17 28
Unknown 13 7
Saint Petersburg 9 0
Beijing 8 0
Ashburn 5 0
Buffalo 2 0
Los Angeles 1 0
Sunnyvale 1 0
Ottawa 1 0
Mountain View 1 0
Tokyo 1 0

{[{ mDataHeader[type] }]} {[{ month[type] }]} {[{ tData[type] }]}
Download Stats



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items