12
Layer-thickness determination and stratigraphic interpretation using spectral inversion: Theory and application Charles I. Puryear 1 and John P. Castagna 1 ABSTRACT Spectral inversion is a seismic method that uses a priori in- formation and spectral decomposition to improve images of thin layers whose thicknesses are below the tuning thickness. We formulate a method to invert frequency spectra for layer thickness and apply it to synthetic and real data using com- plex spectral analysis. Absolute layer thicknesses significant- ly below the seismic tuning thickness can be determined ro- bustly in this manner without amplitude calibration. We ex- tend our method to encompass a generalized reflectivity se- ries represented by a summation of impulse pairs. Applica- tion of our spectral inversion to seismic data sets from the Gulf of Mexico results in reliable well ties to seismic data, ac- curate prediction of layer thickness to less than half the tun- ing thickness, and improved imaging of subtle stratigraphic features. Comparisons between well ties for spectrally in- verted data and ties for conventional seismic data illustrate the superior resolution of the former. Several stratigraphic examples illustrate the various destructive effects of the wavelet, including creating illusory geologic information, such as false stratigraphic truncations that are related to later- al changes in rock properties, and masking geologic informa- tion, such as updip limits of thin layers. We conclude that data that are inverted spectrally on a trace-by-trace basis show greater bedding continuity than do the original seismic data, suggesting that wavelet side-lobe interference produces false bedding discontinuities. INTRODUCTION According to the Widess 1973 model, seismically thin layers be- low one-eighth of a wavelength in thickness cannot be resolved. However, such thin layers might be significant reservoirs or impor- tant flow units within reservoirs. Exploration and development geo- physicists frequently are faced with the task of inferring layer thick- ness for layers such as these where the top and base of the layer can- not be mapped distinctly. Consequently, determining layer proper- ties for such seismically thin beds is of great interest in exploration and development applications. Although tuning-thickness analysis based on the theory of Widess 1973 and Kallweit and Wood 1982 has been the thickness-map- ping method of choice for several decades, Partyka et al. 1999, Par- tyka 2005, and Marfurt and Kirlin 2001 demonstrate the effec- tiveness of spectral decomposition using the discrete Fourier trans- form DFT as a thickness-estimation tool. However, such methods have difficulty with thin layers if seismic bandwidth is insufficient to identify the periodicity of spectral peaks and notches unambiguous- ly. This difficulty motivates the development of methods that do not require precise identification of peaks and troughs within the seismic bandwidth. Partyka 2005, Portniaguine and Castagna 2004, 2005, Puryear 2006, Chopra et al. 2006a, 2006b, and Puryear and Castagna 2006 show that inversion of spectral decompositions for layer properties can be improved when reflection coefficients are deter- mined simultaneously. The result is a sparse-reflectivity inversion that can be parameterized to provide robust layer-thickness esti- mates. Such a process, called spectral inversion, produces results that differ from conventional seismic inversion methods. In this article, we discuss the basic theory of spectral inversion, develop a new spectral-inversion algorithm, and show field exam- ples of improved bed-thickness determination and enhanced strati- graphic imaging that can be achieved with the process. Widess model The Widess 1973 model for thin-bed reflectivity teaches that the fundamental limit of seismic resolution is /8, where is the wave- length. Essentially, constructive wavelet interference and measured amplitude in the time domain peak at /4. The waveform shape and peak frequency continue to change somewhat as amplitude decreas- es to /8, at which point the waveform approximates the derivative of the seismic wavelet. As the layer thins below /8, the waveform Manuscript received by the Editor 18 April 2007; revised manuscript received 31 October 2007; published online 27 February 2008. 1 University of Houston, Department of Geosciences, Houston, Texas, U.S.A. E-mail: [email protected]; [email protected]. © 2008 Society of Exploration Geophysicists. All rights reserved. GEOPHYSICS, VOL. 73, NO. 2 MARCH-APRIL 2008; P. R37–R48, 22 FIGS. 10.1190/1.2838274 R37

Spectral Inversion

  • Upload
    akhedy

  • View
    24

  • Download
    1

Embed Size (px)

DESCRIPTION

inversi

Citation preview

  • L tratu ap

    C

    such as false stratigraphic truncations that are related to later-

    lHt

    pn

    n

    ta

    pttfhilr

    b

    pm

    tmates. Such a process, called spectral inversion, produces results

    GEOPHYSICS, VOL. 73, NO. 2 MARCH-APRIL 2008; P. R37R48, 22 FIGS.10.1190/1.2838274al changes in rock properties, and masking geologic informa-tion, such as updip limits of thin layers. We conclude that datathat are inverted spectrally on a trace-by-trace basis showgreater bedding continuity than do the original seismic data,suggesting that wavelet side-lobe interference produces falsebedding discontinuities.

    INTRODUCTION

    that differ from conventional seismic inversion methods.In this article, we discuss the basic theory of spectral inversion,

    develop a new spectral-inversion algorithm, and show field exam-ples of improved bed-thickness determination and enhanced strati-graphic imaging that can be achieved with the process.

    Widess modelThe Widess 1973 model for thin-bed reflectivity teaches that the

    fundamental limit of seismic resolution is /8, where is the wave-ayer-thickness determination and ssing spectral inversion: Theory and

    harles I. Puryear1 and John P. Castagna1

    ABSTRACT

    Spectral inversion is a seismic method that uses a priori in-formation and spectral decomposition to improve images ofthin layers whose thicknesses are below the tuning thickness.We formulate a method to invert frequency spectra for layerthickness and apply it to synthetic and real data using com-plex spectral analysis.Absolute layer thicknesses significant-ly below the seismic tuning thickness can be determined ro-bustly in this manner without amplitude calibration. We ex-tend our method to encompass a generalized reflectivity se-ries represented by a summation of impulse pairs. Applica-tion of our spectral inversion to seismic data sets from theGulf of Mexico results in reliable well ties to seismic data, ac-curate prediction of layer thickness to less than half the tun-ing thickness, and improved imaging of subtle stratigraphicfeatures. Comparisons between well ties for spectrally in-verted data and ties for conventional seismic data illustratethe superior resolution of the former. Several stratigraphicexamples illustrate the various destructive effects of thewavelet, including creating illusory geologic information,According to the Widess 1973 model, seismically thin layers be-ow one-eighth of a wavelength in thickness cannot be resolved.owever, such thin layers might be significant reservoirs or impor-

    ant flow units within reservoirs. Exploration and development geo-

    la

    pe

    o

    Manuscript received by the Editor 18April 2007; revised manuscript received 31 O1University of Houston, Department of Geosciences, Houston, Texas, U.S.A. E-m2008 Society of Exploration Geophysicists.All rights reserved.

    R37igraphic interpretationplication

    hysicists frequently are faced with the task of inferring layer thick-ess for layers such as these where the top and base of the layer can-ot be mapped distinctly. Consequently, determining layer proper-ies for such seismically thin beds is of great interest in explorationnd development applications.

    Although tuning-thickness analysis based on the theory of Widess1973 and Kallweit and Wood 1982 has been the thickness-map-ing method of choice for several decades, Partyka et al. 1999, Par-yka 2005, and Marfurt and Kirlin 2001 demonstrate the effec-iveness of spectral decomposition using the discrete Fourier trans-orm DFT as a thickness-estimation tool. However, such methodsave difficulty with thin layers if seismic bandwidth is insufficient todentify the periodicity of spectral peaks and notches unambiguous-y. This difficulty motivates the development of methods that do notequire precise identification of peaks and troughs within the seismicandwidth.

    Partyka 2005, Portniaguine and Castagna 2004, 2005, Puryear2006, Chopra et al. 2006a, 2006b, and Puryear and Castagna2006 show that inversion of spectral decompositions for layerroperties can be improved when reflection coefficients are deter-ined simultaneously. The result is a sparse-reflectivity inversion

    hat can be parameterized to provide robust layer-thickness esti-ength. Essentially, constructive wavelet interference and measuredmplitude in the time domain peak at /4. The waveform shape andeak frequency continue to change somewhat as amplitude decreas-s to /8, at which point the waveform approximates the derivativef the seismic wavelet. As the layer thins below /8, the waveform

    ctober 2007; published online 27 February 2008.ail: [email protected]; [email protected].

  • ddfm

    tc

    hWs

    pm

    ho

    G

    tc

    ir

    u

    c

    tc

    b

    tgc

    ts

    tttn

    b

    gte

    iqa

    qn

    p

    s

    lr

    n

    a

    u

    a

    flt

    idr

    e

    Fapt

    Frhn

    FtIna

    R38 Puryear and Castagnaoes not change significantly, but amplitude steadily decreases, asemonstrated in Figure 1. In this figure, amplitudes are obtainedrom the convolution of a 30-Hz Ricker wavelet with a wedgeodel.From this point of view, there are no means to differentiate be-

    ween amplitude changes associated with reflection-coefficienthanges and thickness changes below /8, making this thickness aard resolution limit for broadband analysis in the time domain.orse yet, in the presence of noise and wavelet broadening, the tran-

    ition between /4 and /8 is obscured, sometimes making /4 aractical limit of resolution. The key assumptions for the Widessodel are that the half-spaces above and below the layer of interest

    ave the same acoustic impedance and that the acoustic impedancef the thin layer is constant.

    eneralized reflectivity modelAlthough the theory of Widess 1973 is valid when the assump-

    ions are satisfied, nature rarely accommodates such strict theoreti-al provisions. The theory of spectral inversion is based on the real-zation that the Widess model for thin-bed reflectivity presupposes aeflectivity configuration that is actually a singularity in the continu-m of possible reflection-coefficient ratios. Any reflection-coeffi-ient pair can be decomposed into even and odd components, withhe even components having equal magnitude and sign and the oddomponents having equal magnitude and opposite sign, as describedy Castagna 2004 and Chopra et al. 2006a, b.

    igure 1. Plot of amplitude versus thickness. Note the increase inmplitude over background as the tuning thickness /4 is ap-roached. Below tuning, the amplitude rolls off nearly linearly, andhe waveform approximates the derivative of the wavelet at /8.

    igure 2. Any arbitrary pair of reflection coefficients r1 and r2 can beepresented as the sum of even and odd components. The even pairas the same magnitude and sign, and the odd pair has the same mag-

    itude and opposite sign. The identity is illustrated in Figure 2. The Widess model assumeshat reflection-coefficient pairs are perfectly odd, which can be aood approximation for certain target classes such as a sand layer en-ased in a shale matrix. However, the assumption of an odd-reflec-ivity pair implies the worst possible resolution for thin beds. Even amall even component in the reflection-coefficient pair can increasehe resolvability of a layer significantly. The improvement in resolu-ion results from the fact that the even component constructively in-erferes as thickness approaches zero. In contrast, the odd compo-ent destructively interferes. Thus, the even component is more ro-ust against noise as thickness approaches zero see Tirado, 2004.

    We calculated peak frequency and peak amplitude from equationsiven by Chung and Lawton 1995. Figure 3a shows the effect ofhinning on the peak frequency of a reflection-coefficient pair withven and odd components. For the model, the total peak frequencyncreases with decreasing thickness and then returns to the peak fre-uency of the wavelet rather than that of the derivative of the wavelets predicted by the Widess model. Interestingly, the total peak fre-uency shows significant and continuous change down to zero thick-ess. Likewise, the total peak amplitude Figure 3b does not ap-roach zero with thickness as predicted by the Widess model.

    The example indicates that the reflection-amplitude trend canhow significant variation from the Widess curve Figure 1 as theayer thickness approaches zero when the even component is nonze-o. Thus, significant information below the Widess resolution limit isot captured by traditional amplitude-mapping techniques, whichssume equal and opposite reflection coefficients. Such examples ofnequal reflection coefficients at the top and base of a layer, whichre the rule rather than the exception for most real-world seismic re-ections events, reinforce the need for a more generalized approach

    o thin-bed amplitude analysis.Based on the fact that spacing between spectral peaks and notches

    s a deterministic function of layer thickness, our objective was toevelop a new algorithm to invert reflectivity using the constant pe-iodicity in the frequency domain. Our development started with thexpression for an impulse pair in the time domain, from which we

    a)

    b)

    igure 3. a Peak frequency and b peak amplitude as a function ofhickness for the even component, the odd component, and the total.n a, there is peak-frequency information below the tuning thick-ess. In b, total peak amplitude approaches the even-componentmplitude below tuning. Layer model parameters are r10.2, r20.1, and f0 30 Hz.

  • fa

    to

    s

    S

    m

    s

    r

    c

    tpts

    btlt

    m

    w

    fiie

    T

    w

    b

    w

    S

    w

    tTttr

    it

    e

    n

    t

    tlia

    a

    i

    w

    F2

    Fc

    Layer-thickness determination R39ormulated a numerical algorithm using complex spectral analysisnd then tested the algorithm on synthetic wedge models. We ex-ended our development to multiple layers, and we tested our meth-d on 3D seismic data from the Gulf of Mexico shelf, comparingeismic data, spectrally inverted seismic data, and well-log data.

    METHODS

    pectral-inversion methodWe applied windowed Fourier transforms to several reflectivityodels to generate data for spectral inversion and used complex

    pectral analysis to formulate the inversion algorithm. The algo-ithm described herein defines the inversion for reflectivity using theonstant periodicity of the amplitude spectrum for a layer of a givenhickness, taking advantage of the fact that spacing between spectraleaks and notches is precisely the inverse of the layer thickness inhe time domain Partyka et al., 1999; Marfurt and Kirlin, 2001. Es-entially, layer thickness can be determined robustly from a narrowand of frequencies with a high signal-to-noise ratio S/N. To provehis concept, note that the entire reflectivity spectrum for a singleayer could be reconstructed from amplitudes at three frequencies inhe absence of noise.

    Beginning with the expression for an impulse pair in the time do-ain as expressed by Marfurt and Kirlin 2001 Figure 4,

    gt r1 t t1 r2 t t1 T , 1

    here r1 is the top reflection coefficient, r2 is the base reflection coef-cient, t is a time sample, t1 is a time sample at the top reflector, and T

    s layer thickness. Locating the analysis point at the center of the lay-r yields

    gt r1 t T2 r2 t T2 . 2aking the Fourier transform of the shifted expression 2 gives

    gt, f r1 expi2 ft T2 r2 expi2 ft T2 , 3

    here f is frequency and gf is the complex spectrum. Simplifyingy using trigonometric identities and taking the real part yields

    Regf 2recos fT , 4here re is the even component of the reflection-coefficient pair.imilarly, the imaginary part of the complex spectrum is

    Imgf 2rosin fT , 5here ro is the odd part of the reflection-coefficient pair.Figure 5 shows plots for both the even and odd reflectivity spec-

    ra corresponding to equations 4 and 5 for a layer with thickness 10 ms and reflection coefficients r1 0.2 and r2 0.1. Al-

    hough both even and odd spectra show the same notch period, thewo are shifted by half of the frequency spacing. For the individualeal and imaginary components, the constant period in the spectrums related to the symmetrical location of the analysis point at the cen-

    er of the layer. This placement effectively divides the reflection-co- Ifficient pair into perfectly odd and even components, thereby elimi-ating the phase variation for each. The effect of violating this condi-ion is discussed inAppendix A.

    To maintain constant periodicity in the spectrum while shiftinghe analysis point away from the layer center, we compute the modu-us of the real and imaginary components of the spectrum, which isnsensitive to phase. Beginning with general expressions for the realnd imaginary time-shifted spectra,

    Ime2i ftgf 2ro sin fTcos2 ft 2re cos fTsin2 ft 6

    nd

    Ree2i ftgf 2re cos fTcos2 ft 2ro sin fTsin2 ft , 7

    t can be shown Appendix B that

    Ot,k GfdGfdf 2Tk sin2 fT , 8

    here Gf is amplitude magnitude as a function of frequency, kre

    2 ro2, and Ot,k is the cost function at each frequency. The so-

    xm (km)

    t(s)

    t2

    t1

    T

    Reflectivity = r1

    Reflectivity = r2

    igure 4. Two-layer reflectivity model from Marfurt and Kirlin,001.

    igure 5. Amplitude versus frequency plots for a even and b oddomponents of the reflection-coefficient pair r1 0.2 and r2 0.1.

    n this example, the even component is dominant.

  • le

    c

    fSS

    bitple

    a

    w

    pdtpe

    2o

    Ns

    fls

    M

    r

    w

    te

    tC

    w

    w

    ttv

    t

    m

    tn

    n

    u

    ts

    ifls

    s

    o

    a

    a

    tm

    n

    c

    r

    Frts

    a

    b

    Ffqssti

    R40 Puryear and Castagnaution to equation 8 occurs when the sum of cost functions Ot,k,valuated at each frequency, is minimized over the range of frequen-ies within the analysis band. One data term exists at every samplerequency, so the performance of the method is determined by the/N over a given analysis band i.e., more frequencies with a high/N yield a more stable and accurate inversion.We found the global minimum of equation 8 for a given analysis

    and by searching physically reasonable model parameters k and Tn two-parameter model space and minimizing the objective func-ion Figure 6. Although it is costly and impractical for more com-licated cases, the global search method guarantees the avoidance ofocal minima for the single-layer case. The remaining model param-ters then are determined by

    roGf24 k cos2 fT , 9re k ro2, 10

    igure 6. The difference or error function between the data and aange of model parameter pairs is calculated. The blue bulls-eye ishe correct model solution T 10 ms, k 0.02. The black arrowhows a local minimum.

    )

    )

    igure 7. a Forward-model and b inverse-model schemes usedor the synthetic. The plots show magnitude of amplitude versus fre-uency. In a, multiplication of the wavelet with the reflectivitypectrum and addition of noise in the frequency domain yields theeismic signal. In b, division of the seismic signal by the wavelet inhe frequency domain yields the noisy reflectivity band. A smooth-5ng filter produces the inversion band for input to the model.nd

    t11

    2i f ln gfr1 r2e2i fT , 11here t1 is the time sample at the top reflector r1 and gf is the com-lex spectrum for the reflection-coefficient pair. Equation 11 can beerived by taking the Fourier transform of equation 1 and solving for

    1. The reflection coefficients r1 and r2 can be recovered by recom-osing the odd and even components of the pair calculated usingquations 9 and 10, the reverse of the operation illustrated in Figure. Thus, it is straightforward to compute the remaining componentsf the layer-reflectivity model from the initial parameters k and T.ote that although the derivation Appendix B of the algorithm as-

    umes that the even reflectivity component is greater than the odd re-ectivity component, the solution is the same for the antithetical as-umption.

    odeling resultsWe test the method by convolving a 30-Hz Ricker wavelet with

    eflection-coefficient pairs that have various ratios. We produceedge models with 4-ms sampling for a predominantly odd reflec-

    ion-coefficient pair, r10.2 and r2 0.1, and a predominantlyven reflection-coefficient pair, r1 0.2 and r2 0.1. The tuninghickness of a thin-bed model with a Ricker wavelet is given byhung and Lawton 1995:

    tR6

    2 f0, 12

    here f0 is the dominant wavelet frequency. For a 30-Hz Rickeravelet, tR 13 ms. The convolution of the wavelet with the reflec-

    ion-coefficient pair in the time domain is equivalent to multiplica-ion with the reflectivity spectrum in the frequency domain. The in-ersion performs perfectly in the absence of noise for layers of anyhickness.

    To achieve a more realistic model, we added noise in the time do-ain, and we measured and controlled the noise level by computing

    he ratio of the area under the spectrum of the signal to that of theoise in the frequency domain. We tested the model with 1% and 5%oise levels. The forward-modeling procedure is illustrated in Fig-re 7a.

    The addition of noise causes instability in the inversion for veryhin layers, partly because the reflectivity spectrum approaches a flatpectrum as T approaches zero. We mitigated this problem by apply-ng the arbitrary constraint0.03k0.03 to ensure that the re-ection-coefficient strength could not exceed what typically is ob-erved in seismograms.After Fourier-transforming the time-domainignal, we removed the wavelet overprint by dividing the magnitudef the amplitude of the seismic signal by that of the wavelet at eachnalysis frequency.

    We tested the inversion at different noise levels while varying thenalysis band and smoothing filter. These experiments showed thathe optimal analysis band and the optimal smoothing filter are deter-

    ined by the noise level. We achieved optimal results for the 1%oise case using a 25-Hz bandwidth sampled at 2-Hz frequency in-rements and centered on the peak signal frequency. A signal is cor-upted by noise, so it was necessary to narrow the bandwidth. For the

    % noise case, we achieved optimal results with a 20-Hz bandwidth.

  • nitc

    n

    ttba

    w

    w

    e

    v

    r

    n

    r

    r

    tKltt

    dps

    e

    tlc

    c

    a

    t

    lo

    a

    ptta

    pllttt

    a

    bttt

    s

    r

    o

    o

    ls

    ts

    lfs

    itl

    E

    tis

    idlf

    a

    b

    Fotn

    Layer-thickness determination R41More smoothing is required to stabilize the derivative operator asoise is added. The process resulting in the smoothed inversion bands illustrated in Figure 7b. We took derivatives of the magnitude ofhe amplitude with respect to frequency at each observation frequen-y sample using a first-order central difference approximation. Fi-ally, we multiplied the magnitude of the amplitude at each observa-ion frequency by the derivative of the magnitude of the amplitude athat frequency and minimized the error between the model definedy equation 8 and the data for the range of frequencies within thenalysis band Figure 6.

    We applied the inversion defined by equation 8 using a 256-ms-indow fast Fourier transform FFT for spectral decomposition toedge models that have predominantly odd and even reflection-co-

    fficient pairs Figure 8. Figure 9a and b shows the result of the in-ersion black compared with the true reflectivity model green anded for 1% and 5% levels of noise, respectively, for the predomi-antly odd reflection-coefficient pair; Figure 9c and d shows theseesults for the predominantly even pair. The tuning thickness is rep-esented by the vertical black line.

    We compared the 5% noise results with the corresponding ampli-ude-mapping results brown using tuning analysis Widess, 1973;allweit and Wood, 1982. The algorithm performed nearly perfect-

    y for both configurations at 1% noise levels as expected, illustratinghe principle that noiseless data could yield extremely high resolu-ion.

    Although the noise level is unrealistically low for most seismicata, the results highlight the importance of meticulous noise sup-ression during acquisition and processing. For thin layers, the re-ults are useful far below tuning. For the 5% noise cases, the absoluterror does not increase significantly for very thin layers because ofhe reflectivity constraint, although the percent error increases asayer thickness decreases. Reflection-coefficient estimates wereomparably accurate.As expected, accuracy deteriorates as noise in-reases beyond 5%. Thus, for a given wavelet peak frequency andnalysis band, noise level rather than tuning thickness determineshe limit of resolution.

    The tuning-analysis predictions are shifted toward thickerayers for the even and odd components because of the assumptionf equal-magnitude reflection coefficients at the top and base. Thessumption in the relationship between amplitude and thickness de-ends on the particular mapping scheme. We plotted the results ofhe amplitude-mapping techniques that generated the least error inhickness prediction. For the predominantly odd pair, the optimalmplitude-mapping method was a simple linear regression fromeak tuning to zero thickness. However, this technique generatedarge errors and negative thickness predictions for the predominant-y even pair. So for the predominantly even pair, we assumed both ofhe reflection coefficients were equal to the peak background ampli-ude for the thick layer and mapped thickness from the tuning ampli-ude to zero thickness.

    We tested the inversion on a reflector model that violates the basicssumptions of the method. The model includes two layers definedy three reflectors, all with reflection coefficients equal to 0.1. Theop layer has twice the thickness of the lower layer. We decomposedhe spectrum using an 80-ms DFT with a Gaussian taper centered onhe thicker layer.

    Figure 10 shows the model green and red and the resulting inver-ion black. The predicted layer thickness is measured from the topeflector to the black line, yielding a thickness value greater than that

    f the thicker layer but less than that of the two layers combined. The ebserved predicted thickness results from the interference of the twoayers in the frequency domain, creating a period corresponding to aingle layer that is slightly thicker than the thickest layer. The reflec-ion-coefficient predictions for the lower reflector are closer to theum of the two base reflection coefficients r2 r3 0.2 for thinayers and to the single base reflection-coefficient value r2 r3

    0.1 for thicker layers. In practice, additional spikes widely spacedrom the reflectors of the layer should be considered noise for theingle-layer model.

    Although we value the single-layer model for its ease of invert-bility, it is necessary to extend the inversion scheme so that it simul-aneously can invert seismograms containing multiple interferingayers for most real cases.

    xtension of the method to multiple layersRecognizing that a seismogram can be represented as a superposi-

    ion of impulse pairs, the inversion for the properties of a single layers extended easily to encompass a general reflectivity-series inver-ion by considering the spectrum versus time acquired using a mov-ng window as a superposition of interference patterns originating atifferent times. The inversion process for reflection coefficients andayer thickness is performed simultaneously for all impulse pairs af-ecting the local seismic response.

    )

    )

    igure 8. Original reflectivity wedge models for a a predominantlydd reflection-coefficient pair and b a predominantly even reflec-ion-coefficient pair. The vertical black line defines the tuning thick-ess for the 30-Hz Ricker wavelet convolved with the reflection-co-

    fficient pair blue.

  • aw

    fa

    v

    p

    w

    til

    a

    c

    F eflectiop eflectiop iginala techniw al blact he reflec

    R42 Puryear and CastagnaLet us represent the reflectivity series rt as a summation of evennd odd impulse pairs:

    rt

    retII t Tt dt

    rotII t Tt dt ,13

    here ret and rot are the magnitudes of the impulse pairs as aunction of time, Tt is the time series of layer time thicknesses, II isn odd impulse pair, and II is an even impulse pair. Assuming a con-olutional seismogram and known wavelet wt, f, a spectral decom-osition of a seismic trace st, f is then

    st, f wt, f

    tw

    tw

    retcos fTt

    irotsin fTtdt , 14here tw is window half-length. The multilayer case involves more

    han two reflectors, so it is necessary to use an objective function fornversion that properly accounts for interference between multipleayers.

    ) b)

    ) d)

    igure 9. Result of the thickness inversion of a predominantly odd rair with a 1% noise and b 5% noise and a predominantly even rair, also with c 1% noise and d 5% noise. The plot shows the ornd red along with the inversion black and the amplitude-mappinghich uses the incorrect assumption of equal reflectivity. The vertic

    he tuning thickness for the 30-Hz Ricker wavelet convolved with tient pair blue.If the wavelet spectrum is known, we can solve for rt and Tt byptimizing the objective function Ot,re,ro,T by

    Ot,re,ro,T

    fL

    fH

    eRest, f/wt, f

    tw

    tw

    retcos fTtdt oImst, f/wt, f

    tw

    tw

    rotsin fTtdtdf , 15here f l is low-frequency cutoff, fh is high-frequency cutoff, and e

    nd o are weighting functions, the ratio of which can be adjusted tond an acceptable trade-off between noise and resolution. For higho/e, the reflectivity approaches the Widess model, and the resolu-

    tion limit becomes /8. We summarize the multi-layer inversion process in Figure 11.

    Multilayer synthetic exampleTo validate the multilayer inversion technique

    in a controlled situation, we generated a modelcontaining several arbitrary layers Figure 12a,from which we generated a synthetic trace Fig-ure 12b. The reflectivity spikes were convolvedwith a 30-Hz Ricker wavelet. The identical tracewith variable random noise is repeated in Figure12 for clarity. We tested the inversion using win-dows of different lengths and centered at differentlocations along the synthetic trace. Assuming noa priori insight, we set the weighting function re-lationship o e and minimized the objectivefunction given by equation 15 using a standardleast-squares conjugate-gradient inversion.

    We inverted the data from a 200-ms Gaussian-tapered Fourier transform, thereby recovering theoriginal model Figure 12c. Shortening the totalwindow length to 100 ms and maintaining theGaussian taper, we computed the Fourier trans-form for time samples between 50 and 150 ms,respectively. The window was shifted one timesample at a time, with the results of each previouswindow acting as a constraint for the next inver-sion. If no new reflectors appeared in the windowor exited the window as it shifted downward, theresult was the same as in the previous window. Ifthere were no reflectors in the window, the algo-rithm inverted the noise, a result that was not par-ticularly problematic because the noise was notamplified.

    n-coefficientn-coefficientmodel greenque brown,k line definesction-coeffi-o

    w

    a

    fi

  • io

    s

    r

    s

    v

    w

    t

    C

    io

    c

    u

    te

    n

    TFdt1

    r

    w

    tia

    tdpc

    a

    Filflts

    Ft

    a

    c) d)

    Fwso1so

    Layer-thickness determination R43The results for the windows centered at 50 and 150 ms are shownn Figure 12d and e. As expected, the shorter windows divide theverall longer series of reflectors into two isolated sets. By superpo-ition, we can sum the two sets of reflectors to obtain the longer se-ies. Although the window length can be varied according to the de-ired result, there are practical limits to the length of an optimal in-ersion window, which often are defined by trial and error. If theindow is too short, frequency resolution suffers; if the window is

    oo long, time resolution is lost see Castagna et al., 2003.

    REAL DATA RESULTS

    omparison to well-log dataWe studied the results of the application of the multilayer spectral-

    nversion process Figure 11 to two seismic data sets from the Gulff Mexico shelf. Well information for lithologic interpretation, in-luding P-wave, resistivity, and deep induction, was available Fig-re 13. We created a synthetic tie between the input seismic data andhe well-log data, stretching the well-log data to time without refer-nce to the inverted data for an unbiased, quantitative layer-thick-ess comparison between the well-log data and the inverted data.he wavelet extracted from the well for the synthetic is shown inigure 14. We also visually compared the well-log data, the invertedata, and the original data to assess the difference in vertical resolu-ion between the inverted data and the original data Figures 15 and6.

    In the original seismic tie to the well Figure 13, we achieved aelatively good fit r 0.64 between the reflectivity convolvedith the extracted wavelet and the seismic trace. However, because

    he seismic data are much lower in frequency than the log data, the fits useful only as an approximation for aligning gross lithologic pack-ges.Agreat deal of useful information is lost to the seismic wavelet.

    The thickness inversion provides a significantly better representa-ion of the layering observed in the log data than the original seismicata does. Figure 15 shows the spectral inversion for reflectivity, dis-layed with a 90 phase shift to emphasize relative impedancehanges and a slight time shift from the original synthetic to providetie to the well with higher fidelity. The need for this time shift be-

    igure 10. Result of the inversion for a two-layer model red, violat-ng the assumptions of the method. For thicker layers, the predictedayer thickness from the inversion black, measured from the top re-ector, falls between the thickness of the thicker layer and that of the

    otal package. As the reflectors converge, the inversion becomes un-

    table. te)

    igure 12. a The original model is convolved with a 30-Hz Rickeravelet to create b a synthetic seismogram. c The spectral-inver-

    ion result using a 200-ms window centered at 100 ms recovers theriginal reflectivity series. d The spectral-inversion result using a00-ms window centered at 50 ms and e the spectral-inversion re-ult using a 100-ms window centered at 150 ms recover the portionsf the reflectivity series contained within the window. By superposi-igure 11. Flowchart for the method of the inversion, where dout ishe output trace, tc is the time sample at the center of the window, andt is the sampling interval.

    ) b)ion, c d e.

  • cu

    ts

    tgv

    dthtbyw

    c

    tbw

    t

    v

    n

    C

    s

    ts

    tiw

    fdt

    FctT

    Ff

    Fdlcsolved at the tuning thickness.

    FcadAw

    R44 Puryear and Castagnaomes apparent only after reflectivity inversion. The lithologic col-mn shows sand and shale packages interpreted primarily from thehree logs see Figure 13. We initially distinguished sands andhales using a gamma-ray log, but we used resistivity, deep induc-ion, and impedance for the bulk of the interpretation becauseamma-ray data were absent in the well with an available P-waveelocity log.

    Generally, sands correspond to higher resistivity and separation ofeep induction and shallow-resistivity curves as a result of mud-fil-rate invasion. In the interval shown in Figure 15, the sands haveigher impedance than the shales. Comparing the lithologic columno the inverted data, the thickness inversion clearly shows layeringelow the tuning thickness the peak frequency of the data is 16 Hz,ielding a one-quarter-wavelength resolution of about 31 ms. Aell-resolved sand-shale sequence that has a total thickness very

    lose to the tuning thickness is indicated also in Figure 15. Althoughhe thickness inversion effectively delineates the layering sequenceelow tuning, it fails to capture gradational impedance changesithin thin layers, as in the case of shale grading to sand, shown in

    he lithologic column.The following example demonstrates the vast improvement in

    ertical resolution for discrete layers that is achieved by using thick-ess inversion rather than the original data.

    omparison to conventional seismic dataViewing the comparison of the thickness inversion to the original

    eismic data in Figure 16, both with a90 phase rotation, it is clearhat boundaries between layers are indistinct on the original seismicection. Layers below the 31-ms tuning thickness are not resolved inhe original seismic data. Geologic detail is obscured by the wavelet-nterference patterns, which become more apparent when comparedith the inversion. A skilled interpreter can decipher meaningful in-

    ormation embedded in the wavelet-interference patterns, but it isesirable to remove these artifacts altogether to allow direct accesso the underlying geology.

    igure 16. Original seismic data left compared with the lithologicolumn center and the thickness inverted data right. Both imagesre phase-rotated by90. Note the failure of the original seismicata to delineate thin layering as compared with the inverted data.lso note that the inverted data do not resolve gradational changesigure 15. Impedance log left compared with spectrally invertedata right. Sands are of higher impedance in this interval. Theithologic section interpreted from the impedance log shows a closeorrespondence to the inverted data. Note the two-layer sequence re-igure 14. The phase and amplitude spectra of the wavelet extractedrom the well. The peak frequency is 16 Hz.igure 13. Well-log data, including deep induction, resistivity, andomputed impedance, long with the synthetic tie blue, the trace athe well red, and the seismic traces surrounding the well black.he correlation coefficient r is 0.64.ithin an individual layer.

  • Ttio

    ts

    c

    tb

    u

    a

    t

    S

    o

    p

    ls

    fiv

    ita

    a

    lTr

    ge

    g

    bw

    ta

    Fsantl

    a

    b

    Fawb

    Fdnvl

    Layer-thickness determination R45hickness comparisonWe quantitatively compared thicknesses from the well-log data to

    hose from the spectral reflectivity inversion, interpreting the layer-ng from the well-log data, which were stretched from depth to timen the original synthetic, and comparing the results to the invertedhicknesses as defined by zero crossings on the90 phase-rotatedpectral reflectivity inversion.

    We performed the interpretation between 1700 and 2900 ms. Theomparison is plotted in Figure 17, with excellent correlation be-ween well-log data thicknesses and the predicted thicknesses pinkelow the tuning line red. The mean error for the two data sets is0.5 ms, indicating that the method is generally accurate and

    nbiased. The square of the correlation coefficient is R2 0.94 withstandard deviation of 3.10 ms, corresponding to precision in

    hickness estimation on the order of 15 ft 4.5 m for this example.

    tratigraphic interpretationFigures 1820 show stratigraphic examples in which both the

    riginal seismic data and the spectrally inverted data have a90hase to highlight layer boundaries.

    An example of the wavelet overprint effect is observed on thearge-scale seismic line comparison of Figure 18. On the originaleismic line, an apparent discontinuity is inferred at about 1315 msFigure 18a. The discontinuity might be interpreted as a localizedault with minimal offset or as a stratigraphic discontinuity in layer-ng. However, comparison with the inverted data Figure 18b re-eals another picture. Although the inverted data show more detailn general, the apparent discontinuity in layering does not exist onhe inverted section despite the fact that the inverted data are gener-ted using a trace-by-trace operation that makes no assumptionbout lateral continuity.

    On the inverted section, the apparent disruption in layering actual-y represents a lateral change of rock properties within a given layer.he discrepancy points to the apparent discontinuity in reflection ar-

    ival times seen on the original data; it is not a geologic feature but aeophysical effect. Specifically, it is a shift in the wavelet-interfer-nce pattern caused by an impedance change that resembles a smalleologic layering discontinuity in the seismic image.

    igure 17. Plot of predicted thicknesses from the inversion pinkquares versus well-log interpreted thickness blue line is a 1:1 di-gonal, showing a strong correlation between the two. The thick-esses were interpreted between 1700 and 2900 ms. The tuninghickness is marked by the red line, and accuracy is maintained be-

    ow one-eighth of a wavelength. aThe smaller-scale image in Figure 19 shows significant lateralreaks in layering that might be interpreted as discrete sand bodiesith possible erosion of previously continuous layers. These fea-

    ures can be caused by different types of downslope transport mech-nisms such as channel incision. The horizons on the original dataFigure 19a are difficult to continue in places black arrows. How-

    )

    )

    igure 18. a The original seismic data show a small discontinuity.b The thickness-inverted data reveal a strikingly continuous layer,strong indication that the geologic discontinuity seen in a is aavelet effect rather than a real subsurface feature. The phase foroth images in90, and red indicates higher impedance.

    a)

    b)

    igure 19. a The original seismic line has significant stratigraphiciscontinuities black arrows that might be interpreted as the termi-i of discrete depositional lobes. b The spectrally inverted data re-eal a laterally continuous layering characteristic of undisturbedayer-cake geology. The phase on both images is90. Timing lines

    re 20 ms, and red indicates higher impedance.

  • ela

    e

    l

    gitt

    s

    gc

    bc

    v

    i

    Mm

    pv

    c

    c

    s

    ps

    s

    tr

    Wt

    sFritap

    a

    b

    Fwfml

    a

    b

    FctTn

    R46 Puryear and Castagnaver, the spectrally inverted data Figure 19b, which assume no re-ationship between neighboring traces, show striking continuitylong the same horizons. Once again, the complex wavelet-interfer-nce pattern creates an illusory geologic scenario that accompaniesimited resolution.

    The previous examples demonstrate artifacts that resemble geolo-y, and Figure 20 shows an example of the same wavelet effect eras-ng geologic information. In the original seismic data Figure 20a,he apparent pinch-out of a low-impedance layer is observed, withhe upper and lower events merging below the resolution of the layer

    a)

    b)

    igure 20. a The original seismic data shows a pinch-out white ar-ow where the thin layer becomes unresolved. b The inverted datamages the pinch-out much farther updip.An apparent erosional fea-ure black circle is resolved on the inversion. The phase on both im-ges is90. Timing lines are 20 ms, and red indicates higher im-edance.

    )

    )

    igure 21. Comparison of a the original zero-phase seismic dataith b the spectrally inverted data, which are phase-rotated90

    rom a. Channels white arrows show the base of channels haveore relief and more curvature on the original seismic data. Timingines are 10 ms, and red indicates higher impedance. lwhite arrow. However, the spectrally inverted data Figure 20bhow the same low-impedance layer imaged much farther updip, to-ether with the resolved bounding layers. In addition, an apparent lo-alized broadening or bulge in the wavelet in the original data justelow the first two timing lines is resolved as a possible erosional in-ision on the inverted data. Such improved detection of stratigraphicariation has significant implications for better reservoir character-zation and delineation.

    We tested the method on a line of data from a shallow Gulf ofexico data set with known large incision features, previouslyapped using the coherence attribute. Figures 21 and 22 show zero-

    hase original seismic images and90 phase-rotated spectral-in-ersion images. Typically, seismic images of channels show signifi-ant relief from the levy to the thalweg, which appeals to the intuitiveoncept of a curved-channel geometry.

    Figure 21a is an example of a pair of adjacent channels, showing atrongly curved geometry on the original seismic data. The90hase-rotated spectrally inverted section of the data Figure 21bhows an alternative image of the channels in which the curvatureeen in the channel profile is less prominent, hinting at the possibilityhat some component of the curvature can be attributed to the rapidock-property changes known to occur across the strike of a channel.

    e believe further investigation of this phenomenon using well con-rol is warranted.

    Figure 22 shows another large channel imaged on the originaleismic data and on the90 phase-rotated spectral inversion, re-

    )

    )

    igure 22. A large channel white arrows show the edges of thehannel imaged on a the original zero-phase seismic data and bhe spectrally inverted data, which are phase-rotated90 from a.he thin-bed layering of the channel edges and overall vertical chan-el extent are imaged more precisely on the inverted data. Timing

    ines are 10 ms, and red indicates higher impedance.

  • stis

    dtIpz

    a

    f

    c

    pv

    n

    s

    m

    tt

    r

    o

    ifltfm

    ffb

    c

    r

    tpm

    s

    iit

    s

    w

    tpe

    a

    c

    v

    X

    gMs

    a

    e

    ttt

    A

    T

    R

    w

    o

    o

    tpo

    s

    p

    Layer-thickness determination R47pectively. The tuning effect in the original data confounds channel-hickness interpretation as the channel thins toward the levies. Theres also an ambiguity in the placement of the top and base boundingurfaces of the channel related to the wavelet phase. The invertedata show a clearer picture of the channel geometry, with constanthinning of the channel-fill wedges toward the edges of the channel.n addition, the top and base bounding surfaces of the channel can beicked more precisely on the inverted data at the sharply definedero crossings with less guesswork in the placement of horizons.

    Thus, in the workflow of seismic interpretation, spectral inversiondds visual information that can contribute to delineating geologiceatures of interest such as channels.

    CONCLUSIONS

    Beginning with a generalized theory of reflectivity, spectral de-omposition is used as a tool to unravel the complex interferenceatterns created by thin-bed reflectivity. These patterns can be in-erted to obtain the original reflectivity. We developed and studiedew analytical methods for spectral inversion based on complexpectral analysis. Spectral inversion yields accurate thickness deter-inations below tuning, using the inverse relationship between

    hickness and the constant periodicity of spectral interference pat-erns.

    Representing the seismogram as a superposition of simple layeresponses constitutes a means of imposing on the inversion the a pri-ri assumption that sedimentary rocks occur as layers with discretenterfaces at the top and base and can be represented as such in a re-ectivity series. When this assumption is valid, the consequence is

    hat on the inverted reflectivity trace, there is geologically meaning-ul information at frequencies outside the band of the original seis-ic data. When this assumption is false, the recovered frequency in-

    ormation outside the band of the original seismic data will also bealse. For example, smooth impedance transitions will be inverted aslocky steps in impedance.

    Spectral shape information obtained from spectral decompositionan be used to drive an inversion with significantly greater verticalesolution than that of the original seismic data, thereby improvinghickness estimation, correlation to well logs, and stratigraphic inter-retation. These results are achieved without using well-log infor-ation in the inversion as a starting model or as a constraint. The re-

    ulting inversion therefore is unbiased by preconceived ideas.As ev-denced by the results of applying the method to real data, spectralnversion has great potential as a practical tool for seismic explora-ion.

    The spectral-inversion methods described in this work demon-trate improvement in vertical resolution; however, we did not useell-log information after the wavelet-removal step. It is desirable

    o investigate the effectiveness of using well-log data to further im-rove vertical resolution of interbedded layers or gradational chang-s within layers that are not revealed by seismic spectral inversionlone. In addition, thickness constraints from spectral inversionould be used as input for more accurate model-based impedance in-ersion.

    ACKNOWLEDGMENTS

    The authors would like to thank Gene Sparkman, Carlos Moreno,

    ianhui Zhu, and Oleg Portniaguine of Fusion Petroleum Technolo- aies for their help and support. Thanks also to Kurt Marfurt and Scottorton for assistance, suggestions, and contributions. Financial

    upport was provided by ExxonMobil and Shell.

    APPENDIX A

    THE SHIFT EFFECT

    When examining the real and imaginary components separately,phase shift occurs if the analysis window is not centered on the lay-r. To study this effect in more detail, we revisited the original equa-ions. The shift theorem says that a time sample shift t away fromhe layer center tc in the time domain is equivalent to a phase ramp inhe frequency domain:

    gtc t e2i ftgf . A-1pplying this equivalency,

    e2i ftgf cos2 ft i sin2 ft2re cos fT i2ro sin fT . A-2

    aking the real component of equation A-2,

    Ree2i ftgf 2re cos fTcos2 ft 2ro sin fTsin2 ft .

    A-3

    earranging yields

    Ree2i ftgf 2rocos fTcos2 ft sin fTsin2 ft 2re rocos2 ftcos fT,

    2ro cos2 ft T/2 2re rocos2 ftcos fT , A-4

    hich has the form of a modulation and represents the spectral plotsf time-shifted models. A similar expression can be derived for thedd component. The phase shift corresponds to a sinusoidal modula-ion of the signal, which can be viewed as an interference pattern su-erimposed on another interference pattern. Furthermore, the periodf the interference pattern is determined by the magnitude of thehift.

    APPENDIX B

    INVERSION-MODEL DERIVATION

    Applying the shift theorem equation A-1 and taking general ex-ressions for the real and imaginary spectra,

    Ime2i ftgf 2ro sin fTcos2 ft 2re cos fTsin2 ft

    B-1nd

  • Ree2i ftgf 2re cos fTcos2 ft 2ro sin fTsin2 ft .

    B-2

    Er

    R

    G

    T

    M

    w

    V

    t

    gr

    r

    t

    t

    Tt

    gRIr

    r

    G

    Variable Definition

    dGf / df Derivative of magnitude of amplitude with respectto frequency

    k Even component of reflectivity squared minus odd

    t

    r

    IIs

    w

    t

    OO

    C

    C

    C

    C

    K

    M

    P

    P

    P

    P

    P

    T

    W

    R48 Puryear and Castagnaxpressing the amplitude spectrum and settingt 0 as a constanteference,

    Gf Ree2i ftgf2 Ime2i ftgf2

    4re2 cos2 fT 4r02 sin2 fT . B-3earranging terms,

    f 2re2 r02cos2 fT r02 cos2 fT r02 sin2 fT .

    B-4aking the derivative,

    dGfdf

    2Tre2 ro

    2cos fTsin fTre2 ro2cos2 fT r02

    . B-5

    ultiplying and simplifying using trigonometric identities yields

    GfdGfdf 4Tre

    2 r02cos fTsin fT

    2Tk sin2 fT , B-6here k re2 ro2.

    NOMENCLATURE

    ariable Definition

    Seismic record timet Time-domain impulse pair1 Top reflector in a two-reflector model2 Base reflector in a two-reflector model1 Time at top reflector in a two-reflector model2 Time at base reflector in a two-reflector model

    Layer two-way traveltime thicknessc Time at layer center in a two-reflector modelt Time shift

    f Frequencyf Frequency-domain impulse responsee Real component of a function

    m Imaginary component of a functione Even component of the reflection coefficiento Odd component of the reflection coefficientf Magnitude of amplitude as a function of frequencycomponent squaredR Tuning thicknessfo Wavelet peak frequencyt Reflection-coefficient series as a function of time

    Convolutional placeholderI Even impulse pairI Odd impulse pairt, f Time- and frequency-varying seismic tracet, f Time- and frequency-varying seismic wavelet

    w Window half-lengthk,T Frequency-varying objective functiont,re,ro,T Time- and frequency-varying objective functione Even-component weighting functiono Odd-component weighting function

    fL Low-frequency cutofffH High-frequency cutoff

    REFERENCES

    astagna, J. P., 2004, Spectral decomposition and high resolution reflectivityinversion: Presented at the Oklahoma Section Meeting, SEG, wild-horse.insinc.com/cseg/castagna/ .

    astagna, J. P., S. Sun, and R. W. Siegfried, 2003, Instantaneous spectralanalysis: Detection of low-frequency shadows associated with hydrocar-bons: The Leading Edge, 22, 120127.

    hopra, S., J. P. Castagna, and O. Portniaguine, 2006a, Seismic resolutionand thin-bed reflectivity inversion: Canadian Society of Exploration Geo-physicists Recorder, 31, 1925., 2006b, Thin-bed reflectivity inversion: 75th Annual InternationalMeeting, SEG, ExpandedAbstracts, 20572061.

    hung, H., and D. C. Lawton, 1995, Frequency characteristics of seismic re-flections from thin beds: Canadian Journal of Exploration Geophysicists,31, 3237.

    allweit, R. S., and L. C. Wood, 1982, The limits of resolution of zero-phasewavelets: Geophysics, 47, 10351046.arfurt, K. J., and R. L. Kirlin, 2001, Narrow-band spectral analysis andthin-bed tuning: Geophysics, 66, 12741283.

    artyka, G. A., 2005, Spectral decomposition: SEG Distinguished Lecture,http://ce.seg.org/dl/spring2005/partykaabstract.shtml .

    artyka, G. A., J. A. Gridley, and J. A. Lopez, 1999, Interpretational aspectsof spectral decomposition in reservoir characterization: The LeadingEdge, 18, 353360.

    ortniaguine, O., and J. P. Castagna, 2004, Inverse spectral decomposi-tion: 74th Annual International Meeting, SEG, Expanded Abstracts,17861789., 2005, Spectral inversion: 75th Annual International Meeting, SEG,ExpandedAbstracts, 16381641.

    uryear, C. I., 2006, Modeling and application of spectral inversion for deter-mination of layer properties: M.S. thesis, University of Houston.

    uryear, C. I., and J. P. Castagna, 2006, An algorithm for calculation of bedthickness and reflection coefficients from amplitude spectrum: 76thAnnu-al International Meeting, SEG, ExpandedAbstracts, 17671770.

    irado, S., 2004, Sand thickness estimation using spectral decomposition:M.S. thesis, University of Oklahoma.idess, M., 1973, How thin is a thin bed?: Geophysics, 38, 11761180.