You are on page 1of 6

www.seipub.

org/ie InformationEngineering(IE)Volume4,2015
doi:10.14355/ie.2015.04.008

FirstBreakoftheSeismicSignalsBasedon
InformationTheory
HongJiuZhang
ResearchinstituteofBPG,CNPC,Zhuozhou072750,China
paulra@126.com
Abstract
FortheseismicdatawithlowSNRs,thefirstarrivalautomaticpickingmethodisveryimportantbutdifficult.Inthepaper,we
proposedanewmethodbasedonthemutualinformationininformationtheory.Themutualinformationbetweensignalsand
noises is zeros, thus random noises have less effects on first arrivals pickup. The paper compares the principle of STA/LTA,
AIC,fractaldimensionofthreekindswiththeproposedmethodforseismicdatafirstbreakpickingmethod,andatthesame
times,thepaperpresentsadetailedtestandverificationofthesimulationdata,andcomparesfirstbreakpickingaccuracyand
efficiencyofthethreealgorithmsthroughactualdatawithdifferentS/Nratios.TheresultsshowthatforthedatawithhighS/N
ratio, first break picking accuracy of these four methods is relatively high. When SNR decreases, firstarrival time that the
proposedmethodpickshashigherprecisionandgoodnoiseimmunity.However,mutualinformationbasedmethodhaslower
efficiency and is limited by algorithm principle. it is difficult to separately pick first breaks for fractal dimension and AIC
method.Soitisaverygoodmethodtoidentifyseismiceventsanddeterminepreliminarilythetimerangeoffirstbreaksbythe
proposedmethod.
Keywords
SeismicData;FirstArrivals;Pickup;InformationTheory;MutualInformation

Introduction
Inseismicexploration,firstbreakpickingisthetaskofdetermining,givenasetofseismictraces,theonsetsofthe
firstsignalarrivalsasaccuratelyaspossible.Ingeneral,thesearrivalsareassociatedwiththeenergyofrefracted
wavesatthebaseoftheweatheringlayerortothedirectwavethattravelsdirectlyfromthesourcetothereceiver
[5].
The accurate determination of the first arrivals onset firstbreak times is needed for calculating the static
corrections, a fundamental stage of seismic data processing. Clearly, the effectiveness of reflection and
refractionbased methods of static corrections depends on the pickingprocess reliability. At the same times,
applications such as nearsurface tomographic static corrections tomographic statics require rapid automated
detectionofthesignalfirst.
Generally, firstbreak quality is related to the nearsurface structure, source type, and signaltonoise ratio S/N
conditions.Asaconsequence,theautomatedpickingoffirstbreakscanbeaverydifficulttaskifdataareaacquired
in complex nearsurfaces scenarios or if the S/N is low. Moreover, if the source wavelet is zerophase as when
vibroseissourcesareused,thesweepcorrelationoftenproducessidelobesthatarrivebeforethefirstbreak,thus
makingthepickingprocessevenmoredifficult.
First arrival pickup of seismic waves, so far, has had a lot of methods. According to the criterion, firstarrival
picking algorithms can be divided into several types of methods, including the coherence method [3, 4]
crosscorrelationmethod[2],neuralnetworkmethod[1]andfractaldimensionmethods[7].Forthecoherenceand
neural network methods, some kinds of patterns are assumed for picking the first arrivals. Therefore, pattern
recognitionofthistypeiseffectiveifasimpleearthmodelexiststomodeltheearthstructure.However,thissimple
earthmodel rarely matches the nearsurface conditions when studied with the detail required of most modern
surveys [6]. The advantages of the crosscorrelation methods are that the algorithm is based on tracebytrace
evaluationofthefirstarrivaltimesandareconsideredtobemostappropriateforthenearsurfacesurveys.
In the paper, we proposed a robust method of firstbreak picking for data sets with high noise levels through

38

InformationEngineering(IE)Volume4,2015 www.seipub.org/ie

theuse of the information theory on seismic records. Using synthetic shot records with various noise levels, we
showedthattheperformanceofthisproposedmethodenhancesfirstarrivals,whichhelpsinpickingthem.This
was particularly true when the noise level was high where picking on raw amplitudes completely fails. The
method can be used to guide better the subsequent careful picking of first arrivals and requires one forward. In
contrast to methods based on tracebytrace picking that often fail to pick some traces, the proposed method
automaticallyinterpolatesmissingpicks.
Basics of Information Theory
Theaverageamountofinformationgainedfromagivendiscretespace X istheentropy Y ,
H ( X ) P xi log P xi (1)
i

Ifthelogistakentothebasetwo, H isinunitsofbits.Entropyofinformationininformationtheorycanbedefined
as follows. For given discrete probability space, which expressed information source and the information source
defined random variable I . The mathematical expectation of I is entropy of information of information source,
whoseunitisbit/symbol.
H ( X ) E[ I ( x)] p( xi ) log p( xi ) (2)
i

Theconventionalmutualinformationhasbeendefinedas:

I X Y X Y f XY x y ln

f XY x y

f X x fY y

dxdy (3)

where, f XY x y isthejointprobabilitydensityfunction(PDF),and f X x and fY y arethemarginalPDFsof


variables X and Y ,respectively.
Mutualinformationcanalsobeequivalentlyexpressedas:
I ( X Y ) H (Y ) H (Y X )

(4)
I ( X Y ) H ( X ) H (Y ) H ( XY )

Actually, the mutual information is very difficult to calculate since the probabilities of variables X and Y are
quite difficult to get. In the paper, we propose a new method to calculate the mutual information based on the
recursiveidea.Inprinciple,themutualinformationisakindofmeasuringhowdependentthevariableof X are
onthevariableof Y .
Bymakingtheassignment, s q x t x t T ,wecanconsiderageneralsystem S Q ,thentheuncertainty

ofmeasurementof q ,given S Q ,is

H Q si Pqs q j si log Pqs q j si (5)


j

where Pqs q j si istheprobabilitythatameasurementof q willyield q j ,giventhatthemeasuredvalueof s ,


is si .Atthesametime,theaverageuncertaintyinameasurementof x atthetime t T canbeshowngiventhat
x hasbeenmeasuredattime t ,

H Q S Ps si H Q si H S Q H S (6)
i

where,
H S Q Pq s q j si log Pq s q j si (7)
i j

H Q istheuncertaintyof q ,and H Q S istheuncertaintyof q givenameasurementof s .Sotheamountthat

39

www.seipub.org/ie InformationEngineering(IE)Volume4,2015

ameasurementof s reducestheuncertaintyof q isaccordingtotheequation(4)

I Q S H Q H Q S (8)
Itisimportantthatmutualinformationisnotafunctionofthevariables s and q ,butthatitisafunctionalofthe
jointprobabilitydistribution Psq .If s and q arethesametowithinthenoise,then S Q specifiestherelative
accuracy of the measurements in bits, i.e., how much information one measurement gives about a second
measurementofthesamevariable.
If S isadelayedimageof Q ,thenadelayphaseportraitgivestheestimatedjointdistribution Psq and I isa
statisticcalculatedontheportraitthatevaluateshowredundantthesecondaxisis.
New Method for First Arrivals
The calculation method of mutual information of the arbitrary A and B two channels on seismic record is
performedwiththesoftwareSmartSignalProcessingdevelopedbyProf.MingYueZHAIatNorthChinaElectric
PowerUniversity,andstepsareasfollows:
(1):Firstly,weshouldintercepttheseismicwavewhoselengthis L on A channel,takendown: {x (n)}n 01,...L 1 ,

L isthelengthofsequences.Thentherangeofsequence x isdividedintoequalpartswhich M isinterval,at


lastwecountthesamplingpointthatfallsintotheeveryintervalandcalculatetheprobabilitydistribution p ( x)
ofsequence {x ( n)}n 01,...L 1 .
(2):Secondly,

{x (n)}n 01,... L 1

whoselengthis L slidesonchannel B withacertainstep.Themutualinformation


{x (n)}n 01,... L 1
and
is taken down when it slides one time, and this mutual information is the one of the
{ y ( n d )}n 01... L 1 .Forthechannel B ,wewillfilltheinsufficientpartswithrandomsequences,andthenwecan
getthemutualinformation I (d ) .
(3):For, I (d ) , d 01,... .Themaximal ( d ) is I ( Dmax ) .Now,wethinkthisisthefirstarrivaltime.ThevalueofL
ismorethan 100 .ThevalueofMisabout 16 to 64 .
In the foresaid algorithm, the seismic mutual information of channel A and channel B need to be calculated.
Because calculation is very complicated, the paper improves the algorithm in order to reduce the amount of
calculation.
Throughthecalculationofmutualinformation,wegetthemaximalvalueofthemutualinformation.Thetimethat
itcorrespondsisfirstarrivaltimeofseismicwaves.Thismethodismoreaccuratethantheratioofenergymethod.
However, it needs to calculate mutual information from the first channel to the last channel, whose amount of
calculation is very huge. But the advantage is that mutual information is accurate and robust, because random
noisesareuncorrelatedwithsignalsofinterestingandthusthemutualinformationiszeroforsuchcase.
Wecanexplainitasameasureoftheamountofinformationonerandomvariablecontainsaboutanotherrandom
variable, thus it is the reduction in the uncertainty of one random variable due to the knowledge of the other.
Mutual information is not an invariant measure between random variables because it contains the marginal
entropies.NormalizedMutualInformationisabettermeasureofthepredictionthatonevariablecandoabout
theother
Applications to Simulated Signals
Inthesection,weappliedtheproposedmethodtoaseismicprofile,whichisgeneratedbythetimedelayedline
models (TDL). The wavelet used in TDL model is Ricker wavelet, with center frequency 10 Hz . The simulated
data without noise is illustrated in Fig. 1, and the noisy version of the simulated data is plotted in Fig. 2. As an
example,Fig.3illustratedtheresultsoftheproposedmethod.Fromthesimulations,wecanseethattheproposed
methodcanobtaintheprecisefirstarrivalsevenunderverylowSNR.

40

InformationEngineering(IE)Volume4,2015 www.seipub.org/ie

Fromabovementionedfigureswecanknowthat,whenSNRisveryhigh,theproposedmethodscanpickupthe
firstarrivalverywell.WiththereductionofSNR,errorofratioofenergymethodchangeslarger,andalotofbad
points appearance. We cannot accurately pick up the first arrival. When SNR is 5dB , the method based on the
mutual information can calculate the mutual information from the first channel to the end channel. At the same
times, it only calculates a part of mutual information. For calculating the part of mutual information, amount of
calculation of mutual information are the same. To evaluate the performances of the proposed method under
differentSNRs,weappliedittothesimulateddatawithdifferentSNRs.TheresultsareillustratedinFig.4.From
suchfigure,wecanseethattheerrorsareverylow.
Conclusions
Firstarrivalspickupisaveryimportantissueinseismicdataprocessing.Inthepaper,weproposedanewmethod
withthehelpofthemutualinformation.Insuchmethod,thereisnoanyassumptionforthedata.Therefore,the
proposedmethodcanbeappliedtoanyseismicdata.Especially,thementionedmethodcanworkverywellunder
verylowSNRsenvironments,becausethemutualinformationbetweendataandnoiseiszero,andthusnoisescan
affectarrivalspickupaslessaspossible.
simulated seismic data without noise in seismic profile
0
0.5
1
1.5

time [s]

2
2.5
3
3.5
4
4.5
5

2000 1500 1000 500


0
500
distance [m]

1000

1500

2000

FIGURE1.THESIMULATEDSEISMICPROFILEWITHTDLMODELWITHOUTNOISE
processed data with SNR= 0
0
0.5
1
1.5

time [s]

2
2.5
3
3.5
4
4.5
5

2000 1500 1000 500


0
500
distance [m]

1000

1500

2000

FIGURE2.THESIMULATEDSEISMICPROFILEWITHNOISE

41

www.seipub.org/ie InformationEngineering(IE)Volume4,2015

Results of different SNR

1
the time/s

2
3

2
3

5
-4000 -2000
0
2000 4000
the trace position
Results of different SNR
0

5
-4000 -2000
0
2000 4000
the trace position
Results of different SNR
0

1
the time/s

the time/s

the time/s

Theory value
0

2
3
4

2
3
4

5
-4000 -2000
0
2000
the trace position

4000

5
-4000 -2000
0
2000
the trace position

4000

FIGURE3.ANEXAMPLEOFFIRSTARRIVALPICKUPWITHTHEPROPOSEDMETHOD
first arrivals pickup errors with the proposed method
0.01
0.009
0.008
0.007

errors

0.006
0.005
0.004
0.003
0.002
0.001
0

10

15
snr[dB]

20

25

30

FIGURE4.ERRORSFORTHEFIRSTARRIVALPICKUPWITHTHEPROPOSEDMETHOD
REFERENCES

[1] S.Assous,P.Elkington,S.Clark,andJ.Whetton.Automateddetectionofplanargeologicfeaturesinborehole
images.GEOPHYSICS,79(1):D11D19,2014.
[2] T.Keho and W.Zhu.Revisiting automatic first arrival picking for large 3D land surveys, chapter 642, pages
31983202.2009.
[3] S.Li, Vladimirsky, A., and S.Fomel. Firstbreak traveltime tomography with the doublesquareroot eikonal
equation.GEOPHYSICS,78(6):U89U101,2013.
[4] J.Sabbione and Velis D. Automatic firstbreaks picking: New strategies and algorithms. GEOPHYSICS,
75(4):V67V76,2010.
[5] MingYue Zhai. Seismic data denoising based on the fractional fourier transformation. JOURNAL OF

42

InformationEngineering(IE)Volume4,2015 www.seipub.org/ie

APPLIEDGEOPHYSICS,109:6270,OCT2014.
[6] MingYue Zhai, Heidi Kuzma, and JamesW. Rector. A new fractalinterpolation algorithm for seismic data
basedoniteratedfunctionsystems.EPL,92(4),NOV2010.
[7] MingYue Zhai, Juan Luis FernandezMartinez, and JamesW. Rector. A new fractal interpolation algorithm
and its applications to selfaffine signal reconstruction. FRACTALSCOMPLEX GEOMETRY PATTERNS
ANDSCALINGINNATUREANDSOCIETY,19(3):355365,SEP2011.

43