What is the Best Seismic Attribute for Quantitative Seismic Reservoir Characterization?Dennis Cooke*1, Arcangelo Sena 2, Greg ODonnell 3 , Tetang Muryanto 4 and Vaughn Ball 4 . 1 ARCO Alaska, 2ARCO Exploration Technology, 3 ARCO Indonesia, 4 Matador Petroleum formerly ARCO ExplorationTechnologySummary phase data, relative impedance data or absolute impedance data. In our opinion, these three attributes (zero phaseIt is possible to generate at least 30 different seismic amplitude, relative impedance and absolute impedance) areattributes from a given seismic data set. This presentation the most useful for quantitative reservoir characterization.addresses the question of which of those post-stackattributes is most appropriate to use for a quantitative 3)Interval attributes are those that are used to quantify aseismic reservoir characterization. Our conclusion is that window of seismic data usually containing more than onean absolute impedance inversion is the best attribute in peak or through. Most seismic attributes fall into thistheory, but, in practice, a relative impedance inversion is category. Examples of interval attributes are number ofmuch more practical. zero crossings, average energy and dominant frequency. These attributes are frequently used when a reservoirsIntroduction seismic reflection(s) are so discontinuous that it is impossible pick the same peak or trough on all traces. AnReservoir characterization is the process of mapping a interval attribute is analogous to a well log cross sectionreservoirs thickness, net-to-gross ratio, pore fluid, porosity, with a number of thin, discontinuous sands that can not bepermeability and water saturation. Traditionally, this has correlated with any certainty. For this reservoir, a net-to-been done in a field development environment using data gross sand ratio map is made instead of individual sandfrom well logs. Within the past few years, it has become (flow) unit thickness maps. A seismic reservoirpossible to make some of these maps using seismic characterization is always improved if all peaks and troughsattributes when those attributes are calibrated with over the reservoir interval can be picked individually andavailable well control. The advantage of using wells and thus have quantitative attributes extracted. If this is notseismic instead of just wells alone, is that the seismic data possible, the use of interval attributes is warranted.can be used to interpolate and extrapolate between andbeyond sparse well control. 4)AVO attributes are those that are generated using a reflections pre-stack amplitudes. Examples of pre-stackThere is a multitude of different seismic attributes that can attributes are AVO gradient, AVO intercept, nearbe generated from a given seismic data set. A quick review amplitude and far amplitude. 3D pre-stack attributes haveof one popular seismic interpretation package shows that only become available recently with the advent ofone can generate at least 30 different seismic attributes affordable pre-stack time migrations. Pre-stack attributesfrom an input seismic survey. Some of these attributes are have a lot of promise, but are beyond the scope of thismuch better than others for reservoir characterization, but presentation.there has not been much discussion of this in thegeophysical literature. The objective of this presentation is This talk will focus on the three main quantitative attributesto try to classify seismic attributes and show which ones (zero phase, relative impedance and absolute impedance)work best for reservoir characterization. and address their respective advantages and disadvantages.One way to organize and understand seismic attributes is to Zero Phase Amplitudesseparate them into the following four categories: All seismic attributes are calculated from the final migrated1)Qualitative attributes such as coherency - and perhaps zero phase dataset (or what is believed to be zero phase).instantaneous phase or instantaneous frequency - are very Clearly, the easiest, fastest, least expensive attribute is thegood for highlighting spatial patterns such as faults or zero phase amplitude. The convolutional model and thefacies changes. It is difficult if not impossible to relate reflection coefficient formula show that a reflectors zerothese attributes directly to a logged reservoir property like phase amplitude can be directly related to the reservoirsporosity or thickness, and thus these attributes are not impedance. A thin-bed tuning curve model shows that zeronormally used to quantify reservoir properties. phase amplitude is also directly related to reservoir thickness. Additionally, gas substitution modeling shows2)Quantitative attributes: The simplest quantitative that a reservoirs zero phase amplitude can be influenced byattributes are the amplitude (of a peak or a trough) on zero changes in pore fluids. A solid theoretical conclusion is that
What is the Best Seismic Attribute for Reservoir Characterization?changes in zero phase amplitude are a function of changes oil sand. The cap rock impedance varies due to lateralin reservoir impedance, thickness and pore fluid. This lithology changes and because it is a waste rock andconclusion has been proven by many successful contains some oil and/or gas.quantitative reservoir characterizations done with zerophase amplitude.Absolute Impedance and its AdvantagesThe absolute impedance attribute can be generated witheither a Seislog ® type impedance inversion (one thatincludes a low frequency background model) or a model-based inversion such, as that first described in Cooke andSchneider (1983). There are two major motivations forusing absolute impedance for reservoir characterization:1)The amplitudes on an absolute impedance datasetdescribe the impedance of the rocks, where the amplitudeson a zero phase dataset describe the impedance contrastbetween rocks. Put another way, the impedance attribute isrelated to the geology while the zero phase attribute is Figure2: Reflection coefficient probability distribution.related to the derivative of the geology. The importance of Calculated using the impedances in Figure 1 and the formula:this difference can not be overstated for the case where the RC = (Z2-Z1)/(Z2+Z1) where Z1 and Z2 are the impedances of the cap and reservoir rock.impedance of both the reservoir and the surrounding rockare changing laterally. Consider Figure 1 which shows the 2)The second major motivation for using absolutedistribution of impedance for both cap rock and reservoir impedance instead of zero phase amplitude concerns therock (gas filled and oil filled) at Prudhoe Bay Field. These amplitude scale and format problem that occurs with zerodistributions can be input into the reflection coefficient phase data. Consider an undrilled gas prospect on one 3Dformula which leads to the reflection coefficient survey, with a second 3D survey that covers a nearby gasdistributions of Figure 2. Figure 1 corresponds to absolute discovery. With zero phase seismic data, the prospectsimpedance data and Figure 2 would correspond to zero amplitudes and the gas discoverys amplitudes can not bephase data (without a seismic wavelet). Clearly, the ability compared (unless a similar empirical scaling has beento discriminate between oil filled reservoir and gas filled applied to both). Furthermore, the gas discoverys logs canreservoir is enhanced in the absolute impedance case. not be compared the amplitudes on the zero phase seismic data. When both 3Ds are converted to absolute impedance, the seismic amplitudes can be compared to each other and to the impedance logs from the gas well. Disadvantages of Absolute Impedance Absolute impedance inversions can be very expensive in terms of both money and time delays. Frequencies in the inversion above the seismic bandpass will be non-unique. And since the input zero phase seismic data does not contain frequencies below the seismic bandpass (which are required for inversion), information at these frequencies must be supplied by the processor. The work that is done to prepare and constrain the low frequency portion of inversions can be very subjective and interpretive. Most often, this work on the low frequencies is not done by the interpreter, but by others who may not communicate to the Figure 1: Probability density functions for the acoustic impedance interpreter the subjective nature of the low frequencies.of Sadlerochit reservoir and Shublik cap rock at Prudhoe BayField. A good way to understand the problem with the lowThe data in Figure 1 are taken from a gas well and an oil frequencies in absolute impedance inversion is to considerwell. As expected, the gas sand has slower impedance that a hypothetical inversion between two wells as in Figures 3A and 3B. Wells A and B at structural highs have tight
What is the Best Seismic Attribute for Reservoir Characterization?and thin reservoir (marked in yellow). A prospectivelocation exists between the wells, but it is not clear if thereservoir there is better or worse than on the highs (and thisis why the inversion is being done). The inversion processrequires input of a low frequency (below seismicbandwidth) impedance for all traces. At wells A and B,this low frequency is taken from the well control. At allother locations, the processor must interpolate, interpret orguess at this low frequency input. At the proposedlocation, this low frequency guess could take the form of alinear interpolation between wells A and B (shown in blackin 3A). Alternatively, the low frequencies at this locationcould be modified to fit the structure of the reservoir (i.e.shifted down to tie the yellow horizon). Additionally, thelow frequency input could be modified to fit hypothetical Figure 3A. Hypothetical inversion example.depositional models. Two possible depositional models:Depositional Model 1): The package of sediments thatsurrounds the reservoir it is a predominantly fluvial system.This implies that locations A and B would havepreferentially received thin, shaley over-bank deposits andthe proposed location would have received more sand.Assuming that sands have a slower velocity than the shaleshere, this depositional model implies that the proposedlocation needs a low frequency input that is lower than thatfound at wells A and B. This models low frequency inputis shown in blue in Figure 3B.Depositional Model 2): This is a predominantly shallowwater marine system and the package of sediments at theproposed location have more shale than at A and B. Again,if the sands are slower than the shales, the proposedlocation would needs a low frequency input that is fasterthan found at wells A and B. This low frequency input isshown in red in Figure 3B. Figure 3B: Three different low frequency impedance trends for theEach of these three different low frequency models are just proposed location in Figure 3A.as correct as the others. And, if their frequency content isbelow the seismic bandwidth, three separate inversions There are numerous ways to calculate a relative impedanceusing them would lead to three significantly different inversion from the zero phase dataset. Perhaps the simplestresults for the full bandwidth absolute inversion. Since method is based on Lindseth (1979) who rewrites theinclusion of the low frequencies can lead to such confusion, reflection coefficient formula to express impedance as theperhaps the best approach is to not include them at all. integral - or running sum - of the reflection coefficients.This leads to an inversion that is restricted to the bandwidth This running sum can also be expressed as a convolutionalof the input seismic - also called a relative impedance filter where the phase spectrum is a 90 degree rotation andinversion. the amplitude spectrum has a -6dB/octave filter. One very easy way to generate an relative impedance dataset is to useRelative Impedance Inversion this 90 degree phase rotation filter.The high cost and uncertain nature of absolute impedance There are two advantages to absolute inversion listedinversions are the result of including the low frequencies in earlier: 1) geology vs. derivative of geology and 2) thethat inversion. If the low frequencies are not used, these scale problem of zero phase dataset. The relativeproblems go away, but the absolute impedance inversion impedance dataset does just as good of a job as the absolutebecomes a relative impedance inversion. impedance on the first problem. However, on first inspection, the relative impedance inversion appears to have the same scale problem as the zero phase dataset it
What is the Best Seismic Attribute for Reservoir Characterization?was generated from. This implies that relative impedances it has drawbacks related to its low frequency content. If thefrom different 3D surveys and from well data can not be low frequencies are removed, the result is a relativequantitatively compared. impedance dataset, which is in practice the best seismic attribute.There are two ways one can address the scale problemassociated with relative impedance. The first way only Acknowledgementsscales a reservoirs relative impedance map and not theentire relative impedance dataset. When doing any The authors would like to thank ARCO Alaska, ARCOreservoir characterization project where a number of wells Exploration Technology and Operations, and ARCOare available, the reservoirs impedance at the well Indonesia for permission to publish this work. Thelocations should always be cross plotted against thereservoirs well log properties. This cross-plotting step interpretations and conclusions discussed in this paper areindicates whether or not the impedances are related to the those of the authors and do not necessarily represent thosewell log properties, and if they are, the cross plot supplies of the Prudhoe Bay Unit Working Interest Owners.the information needed to calibrate relative impedance andremove its scale problem. For example, if a cross plot Referencesbetween a reservoirs relative impedance and reservoirporosity-feet shows a linear trend of the sort: Cooke, D.A. and Schneider W.A., Generalized Linear porosity-feet = A*(relative impedance) + B Inversion of Reflection Seismic Data, Geophysics, Vol. 48,then a map of the reservoirs relative impedance can be No. 6 (June 1983) P. 665-676transformed into porosity-feet by multiplying by A andadding B. This solves the relative impedance scale Cooke D.A. and Muryanto, T., Reservoir Quantification ofproblem. B Field, Java Sea via Statistical and Theoretical Methods, Submitted for presentation at the 1999 SEG InternationalNote that for an absolute impedance dataset, the inversion Exposition and Meeting, Houston, TX USAstep incorporates the low frequency information and scalesthe input data to absolute impedance, but it is then rescaled Lindseth, R. , 1979 Synthetic Sonic Logs - A process forto porosity-feet with the cross-plotting. The first scale step Straigraphic Interpretation: Geophysics, 44, 3-26.for absolute impedance dataset is thus redundant.The second method to scale a relative impedance dataset isused when there are not a sufficient wells to make a crossplot and/or the cross plot does not give a linear trend. Thismethod simply rescales the relative impedance data so thatits RMS amplitude for over a large user-defined depth andmap window is constant (usually = 1.0). This RMS rescaleis only valid if the earths impedance averaged over a largewindow is also constant. This scale process allowscomparison of amplitudes on the relative inversion withrelative impedance amplitudes from well models. Anexample of this is shown in figure 4 which comes fromCooke and Muryanto (1999). Another quantitative tool thatis available with this type of scaling is to apply it to all theseismic data over known oil and/or gas reservoirs for abasin. This allows one to build a database that can besorted by fluid type or reservoir or reservoir thickness.This database tool can be very useful for quantifyingexploration risk.ConclusionA quantitative seismic reservoir analysis needs to be doneusing a seismic dataset whose format allows easycomparison between well data and different seismicdatasets. This can be done with absolute impedance data,scaled impedance data or scaled zero phase data. The Figure 4. Tuning curves made from synthetic relative impedanceabsolute impedance data is theoretically the best option, but data scaled to match amplitudes with 3D survey.