You are on page 1of 9

CHINESE JOURNAL OF GEOPHYSICS Vol.54, No.

3, 2011, pp: 375383

THE UNCERTAINTY ANALYSIS OF THE KEY FACTORS THAT AFFECT THE AVO ATTRIBUTES IN SANDSTONE RESERVOIR
HE Tao1 , SHI Ge1 , ZOU Chang-Chun2 , REN Ke-Ying3 , PEI Fa-Gen2
1 Key Laboratory of Orogenic Belts and Crustal Evolution (School of Earth and Space Sciences, Peking University), Ministry of Education, Beijing 100871, China 2 Key Laboratory of Geo-detection (China University of Geosciences, Beijing), Ministry of Education, Beijing 100083, China 3 Zhanjiang Branch of China National Oshore Oil Corporation Ltd., Guangdong Zhanjiang 524057, China

Abstract Traditional seismic AVO forward study usually uses constant parameters to construct rock physics models, and such parameters in real formations have uncertainties in an exploration area. This work employed laboratory measurements of cores from the target formation to simplify the rock physics model with relationships of reservoir sandstone porosities and P- and S-wave impedances under a dry condition. To study inuences of model parameter uncertainties, the probability density functions of key model parameters were introduced based on core measurements and well logging analysis. Using the Monte-Carlo method and Gassmann uid replacement technique, the AVO responses of models were obtained for saturation uids as brine, oil and gas. The comparative analysis indicates that the velocity uncertainty of covering mud is the primary factor aecting the uid clusters distribution and deviation from water background trend on AVO intercept-gradient crossplots, while the porosity uncertainty of sand is also signicantly aecting but less than covering mud. Thus, the correct interpretation of AVO anomalies for real seismic data requires forward simulation on a probabilistic rock physics model to acquire the information on reservoir physical properties and uids as much as possible. Key words Sandstone reservoir, AVO crossplot, Uncertainty, Probability density function, Rock physics model

1 INTRODUCTION Based on simplied Zoeppritz equations, prestack seismic AVO technique has been widely used in petroleum exploration to detect hydrocarbon-induced anomalies in terms of the relationships between AVO attributes[1,2] . Among them AVO intercept-gradient crossplot is a basic form to directly show the characteristics for dierent combinations of rocks and saturation uids[3,4] . However, the acoustic attribute of in-situ rocks is a complex function of matrix properties, pore structure, clay content, uid property, pressure, temperature, and heterogeneity[5] , resulting in non-unique interpretations for AVO intercept-gradient crossplot[6] . Thus, the rock physics model is commonly used to predict the AVO responses of target formation for dierent saturation uids (brine, oil, or gas) with prior information[7] , and then corresponding rock physics templates[8] are made to reduce the multiplicity of solutions in interpretation of real seismic data. Parameters of traditional AVO forward modeling generally uses values from laboratory core measurements and well logging averages[7,9] , i.e., the rock physics parameters are xed. However, cores are discrete samples and the possible dierences should be considered between coring and non-coring parts even for the same drilling section when applying the accurate core laboratory measurements to other depths. On the other hand, sand reservoir normally has a covering layer of mud/shale, of which the information of density and velocity is in practice extracted from loggings since these samples are hard to core and easily damaged during laboratory measurements, and the associated uncertainties with large change in well loggings often arise due to well wall collapse at mud section. Also, well loggings usually are only laterally valid within the small area around the drilling hole (although they are vertically continuous), especially in the case of thin sandstone layers of continental sedimentation, where the clay contents generally vary rapidly transversely, resulting in small density
E-mail: taohe@pku.edu.cn

376

Chinese J. Geophys.

Vol.54, No.3

change but large P- and S-wave velocity changes[10] . When extrapolating the rock physics parameters from drill holes to a large seismic exploration volume, therefore, the composite eect should be considered for combining uncertainties of density, porosity, oil saturation ratio, layer thickness and so forth. In recent years many researchers have made great eorts addressing the uncertainties of rock physics model parameters. The Monte-Carlo method and Gassmann uid replacement technique[11] were used to obtain the AVO intercept-gradient distribution for uids of oil, gas and brine. Since the brine cluster constructed the background trend across the origin of crossplot, the distance of AVO point away from the background trend may reect the type and the content of occurring hydrocarbon[12,13] . Furthermore, the probability of dierent reservoir uids could be quantitatively determined using Bayes theory[14,15] . However, only few detailed studies of relationship between the brine background trend and the parameter uncertainties of the rock physics model are reported. In this work, 9 cores from W3 formation in WXS depression of the South China Sea were measured under the laboratory simulation of formation condition. Based on these measurements we constructed the rock physics model with reservoir sandstone porosity, which is the key factor between density and velocity of samples, and then used the Monte-Carlo method associated with probability density functions of model parameters to study the inuences of model parameter uncertainties on AVO intercept-gradient distribution and the brine background trend. The concluded rules of this study may be helpful to guide AVO analysis and interpretation on real seismic data set. 2 MEASUREMENT OF WAVE VELOCITY IN RESERVOIR SANDSTONE AND FLUID REPLACEMENT

In the W3 formation of WXS depression, the pore uid of reservoir sandstone is the mixture of brine (26 g/L) and light oil (0.7 g/cm3 ). Since the lack of S-wave acoustic logging in the study area and unknown oil saturation ratio for the measured P-wave acoustic logging, we built the empirical relationships of reservoir sandstone samples physical properties and P- and S-wave impedances by measurements under a dry condition[16,17] , and then applied Gassmann equation[18] to calculate model velocities for dierent oil saturation ratios. There were 3 advantages in this method: (1) The relationship of porosity and P- and S-wave impedances is valid at any interpolated value. (2) The AVO response of arbitrary oil/brine ratio could be modeled, which can not be simulated in laboratory. And (3) it is well known that P- and S-wave velocities of uid-saturated rocks are frequency dependent, or dispersion[19] . The same sample may have dierent velocity values between dierent measurement bands, say surface seismic frequencies (< 200 Hz), logging acoustic frequencies ( 104 Hz), and laboratory ultrasonic frequencies (105 106 Hz). Although it has not been rigorously tested for all rocks, many observations found that the dierence between the ultrasonic and low-frequency velocities of dry sandstones was normally within the typical experimental accuracy of about 1 percent[20] . Thus, the uid replacement based on the dry measurements could approximately satisfy the zero-frequency assumption of Gassmann theory. Figure 1a and 1b show very good linear relationships of density vs. porosity and P- and S-wave impedances vs. density with following equations dry = 0.038 + 2.763 (R2 = 0.9754), Zpdry =dry Vpdry = 8584dry 9690 (R2 = 0.9993), Zsdry =dry Vsdry = 6544dry 6501 (R = 0.9918),
2

(1) (2) (3)

where subscript dry denotes dry condition, density is in g/cm3 , and velocity V is in m/s. Thus, for any arbitrary porosity within the core sample range of 12% to 25%, the dry density dry and velocities Vpdry and Vsdry of W3 formation sandstone could be well estimated from Eqs.(13). Then the Gassmann uid replacement technique was used to calculate dierent uid saturated sandstone velocities at given model porosity . The basic form of Gassmann equations[18] are

He T et al.: The Uncertainty Analysis of the Key Factors that Aect the AVO Attributes in Sandstone

377

Fig. 1 Relationships between physical parameters of reservoir sandstone core samples in dry condition and comparisons of laboratory measurements and uid replacement results
High linear relationships of (a) porosity and density and (b) density and P- and S-wave impedances. Close P- and S-wave velocity values of uid replacement results and laboratory measurements for (c) brine and (d) oil saturated samples.

Ksat Kdry Kf = + , Ks Ksat Ks Kdry (Ks Kf ) sat =dry ,

(4) (5)

where Ks , Kf , Kdry , Ksat , dry and sat are rock matrix bulk modulus, pore uid bulk modulus, dry rock bulk modulus, uid-saturated rock bulk modulus, dry rock shear modulus, and uid-saturated rock shear modulus, respectively. Based on the thin sample section under the optical microscope, the rock matrix bulk modulus Ks was estimated using Hashin-Shtrikman equation[21] according to the mineral (mainly quartz, feldspar and clay) contents. Thus, given the model porosity , Kdry and dry can be calculated from dry , Vpdry and Vsdry 4 2 2 Kdry =dry Vpdry Vsdry , 3
2 dry =dry Vsdry .

(6) (7)

378

Chinese J. Geophys.

Vol.54, No.3

Dening the hydrocarbon (oil/gas) saturation of model as Shc , the density of hydrocarbon-brine mixture f and uid-saturated rock sat can be calculated by f =Shc oil + (1 Shc )brine , sat =dry + f , while the bulk modulus of the uid Kf is given by Reuss equation Shc 1 Shc 1 = + . Kf Khc Kbrine Finally the uid-saturated rock velocities can be calculated using above results Vpsat = 4 Ksat + sat 3 sat , (11) (12) (10) (8) (9)

Vssat = sat /sat .

Therefore, the approximate velocities of W3 formation sandstone with pore uid of dierent hydrocarbon/brine ratios can be determined by porosity , which largely simplies the rock physics model construction process. Fig. 1c and 1d compare P- and S-wave velocities of model calculations and laboratory measurements for these samples saturated by brine and light oil, respectively, showing that the calculations matched the measurements quite well with the errors of P-wave velocities <1.5% and S-wave <2%. This validates that the above method could obtain the uid-saturated velocities close to the W3 formation condition, and thus satised the sandstone reservoir models of dierent porosities and mixed uid. 3 PROBABILITY DENSITY FUNCTIONS OF MODEL PARAMETERS A simple mud-sandstone-mud rock physics model was constructed for demonstration purpose. The probability density functions of key model parameters were determined from the integration of regional geology, core sample measurements and well loggings (Table 1). The main characteristic indicators of covering mud are density and P- and S-wave velocities. Since the mud samples of W3 formation were damaged during coring and laboratory measurements, the well logging was used to obtain the brine-saturated parameters. Base on the loggings from 7 drill holes, the density and P-wave velocity of covering mud were normal distribution and thus could be characterized by an average value and a varying range of standard deviation (Table 1). In the study area there is only one well having S-wave logging, which shows the highly linear relation of S-wave velocity and P-wave velocity mud Vsmud = 0.85Vp 1016 (R2 = 0.895). (13) Compared to the real S-wave velocity, many pervious studies also indicate that the calculated value with such empirical equation generally has a random error within 5%. Thus, the S-wave velocity of covering mud in the model is not an independent variable, but the total value of calculation result (Eq.(13)) plus a 5% white noise. For reservoir sandstones the main characteristic indicators could be further divided into three dierent types: matrix properties (density, bulk modulus and porosity), reservoir layer thickness and pore uid. The observations of the thin sections indicate that the main mineral content (quartz, feldspar, detritus, etc.) was statistically in normal distribution, and thus the matrix density and bulk modulus calculated from grain percentage were also normal distributions (Table 1). The sandstone porosities in the limited area were approximately normal distribution based on integrated analysis of core measurements and well logging interpretations: the standard deviation was almost a constant value of 0.025, but the average value ranged from 0.1 to 0.25 for dierent locations. The reservoir sandstone thickness was a uniform distribution between 0 to maximum 25 m for real condition. Similarly the density and bulk modulus of mixed uid were uniform distributions for hydrocarbon saturation ratios varying from 0 to 1.

He T et al.: The Uncertainty Analysis of the Key Factors that Aect the AVO Attributes in Sandstone
Table 1 Probability density functions of rock physics model parameters
Model Covering mud No. 1 2 3 4 Reservoir sandstone 5 6 7 8 Parameter Density mud
mud Vp Vsmud

379

Probability density function Normal distribution Normal distribution Calculated from Eq.(13) Normal distribution Normal distribution Normal distribution Uniform distribution Uniform distribution

Average 2.42 g/cm3 3166.4 m/s 2.65 g/cm3

Varying range Standard deviation 0.08 g/cm3 Standard deviation 300 m/s Add white noise 5% Standard deviation 0.02 g/cm3 Standard deviation 2 GPa Standard deviation 0.025 025 m 01

P-wave velocity S-wave velocity

Matrix density s Matrix bulk modulus Ks Porosity Thickness d Hydrocarbon saturation ratio Shc

40 GPa 0.1 0.25 12.5 m 0.5

4 RANDOM FORWARD MODELING BY THE MONTE-CARLO METHOD Once the probability density functions of model parameters were determined as listed in Table 1, the MonteCarlo method was used to produce a lot of random models satisfying these prior constraints. For any single model since the parameter values are known, the theoretical AVO intercept and gradient of synthetic prestack seismic gather can be calculated using the statistical wavelet of real seismic data and simplied Zoeppritz equations. Amount of random model results altogether composed a simulation ensemble on AVO interceptgradient crossplot, fully showing the eect of model parameters uncertainties. Since in the model velocities of reservoir sandstone were determined by porosity, the average porosity values of 0.1, 0.15, 0.2 and 0.25 were selected to calculate the typical model AVO responses within the porosity range of W3 formation sandstone. Fig. (2a2d) was a conventional AVO forward model result, which used xed mud and sandstone parameters, i.e., the average values of parameters 16 in Table 1 to analyze AVO characteristics at the top sandstone interface of the model when the reservoir thickness (parameter 7) and hydrocarbon (oil/gas) saturation ratio (parameter 8) changed. This implies that the deviations of rock properties from their averages are small enough within the exploration scope and can be ignored. At this conguration the AVO ensembles of dierent model porosities were mainly in clusters of radial shape from the origin of intercept-gradient crossplot. With the increasing of porosity from 10% to 25%, the corresponding AVO clusters rotated clockwise with intercepts changing from positive to negative values while gradients remain to be negative. Thus, the AVO anomaly transferred from type I to type III. For the same porosity model, clusters of oil (green) and gas (red) were in fan shape with very small overlapping areas, while the concentrated brine cluster (blue) became the linear background trend. Thus, such AVO intercept-gradient crossplots idealistically can not only indicate the approximate reservoir sandstone porosity, but also present denite uid information. However, the AVO clusters of the real prestack seismic processing result are rarely seen such fan shape radiating from the origin, but normally present the strap-like clouds extending along quadrant II-IV direction of AVO intercept-gradient crossplots as shown on Fig. (2e2h), corresponding to the random models fully considering the uncertainties of all 8 parameters according to their probability density functions listed in Table 1. These plots show that the associated uid type of AVO straps transferred from gas to oil to brine in sequence from bottom left to the origin. Whats more, the straps got wider as porosity increasing. The brine clusters in blue color are still the background trend, extending across the origin or nearby, and the AVO responses departing from this trend become the anomaly and may indicate the hydrocarbon accumulation. Thus, the departing or overlapping degree of AVO clusters for dierent uids is determined by the probability distribution of the rock physics model parameters; the more depart the clusters, more easily distinguish the hydrocarbon. 5 INFLUENCES OF MODEL PARAMETER UNCERTAINTIES Since the parameters 7 and 8 used in the probabilistic model (Fig. (2e2h)) were the same as the conventional model (Fig. (2a2d)), the signicant dierence of AVO characteristics between the two models were

380

Chinese J. Geophys.

Vol.54, No.3

Fig. 2 Comparisons of AVO responses at top sand interface of the mud-sand-mud model for dierent sandstone porosities
(a)(d) only consider the variations of reservoir thickness and hydrocarbon (oil/gas) ratio in the uid, while (e)(h) consider variations for all model parameters listed in Table 1.

apparently caused by changing parameters 16 from xed values to probability density functions. Using the probabilistic model with average porosity of 25% as an example, the dierent inuences of parameters 16 on AVO responses were further analyzed. The changes of density and bulk modulus of sandstone matrix were tested for normal distributions (Table 1) while keeping other parameter conguration as the conventional model. For the standard deviation of these two parameters as large as 5% of the average values, the corresponding AVO results in Fig. 3a had the uid cluster patterns very similar to those in Fig. 2d; the fan shapes only slightly expanded and overlapped each other. This indicates that the matrix grain contents of the reservoir sandstones were not the key factors that aected AVO response of W3 formation. Using the same way the eect of porosity uncertainty was tested also for normal distribution with average value of 0.25 and standard deviation of 0.025, and the results are shown in Fig. 3b. In this case the standard deviation was as large as 10% of the average value, and the AVO clusters of dierent uids spread signicantly, resulting in the transition from the fan shapes of Fig. 3a to the strap-like trend along quadrant II-IV direction of Fig. 3b. Note that the AVO intercept-gradient points were still concentrated within quadrant III in this situation. For the uncertainties of covering mud, Figs. 3c and 3d show the AVO results of normal distribution for

He T et al.: The Uncertainty Analysis of the Key Factors that Aect the AVO Attributes in Sandstone

381

mud density (Table 1) and adding 5% white noise to S-wave velocity, respectively. The uid clusters show similar degrees of spreading for these two parameters uncertainties. Note that the spreading degree of the AVO clusters for these two parameters was larger than the eect of sandstone matrix (Fig. 3a) but much less than the one of porosity (Fig. 3b). Considering the P-wave velocity of covering mud was normal distribution, Figs. 3e and 3f show the AVO responses for standard deviation approximately as 5% (150 m/s) and 10% (300 m/s) of average P-wave velocity (3166.4 m/s), respectively. Compared to Fig. (3a3d), the uid clusters changed shape signicantly and spread much along quadrant II-IV direction. Since the well loggings indicated that the covering mud velocity at target formation generally varied within the 10% of average values, Fig. 3f was close to the real eld case and showed that the AVO responses of gas, oil and brine all spread out of quadrant III of crossplot and became three parallel straps in similar length extending along quadrant II-IV direction. Comparing Fig. 2h, the case of total eect of all 8 parameters in their probability density function forms, with the plots of Fig. 3, it is clear that Fig. 3f resembles Fig. 2h very much. Thus, the P-wave velocity uncertainty of covering mud (Parameter 2) dominated the AVO intercept-gradient distribution of W3 formation and surpassed the integrated eect of all other parameters in Table 1. In the situation of stable covering mud with velocity changing less than 2% (e.g., marine sediment), the porosity uncertainty of sandstone could become the primary factor that aects the AVO cluster shapes of dierent uids (refer to Fig. 3b).

Fig. 3 The eect of single parameters probability density function on AVO response for model with average sandstone porosity of 25%
(a) Reservoir sandstone has normal distribution for both matrix density and bulk modulus. (b) Reservoir sandstone has normal distribution for porosity. (c) Covering mud has normal distribution for density. (d) Covering muds S-wave velocity is added with 5% white noise. (e) Covering mud has normal distribution for P-wave velocity with standard deviation of 150 m/s, or 5% of average velocity. (f) Covering mud has normal distribution for P-wave velocity with standard deviation of 300 m/s, or 10% of average velocity.

382 6 CONCLUSIONS

Chinese J. Geophys.

Vol.54, No.3

Using the laboratory measurements of core samples, rock physics model construction was simplied based on highly linear relationship of P- and S-wave impedances under the dry condition and reservoir sandstone porosities of target formation. After introducing appropriate probability density functions into model parameters, the Monte-Carlo method and Gassmann uid replacement technique were used to analyze the uncertainty eect of key model parameters on AVO responses for saturation uids such as brine, oil and gas. The AVO clusters of brine-saturated models pass through the origin (or nearby) on intercept-gradient crossplot and become the background trend; the distances of oil or gas points apart from the background trend represent uid discrimination ability of seismic AVO. Under the condition of stable sedimentation, covering mud and reservoir sandstone have little content change laterally, and the corresponding AVO distributions of dierent uids are the fan-like shape emitting from the crossplot origin; the locations and extending characteristics of these clusters can indicate the magnitude of the reservoir porosity and type of saturation uid. When the covering mud is not stable as the case of W3 formation in this study, however, the velocity uncertainty of mud is the primary factor aecting the AVO distributions and covers the inuences from physical properties of reservoir sandstone (e.g., porosity, matrix properties), resulting in the AVO clusters of dierent uids spreading in parallel straps along quadrant II-IV of crossplot. Thus, the uncertainties of model parameters should be fully considered for the changes of AVO responses when applying rock physics modeling results of discrete, nite areas to the enlarged scale of seismic exploration range. The correct interpretation of AVO anomalies for reservoir and uids need comparison of results from both the probabilistic rock physics model and real seismic data. ACKNOWLEDGMENTS This work was sponsored by the National Natural Science Foundation of China (40830423, 40904029), the Scientic Research Foundation for the Returned Overseas Chinese Scholars, State Education Ministry, and CNOOC Zhanjiang Research Project (Z2008SLZJ-FN0158). REFERENCES
[1] Zheng X D. Approximation of Zoeppritz equation and its application. Oil Geophysical Prospecting (in Chinese), 1991, 26(2): 129144 [2] Hampson D. AVO inversion, theory and practice. The Leading Edge, 1991, 10(6): 3942 [3] Castagna J P, Swan H W. Principles of AVO crossplotting. The Leading Edge, 1997, 16(4): 337344 [4] Sams M. Yet another perspective on AVO crossplotting. The Leading Edge, 1998, 17(7): 911917 [5] Wang Z Z. Fundamentals of seismic rock physics. Geophysics, 2001, 66(2): 398412 [6] Drufuca G, Mazzotti A. Ambiguities in AVO inversion of reections from a gas-sand. Geophysics, 1995, 60(1): 134141 [7] Ma Z G, Guan L P, He Z H, et al. Reservoir characterization using seismic attributes optimized with rock physics modeling. Acta Petrolei Sinica (in Chinese), 2003, 24(6): 3539 [8] Chi X, Han D. Lithology and uid dierentiation using a rock physics template. The Leading Edge, 2009, 28(1): 6065 [9] Huang K, Xu Q Z, Yang X H, et al. Oil, gas and water distribution detection forward modeling with elastic parameters of core samples under simulated condition. Chinese J. Geophys. (in Chinese), 1998, 41(S1): 414421 [10] Shi G, Yang D Q. The regression analysis study on velocity and porosity, and clay content of rocks. Acta Scicentiarum Naturalum Universitis Pekinesis (in Chinese), 2001, 37(3): 379384 [11] Russell B H, Hedlin K, Hilterman F J, et al. Fluid-property discrimination with AVO: A Biot-Gassmann perspective. Geophysics, 2003, 68(1): 2939 [12] Simm R, White R, Uden R. The anatomy of AVO crossplots. The Leading Edge, 2000, 19(2): 150155

He T et al.: The Uncertainty Analysis of the Key Factors that Aect the AVO Attributes in Sandstone

383

[13] Wang W H, Jiang Z X, Pan R F. AVO crossplot analysis and its application. Journal of Xian Petroleum Institute (in Chinese), 2003, 18(2): 58 [14] Malinverno A, Briggs V A. Expanded uncertainty quantication in inverse problems: Hierarchical Bayes and empirical Bayes. Geophysics, 2004, 69(4): 10051016 [15] Gao J R, Teng J W, Li M, et al. AVO uid inversion: theory and practice. Petroleum Exploration and Development (in Chinese), 2006, 33(5): 558561 [16] Pei F G, Zou C C, He T, et al. Fluid sensitivity study of elastic parameters in low-medium porosity and permeability reservoir rocks. Applied Geophysics, 2010, 7(1): 19 [17] He T, Zou C C, Pei F G, et al. Laboratory study of uid viscosity induced ultrasonic velocity dispersion in reservoir sandstones. Applied Geophysics, 2010, 7(2): 114126 [18] Gassmann F. Uber die Elastizitt porser Medien. Veirteljahrsschrift der Naturforschenden Gesellschaft in Zrich, a o u 1951, 96: 123 [19] Wuenschel P C. Dispersive body waves-an experimental study. Geophysics, 1965, 30(4): 539551 [20] Winkler K W. Estimates of velocity dispersion between seismic and ultrasonic frequencies. Geophysics, 1986, 51(1): 183189 [21] Hashin Z. Analysis of composite materials-A survey. Journal of Applied Mechanics, 1983, 50(3): 481505

You might also like