Anda di halaman 1dari 5

AVO classification of lithology and pore fluids constrained by rock physics depth trends

PER AVSETH, HARALD FLESCHE, and AART-JAN VAN WIJNGAARDEN, Norsk Hydro Research Centre, Bergen, Norway

Elastic properties of rocks are strongly

influenced by local geologic trends and can change markedly even within a sedimentary basin. Critical geologic factors that control elastic properties are either related to depositional environment or burial history. Knowledge about the expected change in seismic response, as a function of depositional or compactional trends, will increase the ability to predict hydrocarbons, especially in areas with little or no well log information. In other words, by understanding the geologic constraints in an area of exploration, one can reduce the range of expected variability in rock properties and hence reduce the uncertainties in seismic reservoir prediction. Figure 1 depicts this problem, where we assume there is well log control only in the shallow interval on the shelf edge. Before extending the exploration into more deeply buried zones, or to more distal deepwater environments, it is important to understand the rock physics trends in the area. In this paper, we present a technique to calculate expected AVO (amplitude variation with offset) responses as a function of lithofacies, pore fluid, and burial depth. First, we calculate expected depth trends in rock physics properties for different lithologies and pore fluids. These trends are calculated from empirical porosity-depth models representing the local burial and compaction history. Next, we calculate the corresponding AVO depth trends from the depth trends in rock properties. Different models are generated based on the knowledge of local geology and depositional environment. AVO uncertainties are included and take into account the expected or observed natural variability in the rock properties. In this way one can obtain AVO probability density functions (pdfs) for any given depth of burial. Finally, the modeled AVO pdfs are used to predict the most likely lithology and pore fluids for different depth intervals from real seismic data. We apply this technique to an unconsolidated, mud-rich deepwater turbidite system offshore West Africa (Figure 2). Rock physics depth trends. In order to understand the expected seismic

Figure 1. Rock physics properties change with depositional environment and burial depth. These geologic trends must be taken into account during hydrocarbon reconnaissance of seismic data.

Figure 2. Seismic stack section intersecting a well penetrating a turbiditic gas and oil field, offshore West Africa. The geologists core description is superimposed showing the vertical sequences of sands and shales encountered in the well (red = poorly sorted conglomerates; orange = clean turbidite sands; yellow = shaly sands; black/white streaks = mixed, thin-bedded sands-shales and laminated shales (heterolithics); green= shale). Gas was encountered in the upper sandy interval, whereas oil was found in the middle sand interval. Brine was encountered in the lower sandy interval. The seismic data are zero-phase where red (with dark extremes) represents negative amplitudes and black represents positive amplitudes.

response of a siliciclastic reservoir, at any given depth, it is of key interest to know the expected contrast in elastic properties between shales and sands as a function of depth. However, rock physics and AVO depth trends can be very complicated depending on mineralogy, lithology, diagenesis, pore pressure, effective stress, and fluid properties. In areas with good well coverage, one can establish empirical rock physics depth trends for different lithologies from statistical regressions to well log data (VP, VS, and density). However, in this paper we want to stress the importance of modeling depth trends. Rock physics models allow for extrapolation of observed trends to depositional settings and depth ranges that are not covered by well log data. This is often the case in an early exploration stage. Furthermore,

modeled depth trends help us to better understand observed depth trends, and to detect anomalous zones that do not follow the expected depth trends, whether these are pressure anomalies, unexpected lithologies, or abrupt diagenetic events. In general, seismic velocities and densities of siliciclastic sedimentary rocks will increase with depth due to compaction and porosity reduction. Consequently, the very first step of the proposed AVO classification technique is to establish local porosity-depth trends for sands and shales. For many sedimentary basins, porosity-depth trends for siliciclastic sediments are shown to follow fairly simple and empirical functions, following exponential or linear forms (see appendix). Applying the empirical equation given in the appendix, porosity-depth trends

1004

THE LEADING EDGE

OCTOBER 2003
Downloaded 14 May 2011 to 203.199.41.231. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

Figure 3. Schematic illustration of porosity-depth trends for sands and shales. Both the sand and shale trends can vary significantly due to composition, texture, pore fluids, temperature, and pressure gradients. Hence, no attempt is done to assign absolute scales. However, there are a few rules of thumb: (1) The depositional porosity of shales is normally higher than that of sands. (2) The porosity gradient with depth is steeper for shales than for sands during mechanical compaction (i.e. at shallow depths). (3) The porosity gradient with depth will be steeper for sands than for shales during chemical compaction (i.e. quartz cementation of sands normally occurs at greater burial depth, beyond 2-3 km).

Figure 4. Predicted seismic depth trends based on empirical porosity trends, compared to well log data from the Tertiary deep-water turbidite field shown in Figure 2.

are calibrated using reasonable critical (depositional) porosity values at the surface, and inverted density logs for any burial depth. At deposition, shales tend to have relatively high porosities compared to sands. Sands will have depositional porosities of approximately 0.4, while shales can have depositional porosities of more than 0.8 (Figure 3). Shaly sands and heterolithics (i.e., mixed sands and shales) can have even lower depositional porosity than
1006 THE LEADING EDGE OCTOBER 2003

0.4, as clay particles will fill the pore space of the sand frame. During early burial, porosity is reduced mainly due to change of grain packing and ductile grain deformation. Shales tend to compact more easily than sands, causing a crossover of the porosity-depth trends of sands and shales. At greater depths, different diagenetic processes occur. Sands lose porosity mainly due to cementation, while bound water is released and intrinsic clay porosity is

reduced in shales. Secondary porosity may occur in sands due to dissolution of mineral grains. Hence, porositydepth trends can become very complex at great depths. The rock physics depth trends (i.e., VP, VS, and density versus depth) corresponding to the porosity-depth trends mentioned above can be rather complex due to the competing effects of porosity, pressure, mineralogy, texture, and pore fluids. In fact, we may observe more than one crossover in velocity-depth trends of sands and shales. Rock physics models can be very useful to better understand these depth trends. However, the models have to be calibrated to local geology before they can be used for further prediction of hydrocarbons and lithology. Geologic constraints include expected lithofacies and facies associations, sand and shale mineralogy (to determine effective elastic moduli and densities for the solid phase), fluid properties (oil density, GOR, gas gravity, brine salinity), and information about pressure and temperature gradients. For unconsolidated rocks we apply HertzMindlin contact theory (see appendix) to calculate elastic moduli of unconsolidated sediments as a function of porosity and pressure. Based on the elastic moduli, we calculate VP and VS versus depth. Density () is calculated directly from the porosity trends. From these parameters we can calculate acoustic impedance and VP/VS ratios versus depth. We calculate depth trends for clean sands, shaly sands, and shales. We assume 100% quartz and 0% clay for the clean sand trend and 80% quartz and 20% pore filling clay (smectite) for the shaly sand trend. Effective mineral moduli are estimated using Hills average (see Mavko et al., 1998). Bulk and shear moduli of quartz are 36.8 GPa and 44 GPa, respectively. Similar parameters for smectite are 15 GPa and 5 GPa. Figure 4 shows calculated trend lines of AI and VP/VS versus depth compared with observed well log data from the well intersected by the seismic line in Figure 2. We observe a very good match between the shale trend and the log data in the shaly intervals (i.e. zones with high gamma ray values), both in terms of acoustic impedance and VP/VS. Deviations from the modeled shale trend may reflect variation in silt content within the shales. The sandy reservoir zone nicely follows the shaly sand trend. Deviations from the clean sand and shaly sand trends reflect both presence of hydrocarbons (the trend lines are calculated for brine-saturated rocks) as well as

Downloaded 14 May 2011 to 203.199.41.231. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

Figure 5. Simple facies association model for mud-rich turbidite channel-levee complex and related interface categories.

individual classes in a binary AVO crossplot. Next, we extract VP, VS, and density for clean brine sand, pure shale, and shaly sand, from the calculated depth trends in the previous section. These are assumed to be the mean values for the different facies at the target level. We assume multiGaussian distributions where the variances are selected based on information from analog areas, or from nearby wells. In this example we have used nearby wells to calculate the variances. Moreover, we use Gassmann theory to estimate the rock properties for gas and oil saturated sands (see Mavko et al., 1998). (Fluid properties used for the turbidite field: Gas gravity = 0.7, oil reference density = 28 API, and brine salinity = 80 000 ppm). The resulting histograms of VP, VS, and density for different facies and fluids are shown in Figure 6. For each interface category, the expected AVO response at a target depth is calculated using a common approximation given by Shuey, valid for angles less than 30: R() R(0) + G sin2 , The zero offset reflectivity, R(0), is controlled by the contrast in acoustic impedance across an interface. The gradient, G, is mainly controlled by the contrast in VP/VS ratio. We do a Monte Carlo (MC) simulation to estimate the distribution of zero-offset reflectivity (R(0)) versus AVO gradient (G), based on the mean and covariances in VP, VS, and for the different interface categories. The structure of the covariance matrix determines the dependencies between the variables VP, VS, and . Normally, there is a higher correlation between VP and VS than between VP and . The resulting AVO scatter plots representative of the target depth, from which the AVO pdfs can be estimated, are shown in Figure 7. Seismic calibration and AVO classification. The final step in the AVO classification technique is to apply the modeled AVO pdfs to predict the most likely facies and pore fluid from seismic data. We did a blind test of the well intersecting the line in Figure 2, using the AVO pdfs derived from the modeled depth trends. R(0) and G estimated from pre-stack gathers along the line are calibrated to the modeled AVO pdfs in Figure 7. We identify a background window in the seismic section near or around the target interval. For the studied turbiditic environment we

Figure 6. Histograms of VP, VS and density for different lithologies and fluids at the target depth level corresponding to the reservoir sands penetrated by the well in Figure 4. The mean values are determined by the depth trends, while the variances are assumed to be depth-independent and are taken from a nearby well.

variability in clay content. A few local peaks of high impedance anomalies may reflect local cementation. Rock physics depth trends for cemented sandstone could be calculated using the Dvorkin-Nur cement model (see Mavko et al., 1998). However, in addition to the porosity, we would then need to know the amount of cement as a function of depth before calculating the elastic properties. To do this requires reliable information from geologists about the expected cement volume at a given depth. Statistical AVO modeling constrained by rock physics depth trends. The estimated acoustic impedance and
1008 THE LEADING EDGE OCTOBER 2003

VP/VS trends in Figure 4 can be used to calculate the expected AVO response with depth, for sand-shale interfaces. The example in this study is from a deepwater turbidite setting, and we assume 12 different interface categories. These are based on realistic layer configurations in a turbiditic environment, and are depicted in Figure 5. This model of facies associations is rather simplified compared to the true sedimentologic observations in the area, but we attempt to reduce the amount of interface categories while still honoring geologic variations that may be seismically significant. If we include too many interface categories, we may introduce too much overlap between

Downloaded 14 May 2011 to 203.199.41.231. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

Figure 7. Modeled AVO scatter plots of R(0) versus G for different interface categories for the target depth level. (See Figure 5 for explanations of the interface categories).

Figure 8. Calibration of AVO attributes extracted from prestack seismic data with modeled AVO pdfs in Figure 7. The black stars represent the AVO attributes estimated from the real data (i.e. crossplot of each sample in upper R(0) versus lower G section), while the colored dots represent the modeled AVO responses for the various interface categories. The modeled heterolithics (i.e., shaly sands) and shales in green and blue, respectively, represent the modeled background trend, which has been calibrated with the background trend in the real data.

assume the background trend to be characterized by interface categories 10-12, as most seismic horizons in a mud-rich turbiditic environment are made up of these categories. We cali1010 THE LEADING EDGE OCTOBER 2003

brate the covariance matrix of R(0) and G for the background trend in the seismic data with the covariance matrix of R(0) and G for the background trend in the model, either by matching the

covariances or by univariate variance matching. This calibration is then applied to seismic data in the target area (Figure 8). After calibrating the seismic data with the modeled AVO pdfs, we perform the AVO classification. We use the Mahalanobis distance to estimate the most likely layer category for each data sample in the data (see appendix). The classification result is shown in Figure 9, where we lump the 12 layer interface categories into 5 facies and fluid groups (tops and bases together). We obtain a good match with the observations in the well (compare with Figure 4). The top reservoir is successfully identified as gas bearing, while zones of oil sands are identified below the gas reservoir. However, significant parts of the reservoir are characterized as water-bearing. This could reflect the great overlap and ambiguities between oil and brine sands in terms of AVO properties. The cap rock is predicted to be predominantly heterolithics and shales. Bear in mind, however, that the final result in Figure 9 represents classification of interfaces, not layers. However, the methodology presented in this paper can also be applied to layer inversion results. For instance, elastic inversion could be classified using pdfs of AI versus VP/VS. In that case, the calibration of the seismic data is not necessary. Also note in Figure 9, that a few data points have been categorized as no class, and depicted in black. In the classification procedure, data points located a certain distance away from any of the modeled interface categories in the R(0)-G crossplot, are rejected. The unclassified units could either represent noise in the data or lithologies/facies not included in the modeling. We suspect these to be thin units of cemented sands. This would be in accordance with the well log data in Figure 4, where we observe a few anomalous high velocity peaks in the sandy target interval. One future extension of this AVO technique will be to include depth trends for cemented sandstone. It is also important to note that fluid properties will be depth dependent. Pressure and temperature control the compressibility of fluids but also the chemical properties of fluids can change with depth. In particular, relative density of oil (API gravity) tends to be depth dependent, where biodegradation of oil decreases with depth. Hence, shallow reservoirs will normally contain relatively thick oil, compared to deeper reservoirs. Trend lines of API gravity versus depth would be valuable information to be included in this AVO classification technique. Another

Downloaded 14 May 2011 to 203.199.41.231. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

clean sandstones (Cl=0) and a factor describing the sensitivity toward increasing clay, respectively. The clay index, Cl, is defined as the volume of total clays relative to the total volume of stable framework grains. Hertz-Mindlin contact theory. Elastic moduli of the dry rock are calculated using the following equations:

Figure 9. The most likely lithology and pore fluid along the seismic section in Figure 2.

future extension will be to include facies transition probabilities and spatial statistics to improve the constraints on the classification of vertical and lateral geologic variations from seismic data. Conclusions. The seismic signature of hydrocarbons can be very different from one depth to another due to different compaction trends for different lithologies. Therefore, it is necessary to include depth as a parameter when we use AVO analysis to predict lithology and pore fluids from seismic data. We have developed a depth dependent probabilistic AVO technique that enables us to predict the most likely lithology and pore fluid from seismic data, even in areas with sparse local well log information. Nevertheless, presence of local well log data will improve on the modeling of AVO pdfs, and give better control on the seismic calibration. The main limitations of the methodology include tuning and overburden effects, as well as the inherent ambiguities in rock physics properties and AVO response. The technique presented in this paper is an extension of the methodology presented by Avseth et al. (2001).
Suggested reading. Seismic reservoir prediction using Bayesian integration of rock physics and Markov random fields: A North Sea example by Eidsvik et al. (TLE, 2002). Seismic reservoir mapping from 3-D AVO in a North Sea turbidite system by Avseth et al. (GEOPHYSICS, 2001). Mapping lithofacies and pore fluid probabilities in a North Sea reservoir: Seismic inversions and statistical rock physics by Mukerji et al. (GEOPHYSICS, 2001). Examination of AVO responses in the eastern deepwater Gulf of Mexico by Smith and Sondergeld (GEOPHYSICS, 2001). Rock physics and AVO analysis for lithofacies and pore fluid prediction in a North Sea oil field by Avseth et al. (TLE, 2001). Statistical

rock physics: Combining rock physics, information theory, and geostatistics to reduce uncertainty in seismic reservoir characterization by Mukerji et al. (TLE, 2001). Geostatistical integration of rock physics, seismic amplitudes, and geologic models in North Sea turbidite systems by Caers et al. (TLE, 2001). Rock physics and seismic properties of sands and shales as a function of burial depth by Avseth et al. (SEG 2001 Expanded Abstracts). The Rock Physics Handbook by Mavko et al. (1998). Significance of geopressure in predicting lithology by Verm et al. (TLE, 1998). Porosity/depth trends in reservoir sandstones: assessing the quantitative effects of varying pore-pressure, temperature history and mineralogy, Norwegian Shelf data by Ramm and Bjrlykke (Clay Minerals, 1994).

where KHM and GHM are the bulk and shear moduli at porosity , respectively; P is the differential pressure; K, G, and v are the bulk and shear moduli of the solid phase, and its Poissons ratio, respectively; and n is the coordination number (i.e., number of grain contacts per grain). The wet rock moduli are calculated using the Gassmann theory (see Mavko et al., 1998). Classification using Mahalanobis distance. The Mahalanobis distance is defined as follows: M2 = (x-i)T i-1(x-i), where x is the sample vector, i are the vectors of means for the different classes, and i is the training data covariance matrix for class i. TLE
Acknowledgments: We thank Norsk Hydro and TotalFinaElf for allowing publication of the data used in this study. Thanks to Erling Vgnes and John Gjelberg, both at Norsk Hydro, for valuable geologic input, and Tapan Mukerji at Stanford University for helpful comments that improved the manuscript. Corresponding author: per.avseth@hydro.com

Appendix. Porosity-depth trends. Ramm and Bjrlykke (1994) suggested a clay dependent regression model for porosity versus depth of sands, due to mechanical compaction: =ce-(+Cl)Z where c is the critical (i.e., depositional) porosity, and and are regression coefficients representing a framework grain stability factor for

OCTOBER 2003

THE LEADING EDGE

1011

Downloaded 14 May 2011 to 203.199.41.231. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

Anda mungkin juga menyukai