INTERPRETERS CORNER

Coordinated byby lAn JAckson Coordinated A AlAn JAckson

Density extraction from P-wave AVO inversion: Tuscaloosa Trend example

Jyoti BehurA, BP America Inc. nurul kABir, richArd crider, and Petr Jilek, BP Exploration and Production Technology Group ellen lAke, BP North American Gas Strategic Performance Unit

ensity extraction from amplitude-variation-with-offset (AVO) inversion is thought to be unstable and difficult. Recent research has, however, shown that it is possible to reliably extract density from P-wave reflection data if the interface has significant contrast. This approach is tested over the Tuscaloosa gas reservoir where petrophysical evidence suggests that density is the primary driver of reflectivity. By adopting a linear as well as a nonlinear approach, inversion is performed for density and P-wave velocity from a 3D prestack seismic data set. The nonlinear inversion results correspond well with the log and production data, thus proving that density can be reliably extracted in this case and used for reservoir characterization. We use the extracted density along with the P-wave velocity to distinguish the low-density porous gas sands from tight sands and shales. Background The Tuscaloosa Trend in South Louisiana is a series of prolific gas fields. The Cretaceous reservoir sands were deposited in a near-shore marine depositional environment. The reservoirs consist of gas-saturated sands along with partially saturated gas sands, tight sands, and shales. Reduction of drilling risk and optimization of production require delineation and separation of gas sands from other lithologies. One possible method for lithology discrimination is through extraction of physical parameters (P-wave velocity, S-wave velocity, and density) using AVO inversion. In isotropic media, P-wave AVO is a function of four quantities: g (background Vp/Vs), P-wave velocity contrast, S-wave velocity contrast, and density contrast across the interface. However, a common impression is that these parameters cannot be uniquely found using AVO inversion; i.e. the parameters cannot be decoupled from one another using AVO inversion without long-offset data. Estimation of density can be particularly difficult (Debski and Tarantola, 1995) and requires a wide range of incidence angles to be stable as pointed out by Li (2005) and Lavaud et al. (1999). Kabir et al. (2006), however, showed that density can be reliably estimated from near and mid offsets when density contrast is the dominant contributor to the reflection coefficient (Figure 1). Consider a hypothetical case of a reflection from an interface separating two media having the same P- and S-wave velocities and differing only in density. In this case, the reflection coefficient at any angle would be due only to the density contrast across the interface (the reflection coefficient would be zero if the density contrast were also zero). Inverting the reflection coefficient would thus yield the density contrast.
772 The Leading Edge July 2010

Figure 1. Exact P-wave reflection coefficients (RPP) for an interface between a shale overlying a gas sand. The model parameters for the shale are Vp = 2560 m/s, Vs = 1036 m/s, and = 2400 kg/m3. The parameters for the gas-sand are Vp = 2743 m/s, Vs = 1524 m/s, and = 2000 kg/m3. The different curves correspond to different densities of the gas-sand varied between 1600 and 2400 kg/m3.

This was also demonstrated by Lavaud et al. who conclude that discriminating different lithologies and fluid saturations is possible if the density contrast is much higher than the P- and S-wave velocity contrasts. Using modeled data, they showed that density was the discriminating factor between the reflections from a 20% and a 100% gas-saturated sand underlying a shale. Under this assumption, Kabir et al. successfully related the estimated density to gas saturation in the Mahogany gas field offshore Trinidad. The same idea is expressed by Li who concludes, "In general, high-porosity reservoirs and reservoirs with better density contrasts or anomalies are appropriate candidates for applying density inversion." The porosity of Tuscaloosa sandstones can be as high as 27%, leading to their low density. Petrographic analysis of these sandstones obtained from outcrops and wells suggests that the dissolution of iron-rich calcite and volcanic rock fragments has resulted in high secondary porosity. In this study, because the density contrast between the Tuscaloosa gas sands and shales is high (shown by petrophysical analysis below), inversion for density provides an additional parameter for reservoir characterization. Petrophysical analysis A crossplot of density and P-wave velocity logs from one well is shown in Figure 2a with the data points color-coded by gamma ray. Note that the low-density and low P-wave velocity sands (red ellipse) are the most productive gas sands (with the low density resulting from high porosity). As water

INTERPRETERS CORNER
Density, on the other hand, can discriminate between gas sands and shales (Figure 2a). Not only can density help distinguish gas-sands from shales, but also, in combination with P-wave velocity, can help separate gas sands from wet and tight sands. Therefore, if one could extract density information from seismic data, gas sands could be separated from tight sands and shales, which is not possible using P-wave velocity or impedance alone. As discussed above, extraction of density from P-wave reflection amplitudes is possible only if it is the primary driver of reflectivity. For the Tuscaloosa reservoir system, the density of most sands is quite different from that of the shales. Their P-wave velocities, however, show a significant overlap. Consequently, we conclude that density is the primary driver of P-wave reflectivity between gas sands and shales in this area. Therefore, from the petrophysical analysis and the work of Lavaud et al., Li, and Kabir et al., it follows that density can be reliably estimated from P-wave reflection amplitudes in this case. Linear versus nonlinear inversion A similar procedure as in Kabir et al. is adopted to separate gas sands from tight sands and shales using density extracted from reflection amplitudes at near-to-mid angles (030). Both a linear and a nonlinear inversion algorithm are used to compare their performance. The commonly used linearized reflection coefficient equation (e.g., Ruger, 2002; Smith and Gidlow, 1987) is used in the linear inversion which is a function of four parameters: average Vp/Vs, the contrasts in Pwave velocity, S-wave velocity, and density. This conventional linearized approximation of the P-wave reflection coefficient is valid only for small contrasts in rock properties across the interface. Because of this limitation, use of this equation in AVO inversion would not yield accurate rock properties, especially in cases where the contrasts in rock properties across the interface (particularly density) are significant. Instead of the above linearized approximation, one could use the full Knott-Zoeppritz equation in nonlinear inversion. This, however, would make the nonlinear inversion unstable because, instead of inverting for four parameters, we would be inverting for six parameters (four velocities and two densities). To overcome this drawback, Lavaud et al. (1999) reparameterized the Knott-Zoeppritz equation in terms of a background parameter and three "contrast" parameters (contrasts in P-wave velocity, S-wave velocity, and density). Their parametrization imposes no limitations on the magnitude of the contrast parameters and is valid for all offsets. Also, this formulation eliminates the need to use any approximation formula and makes it possible to perform AVO inversion using the full Knott-Zoeppritz equations. Therefore, we use the reparameterized Knott-Zoeppritz equation of Lavaud et al. in the nonlinear inversion. Because nonlinear inversion is based on the exact AVO equation, it is likely to yield more accurate results than linear inversion. For the same reason, individual parameters are better resolved, even though this inversion is more computer intensive. Nonlinear inversion, however, can be more unstable compared to linear inversion, and its solution might get
July 2010 The Leading Edge 773

Figure 2. Crossplots of P-wave velocity and density, color-coded by (a) gamma-ray measurements and (b) P-wave impedance from a well in the study area. The gas sands have densities significantly different from the shales; their P-wave velocities, however, are similar. The P-wave impedance of the gas sands is also coincident with that of the shales. Therefore, density can help distinguish the gas sands from shales.

saturation increases, the density and P-wave velocity of these porous sands (green circle) increases, with tight sands (blue circle) having the highest density and velocity. Shales (black ellipse), on the other hand, have high densities and a wide range of P-wave velocities. The lowest density gas sands and the partially saturated gas sands have densities that are significantly different from that of the shales; their P-wave velocities, however, show a substantial coincidence. Therefore, P-wave velocity alone would not help distinguish the gas sands from shales. Similarly, P-wave impedance also cannot differentiate between gas sands and shales as is clear from Figure 2b. If the reservoir were devoid of shales, P-wave impedance would be enough to distinguish the gas and tight sands. However, because the Tuscaloosa reservoir system contains shales with P-wave impedance similar to that of gas sands (Figure 2b), P-wave impedance alone would not help locate the gas-saturated sands.

INTERPRETERS CORNER

Figure 3. Comparison of performance of nonlinear (blue) and linear (magenta) inversions for density (left), P-wave velocity (middle), and acoustic impedance (right). White curves represent synthetics calculated from well logs. Nonlinear inversion shows a better match between synthetics and well logs in all three panels.

trapped in a local minimum, thus yielding erroneous results. In this case, however, because we have good initial estimates of velocities and densities from the well logs, the nonlinear inversion is more stable. The objective function used by this nonlinear algorithm is the least-squares error between the observed and computed data (l 2-norm). Detailed mathematical formulation of the algorithm is given in Lavaud et al. A comparison between the two inversions is shown in Figure 3.The white curves represent synthetics calculated by computing contrasts in the individual parameters from the well logs and then convolving them with a wavelet. Note the good match between the log curves and the nonlinear inversion results for all three parameters. Linear inversion, on the other hand, does not show a good match with the log curves, especially for P-wave velocity. Therefore nonlinear inversion performs better than linear inversion in this case. From now on, only nonlinear inversion results will be discussed. Nonlinear inversion results Figure 4 shows the density-contrast section obtained from nonlinear inversion. Note that this is a contrast section, not a layer-parameter section; this section shows only when there is a contrast in density across an interface between two layers. The two red arrows point to locations where density is decreasing with depth (as seen on the density logs). This implies a negative density contrast which matches well with the inverted contrast section. Good agreement can also be seen at locations indicated by the two black arrows. Here the density increase on the log overlaps with a positive contrast on the inversion result. To find layer properties, colored inversion is performed
774 The Leading Edge July 2010

Figure 4. Density-contrast section obtained from nonlinear inversion. Horizons A and B are marked in yellow and red, respectively, along with one dry and two producing wells. Hot colors represent positive contrast; cool colors represent negative contrasts. The density log is white and the P-wave velocity log is magenta.

on the inverted contrast sections. Colored inversion involves computing an operator using contrast volumes and well logs followed by application of that operator on the whole contrast volume to obtain layer properties (Lancaster and Whitcombe, 2000). Figure 5a shows the interval density of layers obtained after colored inversion. Note the good match between the density log (in white) and the inversion result at the two locations marked by arrows. The logs show low-density values which are in agreement with the inversion. Near well

INTERPRETERS CORNER

Figure 5. (a) Density and (b) P-wave velocity sections after colored inversion. Hot colors represent high values; cool colors represent low values. The density log is white and the P-wave velocity log is magenta.

1, the inversion shows a high-density body just below the B horizon, which cannot be a porous sand according to our petrophysical analysis. This is corroborated by drilling and production results. Wells 2 and 3 are good producers as they pass through low-density porous sands while well 1 is a dry hole with nonporous sands. The inverted P-wave velocity after colored inversion is shown in Figure 5b. Note that the same locations as those marked by arrows in the density section (Figure 5a), have low P-wave velocities. According to the petrophysical analysis, these should be porous gas-sands. Figure 6a is the density map extracted for the B sand. The gas field in this area is trapped on a large faulted structural high. The beds are predominantly dipping to the north and northwest as seen from the contours. The best B production comes from the northeast side of the structure where there is a substantial correlation to the low-density areas shown on the extracted density map. Wells 2 and 3 both lie in these low-density areas. The low-density anomalies on the east and west sides of the field (denoted by white arrows) probably

Figure 6. (a) Density and (b) P-wave velocity maps extracted for the B sand. Hot colors represent high values while cool colors represent low values. The contour in red denotes the approximate gas/water contact for the B horizon.

imply porous sands, but these areas do not have any well penetrations. The combination of low density and low P-wave velocity is a high-probability indicator of a productive gas sand. For the highly productive portion of the B horizon, low-velocity regions (Figure 6b) overlie low-density regions. The two productive wells (2 and 3) lie in low P-wave velocity regions while the dry hole (well 1) is in the high P-wave velocity region. The cumulative production from the B horizon in the well 2 fault block is 57 bcf and in the well 3 fault block it is 139 bcf. This can be contrasted with the western fault block which has both high density and high P-wave velocities; this block has produced only 2.4 bcf of gas. For the low-density anomalies noted earlier on the east and west sides of the field (denoted by white arrows), there are no distinct features on the P-wave map. At some locations, the P-wave velocity is high while at others it is low. For this low density and erratic velocity, the fluid properties for these anomalies cannot be

776

The Leading Edge

July 2010

INTERPRETERS CORNER
accurately determined. However, as these areas are both structurally low to the gas-water contacts in the field, the sands are undoubtedly wet here. Discussion Even though density inversion was successful in this field, this technique should not be blindly applied to all areas. A thorough petrophysical analysis is critical to understand the feasibility of this approach. For example, in this study the anomalously high porosity of the Tuscaloosa sandstones results in densities much lower than the surrounding shales. Such a large density contrast makes AVO inversion for density feasible. Care should be taken while interpreting the results as a mismatch was observed at one well outside the field (not shown). The well passes through a thick wet sand but inversion results show a high-density region. The thick sand zone has many interbedded shales and thus interference of reflections and tuning might be yielding wrong inversion results. There might be other explanations for this contradictory result: data not zero-phased at the well; inversion trapped in a local minima; contamination by multiples; and assumptions such as absence of attenuation and anisotropy violated. More analysis needs to be done to understand the cause of this mismatch. The important lesson is to be careful while interpreting results; geological information and other evidence should be taken into account to improve the chances of success. Conclusions Petrophysical analysis of the Tuscaloosa reservoir suggests that the fully and partially saturated gas sands can be separated from the interbedded tight sands and shales if both density and P-wave velocity are known. Though density inversion is usually unstable, in this case it is feasible because of the significant contrast in density between the gas sands and shales caused by the anomalously high porosity of the sandstones. This makes density the primary driver of reflectivity, thereby making AVO inversion for density stable. The extracted density and P-wave velocity (from nonlinear AVO inversion) showed reasonable correspondence with log data. The productive and dry zones interpreted from the inversion results showed good agreement with drilling and production results.
References
Debski, W. and A. Tarantola, 1995, Information on elastic parameters obtained from the amplitudes of reflected waves: Geophysics, 60, 14261436. Kabir, N., R. Crider, R. Ramkhelawan, and C. Baynes, 2006, Can hydrocarbon saturation be estimated using density contrast parameter?: CSEG Recorder, 3137. Lancaster, S. and D. Whitcombe, 2000, Fast-track colored inversion: 70th Annual International Meeting, SEG, Expanded Abstracts, 19, 15721575. Lavaud, B., N. Kabir, and G. Chavent, 1999, Pushing AVO inversion beyond linearized approximation: Journal of Seismic Exploration, 8, 279302. Li, Y., 2005, A study on applicability of density inversion in defining reservoirs: 75th Annual International Meeting, SEG, Expanded Abstracts, 24, 16461649. Ruger, A., 2002, Reflection coefficients and azimuthal AVO analysis in anisotropic media: SEG. Smith, G. and P. Gidlow, 1987, Weighted stacking for rock property estimation and detection of gas: Geophysical Prospecting, 35, 9931014.

Acknowledgments: We thank BP for permission to publish this paper. We also thank Seismic Exchange, Inc. for providing the seismic data. The seismic data are owned or controlled by Seismic Exchange, Inc.; the interpretation is that of BP. The thorough reviews of Alan Jackson and Rebecca Latimer helped improve the manuscript greatly. Corresponding author: Jyoti.Behura@bp.com

July 2010

The Leading Edge

777