Вы находитесь на странице: 1из 5

Use of elastic prestack seismic data for reservoir imaging and characterization

Karine Broto, Thierry Tonellot, Andreas Ehinger, Frdrique Fournier, Vincent Richard, Institut Franais du Ptrole, France, Jan H. Kommedal, BP Exploration, UK

Abstract
There is a clear trend in the geophysical industry towards the development of integrated seismic processing, in particular vertical integration from acquisition to reservoir characterization but also horizontal integration by use of well data and geological information. Whereas inversion formalisms provide a natural framework for the latter, adequate choices have to be made for ensuring the former. Velocity models required for imaging PP and mode converted PS waves are built in a data consistent way by means of traveltime tomography. These velocity models are then used to obtain seismic data sorted by incidence angle. The following step is a stratigraphic inversion of limited angle stacks in order to quantitatively estimate P and S impedance variations at the reservoir level. The techniques involved in this global approach have been applied to data sets from the North Sea and from offshore West Africa. Consistent PP and PS images have been obtained from tomographic inversion. P and S impedances have been estimated and interpreted in connection with the reservoir properties.

Velocity model determination


Velocity model determination is the first step on the way from seismic field data to reservoir characterization. Our velocity model determination method is based on prestack traveltime tomography. This iterative inversion method searches for a model that minimizes an objective function that measures the misfit between the traveltimes computed by ray-tracing in the current model and the observed traveltimes (interpreted reflection events in the prestack data). Besides, regularization and other geological and geophysical a priori information can be introduced into the tomographic objective function. Our traveltime tomography software jerry1,2 uses a blocky velocity model representation. Whereas no correlations between the (spatially varying) P and S block velocities are assumed, the P and S models share the same topology (i.e. the same block limits). This implies that the interpreter has to recognize and interpret PP and PS reflection traveltimes generated at the same subsurface interface. Note that this joint and consistent interpretation ("pairing") can be sensibly facilitated in the depth domain3 and the corresponding PP and PS traveltimes can then be obtained by kinematic demigration4. Pairing in the prestack domain ensures that the PP and PS reflectors are imaged at the same depth positions ("co-depthing"). Since traveltime tomography allows the simultaneous processing of PP and PS traveltimes5 (contrary to most methods used in the industry which are intrinsically limited to a sequential approach) it is also guaranteeing that both the PP and the PS depth images are optimally

focused ("co-focusing"). Nevertheless, for a successful minimization of the PP and PS traveltime misfits, the forward problem (i.e. ray-tracing) has to describe the real wave propagation in the earth as accurately as possible. In particular, when present, anisotropy has to be taken into account. Our traveltime tomography software allows to handle TI (Transverse Isotropy) anisotropy, which is the most encountered type of anisotropy in seismic exploration. We use a new approximation of the phase velocity equations which is physically valid for a large range of anisotropy values and which is still simple to handle6. The anisotropy parameters can be determined either by tomographic inversion or simply taken into account for the ray-tracing7. Note, however, that due to the generally strong underdetermination of one of the anisotropy parameters (), not to invert for it but rather to use a fixed value if any external information allows to define such a value . We applied this methodology on one 2D line of a 3D 4C OBC data set recorded over the Norwegian North Sea . Except for the layer above the reservoir, PP and PS results obtained by classical isotropic inversion are very satisfactory: the differences between observed and calculated traveltimes are very low and the P and S velocity distributions are geologically plausible. As for the layer above the reservoir, the determination of the anisotropy parameters turned out to be the key factor for obtaining at the same time sufficiently small traveltime misfits and geologically plausible layer velocities.
8 8

Prestack stratigraphic inversion and interpretation


As seismic amplitude variation with incidence angle holds information on P and S impedances, AVA (amplitude versus angle) analysis is used regularly in the conventional interpretation process, specially for offshore applications. This analysis is based on the Zoeppritz equation (or on some approximation thereof), and is applied on prestack seismic data sorted by incidence angle, after preserved amplitude processing and NMO correction. This kind of sorting can be achieved through ray tracing after the precise estimation of the velocities, found by using for instance traveltime tomography. After the sorting, limited angle stacks data can then be generated and processed. Conventional AVA inversion methods consist in a two-step procedure: first, prestack attributes such as intercept, gradient or Connolly's9 elastic impedances are extracted from the angle data. Then these attributes are inverted for P and S impedances. In general, these methods are very sensitive to seismic noise and to pre-processing issues. Moreover, depending on the noise level and on the angle range, it is well known that the S impedance is poorly determined. We propose a one-step, 3D model-based prestack stratigraphic inversion methodology10 in which all the angle data available are simultaneously and directly inverted for P and S impedances. Our methodology is based on a formalism which uses a priori information to constrain the elastic parameters and to improve their quantitative determination. The first part of the methodology consists in a detailed well-to-seismic calibration, in order to compute one single seismic wavelet for each angle stack. Each wavelet will thus be able to compensate for some pre-processing issues, like wavelet variations with angle, NMO stretch or tuning, which may cause the determination of the elastic parameters to become less reliable. This step by step multi-well

calibration11 provides an optimal seismic wavelet which best explains the observed seismic traces at and around the wells for each angle stack. The second part of the methodology consists in the construction of an a priori model. First, we design a geometrical framework using both structural and stratigraphic information: from interpreted horizons and faults we define geological macro units. Then, within each unit, we define a depositional mode which describes the internal geometrical framework. Next, a 3D a priori model is build for each elastic parameter, by interpolating well logs along the geometrical framework, using for example a kriging method. The third and last part of the methodology consists in the simultaneous stratigraphic inversion of all the angle data. This inversion is achieved by finding the model which minimises an objective function made up of two terms. The first one measures the misfits between the observed angle data and the synthetic angle data. These synthetic data are computed, for a given model and a given incidence angle, by convolving the corresponding Zoeppritz coefficient series by the previously calculated angle wavelet. The weight of this term in the objective function depends on the noise level of the observed seismic data. The second term of the objective function measures the distance between the current model and the a priori model. This term allows to take into account our confidence in the a priori model not only in terms of values but also in terms of 3D geometry. In conclusion, our 3D stratigraphic AVA inversion approach incorporates a 3D geologically oriented filter within the joint angle stack inversion processing, in order to quantitatively estimate an optimal multi-parameter model in P and S impedances which best explains all the angle data available. This model can then be interpreted in order to estimate the reservoir properties. In most cases, P and S impedances, or other related attributes such as Poisson coefficient, or Lam parameters, are not related to reservoir characteristics in a simple way. Moreover, the difference of scale between the band limited seismic attributes estimated from the stratigraphic inversion and the reservoir heterogeneities described at the wells is an additional difficulty when interpreting the inversion results. So, very often, the interpretation cannot be based on a simple deterministic relationship between the inverted impedances and the reservoir characteristics. We propose to use a statistical calibration methodology12 for relating the impedances after inversion to relevant reservoir characteristics, such as average porosities or water saturations, lithofacies proportions, etc. A statistical relationship is built using estimation techniques, or discriminant analysis if this relationship is highly non-linear, in the vicinity of wells where seismic parameters and reservoir information are both available. From the analysis of the calibration data set, it is also possible to select which reservoir property is related to the seismic information, and what are the best seismic attributes to be used in the interpretation process. In a second step, this relationship is applied to the seismic attributes of the traces far from any well to predict the reservoir characteristics at these locations.

Furthermore, an uncertainty is also attached to the predicted reservoir property, since the calibration is built within a probabilistic frame. We have applied our 3D AVA stratigraphic inversion and interpretation methodology on a 3D marine data set from a deep-offshore turbiditic reservoir. Five angle stacks were available after preserved amplitude processing and NMO corrections. They correspond respectively to the angle ranges 0-6, 612, 12-18, 18-24 and 24-30 degrees. Each angle stack is made up of 141 lines, with 251 traces (cdp's) per line. Well logs (P and S impedances) were available at three wells and an interpretation of the main reflections in the near angle stack was also available. After the multi-well calibration step of our methodology, we have built the a priori elastic model and inverted the five angle stacks. On the deep-offshore turbiditic reservoir under study, it was of particular interest to predict the sand and shale distribution from the inversion results. For this, we have first built a lithofacies columns from the well logs and we have then defined cells consistent with the seismic bandwidth. We then associated to each cell a pseudo rock-type based on the dominant lithofacies in the cell. A total of five pseudo rock-types has been defined, ranging from hemipelagic shales to coarse sands. In order to establish the link between the pseudo rock-types and the available seismic attributes we used discriminant analysis. It showed, through the discriminant analysis performance criteria, that, among the different attributes that could be used (impedances, Lame parameters, etc), the P and S impedances were the most efficient ones to identify the five pseudo rock-types. It also illustrated the contribution of S impedances that clearly identified some sand categories that the sole P impedance could not discriminate from shales. Finally, the assignment function built at wells with discriminant analysis was applied to the complete P and S impedance volumes allowing us to predict complete volumes of pseudo rock-types associated with a probability of good assignment. Using this probability as a reliability index, it was possible to derive a minimum and maximum extension for the connected sands, and also to study the connectivity of non-reservoir facies, both information being crucial for the reservoir development.

Acknowledgements
The reflection tomography software is developed in the framework of the Kinematic Inversion Methods consortium project (KIM). The authors hereby acknowledge the support provided by the sponsors of this project. We are grateful to BP Norway for providing the 2D-4C North Sea data set and for the permission to publish. We also thank Alexandre Stopin for precious discussions, Bertrand Duquet for providing the PP and PS isotropic 2D prestack depth migration code, Hery Rakotoarisoa and his team for the anisotropic one.

References
1. Jurado, F., Sinoquet, D. and Ehinger, A., 1996, 3D reflection tomography designed for complex structures, 66th Ann. Internat. Mtg, Soc. Expl. Geophys., Expanded Abstracts. 2. Jurado F., Lailly, P. 81. 3. Broto, K., Ehinger, A., Kommedal, J. H., Folstad, P., 2001, Tomographic inversion of PP and PS reflections: a 2D North Sea case study, 71th Ann. Internat. Mtg., Soc. Expl. Geophys., Post Convention Workshop on Elastic Rock Properties - The Inversion of the Seismic Vector Field. 4. Ehinger, A. and Lailly, P., Velocity model determination by the SMART method, Part 1: Theory, 65th Ann. Internat. Mtg, Soc. Expl. Geophys., Expanded Abstracts. 5. Stopin, A. and Ehinger, A., 2001, Joint PP PS tomographic inversion of Mahogany 2D-4C OBC seismic data, 71th Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts. 6. Stopin, A., 2000, Comparison of v() equations in TI medium: Fractures, Converted Waves, and Case Studies: Ninth International Workshop on Seismic Anisotropy, paper 20. 7. Stopin, A., 2000, Bending ray-tracing and reflection tomography in TI medium: Fractures, Converted Waves, and Case Studies: Ninth International Workshop on Seismic Anisotropy, paper 25. 8. Broto, K., Ehinger, A., Kommedal, J. H., Folstad, P., 2001, Depth consistent PP and PS imaging using tomographic velocity models, to be submitted to The Leading Edge. 9. Connolly, P., 1999, Elastic impedance: The Leading Edge, Vol. 18, No. 4. 10. Tonellot, T., Mac, D. and Richard, V., 2001, Joint stratigraphic inversion of angle-limited stacks. 71st Ann. Internat. Mtg., Soc. of Expl. Geophys., Expanded Abstracts. 11. Lucet, N., Dquirez, P-Y., and Cailly, F., 2000, Well to seismic calibration: A multiwell analysis to extract one single wavelet: 70th Annual Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts. 12. Fournier, F., Rozanski, J., Doligez, B., 2000, How to constrain 3D reservoir stochastic simulations with seismic: a case study in carbonate deposits, GeoArabia, Vol. 5, No. 1. and Ehinger, A., 1998, Fast 3D two-point raytracing for traveltime tomography, Proceedings of SPIE, Mathematical Methods in Geophysical Imaging V, pages 70-

Вам также может понравиться