68th EAGE Conference and Exhibition incorporating SPE EUROPEC 2006

Two sets of 4D seismic data gave major new insights into the structure and dynamic behaviour of the Gannet C oil and gas reservoir in the UK Central North Sea. The 4D data revealed major extensions of reservoir units previously presumed to be absent or thin over much of the reservoir. Furthermore, in a subsea field with significant uncertainty of production allocation, 4D also proved an invaluable history matching parameter for the dynamic model. Together, the dynamic model and the 4D data gave rise to the identification of one recompletion opportunity and two infill well opportunities, to produce oil volumes in existing and newly identified reservoir sands.

This paper presents a 4D simultaneous inversion case study from the Marlim field, offshore Brazil. The paper aims at illustrating the influence of processing for time-lapse AVO inversion.
An AVO inversion of the Marlim 4D seismic dataset has been performed on an original dataset, not acquired or processed for time-lapse purposes. A seismic pre-conditioning processing sequence was applied to the post-stack migrated angle stacks to improve the seismic repeatability. The sequence included the creation of a 'common' seismic dataset generated from the coherent signal between base and monitor data. The common dataset is characterized by higher signal-to-noise ratio than both base and monitor vintages. The pre-conditioning efficiently improves the seismic repeatability of the vintages. However, the inversion results show a significant amount of noise contaminating the 4D signal.
For this reason, on behalf on Petrobras, the 4D dataset has been subsequently reprocessed in parallel by CGG. The reprocessing involved the application of the same processing sequence to the two vintages. The AVO inversion is being repeated on the new dataset. The 4D parallel reprocessing dramatically increases the repeatability of the two vintages allowing an efficient reduction of the 4D noise for the inversion.

Streamer positioning technology has evolved from streamer shape computations based on magnetic compasses to hybrid systems combining compasses with limited acoustic sensors to improve positioning errors. Correlated positioning errors can have a detrimental impact on the resolution of 3D surveys. The potential impact on 4D seismic is even more serious. These problems can be addressed by the use of more rigorous fully cross-braced acoustic positioning networks. Such networks can improve positioning accuracy by 60% and facilitate the use of steerable streamers for minimizing streamer separation, streamer offset from obstructions and errors in repeatability for 4D seismic monitoring surveys.

Conventional approaches in time-lapse studies often ignore seismic transmission effects such as attenuation. For example, gas injection may produce changes in the factor Q. Consequently, amplitude and phase variations in time-lapse seismic data may be wrongly interpreted To correct such spectral distortion, we present a cross-equalisation technique based on differential Q-controlled calibration. The methodology should be applied in a 4D context when frequency attenuation variation occurs inside a reservoir or in the overburden.
The spectral characterisation of the data is tested for two different techniques: Multi-taper and Wavelet decomposition. We present the advantages and the disadvantages specifically for time-lapse studies.
The cross-equalisation is defined as "attribute-driven-processing". Using the constant Q definition, an attribute, called 4DQ, is computed simultaneously on the base and monitor with linear fitting of the logarithmic spectral ratio.
The calibration, controlled by the 4DQ attribute, removes the effect of the absorption variation.
The methodology is tested on real data. The measured 4DQ attribute shows a clear spatial correlation with the reservoir in production, but its direct interpretation seems to be critical. Furthermore, the calibration process and a single-Q compensation are applied providing a high-resolution 4D signature.

Hydrocarbon bearing reservoir rocks often contain a mixture of several fluids in their porespace (e.g. oil, water and gas). If the pore fluids are immiscible and form pockets on a mesoscopic length scale (exceeding the typical pore size but still small compared to seismic wavelength), the fluid saturation is referred to as patchy saturation. Elastic waves travelling through such a rock will exhibit a characteristic frequency-dependent attenuation and velocity dispersion. These dynamic effects are believed to contribute significantly to the overall characteristics of the seismic wavefield. We numerically simulate wave propagation in a partially saturated rock model containing water with gas inclusions and extract attenuation and dispersion from synthetic seismograms. The results are compared to a theory of frequency-dependent attenuation and velocity dispersion in partially saturated media. Our numerical results are in reasonable agreement with those predicted theoretically. Furthermore, we are able to accurately infer information about the size of the patches from the extracted frequency-dependent attenuation. This underlines the possibility that the size of the fluid patches can be estimated from seismic data.

Ultrasonic P- and S-wave velocity (Vp, Vs) and attenuation (Qp-1, Qs-1), including azimuthal variations in Vs (wave propagation parallel to bedding), were measured on 12 core samples from a Russian carbonate reservoir at 40 MPa effective pressure using the laboratory pulse echo-system. While relationships between velocities, attenuations and porosity and permeability do not reveal any significant features, cross-plots of the ratios Vp/Vs and Qp-1/Qs-1 allow some categorisation of these carbonate rocks in terms of porosity and permeability ranges. In particular, low porosity (< 5%), low permeability (< 0.1 mD) carbonate rocks plot in a separate group to intermediate porosity and permeability carbonate rocks. The results show the value of combined velocity and attenuation datasets to reservoir characterisation.

Discrete element method (DEM) is used as a tool for modeling seismic wave propagation as well as stresses and strains associated with reservoir depletion. The advantage of DEM lies in its ability to model strain localization and evolution of fractures, which can be directly linked to time-lapse seismics and micro-seismic events. For this purpose, DEM is tested to verify its ability to model wave propagation. Then a simplified 2D synthetic model for typical North Sea field is constructed. The simulated geomechanical response to reservoir depletion is compared to that obtained by use of the Finite element method. Finally a simple 4D seismic profile is created for the reservoir top.

We present a depth-domain processing flow-chart based on iterative ray+Born migration/inversion and Very Fast Simulated Annealing (VSFA) inversion. The quantitative migrated image computed by ray+Born inversion is used as an input in the VSFA inversion to build a structural velocity model. Aim of the VSFA inversion is to remove limited-bandwidth effects in the migrated image resulting from the limited bandwidth of the source and the limited aperture coverage. The input data of the VSFA inversion are velocity profiles of the migrated images and the output models are the corresponding structural velocity profiles. The forward problem associated with the VSFA inversion is approximated by a time-domain convolution with the source wavelet, which makes a global exploration of the model space to be possible.
The flow-chart is assessed thanks to the Marmousi model. We computed 9 iterations of single-arrival ray+Born modelling/inversion, which allowed us to derive an accurate true-amplitude migrated image. Second, we applied the VSFA inversion using as input all the vertical profiles of both the true perturbation model (obtained by band-pass filtering the Marmousi model) and the ray+Born migrated image. The output is a structural velocity model, which mimics the true Marmousi model.

In seismic reflection tomography, the velocity model of the subsurface is updated by back-projecting travel-time residuals along ray-paths. The travel-time residuals are picked from the seismic data itself and the methodology used to gather these picks is a fundamental part of any velocity inversion workflow. In particular, the density at which the residuals are represented in the four-dimensional data space (inline, crossline, offset and depth/time) appears to have a significant effect on the precision of the velocity updates that are output from the tomographic inversion. Utilising dense, hyperbolic- (or parabolic-) fitting means that the residuals are finely sampled in the data space but does not necessarily represent their true values with great accuracy. Dense, non-hyperbolic fitting offers a similarly fine sampling but with greater adherence to the true residual value. These two methodologies have been compared and contrasted on a complex synthetic dataset. It can be seen from this comparison that the dense, non-hyperbolic tomography offers greater potential for resolving small-scale velocity heterogeneities in the Earth.

The resolution of first arrival traveltimes tomography is limited by the size of the first fresnel zone for each ray, the experimental device and the structure studied itself. This leads to uneven ray coverage and locally varying resolution. Besides, the sensitivity kernel of each ray induce high frequency information even where theoretical resolution is poor.Smoothing constraints are often added to the tomographic system to deal with these problems. We propose an alternative adaptive parametrization based on second generation wavelet transform, as introduced by Wim Sweldens.

We explain the principles and interesting properties of wavelet transform and show how we insert it in the tomography process. Then we present a synthetic example showing difficulties generated by rays in velocity model recontruction. We present the effects of classical gaussian smoothing and of our adaptive parametrization using wavelets.We show that wavelet transform allows a better control of the resolution power depending on the parameter location and that it is more adaptive than classical methods.