72nd EAGE Conference and Exhibition incorporating SPE EUROPEC 2010

Full waveform inversion (FWI) has received an increasing amount of attention thanks to its ability to provide a high resolution velocity model of the subsurface. The computational cost still presents a challenge, however, and the convergence rate of the FWI problem is usually very slow without proper preconditioning on the gradient. While preconditioners based on the Gauss-Newton Hessian matrix can provide significant improvements in the convergence of FWI, computation of the Hessian matrix itself has been considered highly impractical due to its computational time and the storage requirements. In this paper, we design preconditioners based on an approximate Gauss-Newton Hessian matrix obtained using the phase-encoding method. The new method requires only 2Ns forward simulations compared to Ns(Nr+1)forward simulations required in conventional approaches, where Ns and Nr are the numbers of sources and receivers. We apply the diagonal of the phase-encoded Gauss-Newton Hessian to both sequential source FWI and encoded simultaneous source FWI. Numerical examples on Marmousi model demonstrate that phase-encoded Gauss-Newton Hessian improves the convergence of the FWI significantly.

Thusfar, waveform-based inversion methods have utilized the spatial distribution of back-projected reflectivities in order to refine velocity (and density) models. While these methods excel at refining a velocity model for high-wavenumber features, they face significant difficulties when attempting to invert for long-wavelength features several kilometres below the depth of acquisition. In this abstract, we develop the theory for a waveform-based inversion method that utilizes an impedance image, rather than a reflectivity image, in order to extract spatial variations in velocity and density. Results are presented for the first stage of an inversion using a 2-D line of dual-sensor, single-streamer, field recordings. It is shown that features with vertical scale sizes up to 0.5 km can be determined at depths up to 4 km, using data with a maximum offset of only 8 km and frequencies below 7.5 Hz.

We have developed a 3D tomographic wavefield inversion code that solves the fully elastic wave equation in the time domain using finite differences. We show results of applying this elastic code to different synthetic 3D problems.

An important ingredient of any tomography-based velocity inversion method is the determination of traveltime differences. When a ray-based method is used, the modelled traveltimes are directly available and the traveltimes in the observed data need only to be picked once. When using wave-equation-based methods, however, the traveltime difference between two complex waveforms needs to be determined at each iteration. A straight-forward approach automatically picks the onset of relevant arrivals, either directly or via a correlation of the two waveforms. If the waveforms are not very similar, however, this approach is problematic. We propose to measure the traveltime difference via a weighted norm of the correlation of the two waveforms. The weighted norm can be used directly as an optimisation criterion for waveform tomography. We illustrate this with a synthetic and real data example.

An inversion scheme that solves for reservoir pressure and fluid saturation changes from time-lapse pre-stack seismic attributes and post-stack seismic time-shifts is presented. It makes use of four equations expressing the changes in zero-offset and gradient reflectivities, compressional and shear waves time-shifts as functions of production induced changes in fluid properties. The method has been successfully tested on a realistic, synthetic reservoir, where seismic data have been modeled before and after 30 years production and water injection. Results show very accurate estimations if information about the vertically averaged reservoir porosity is available. The use of the gradient reflectivity equation causes biased estimations of real changes in saturation and strong leakage between the two different parameters. However, if the equation related to the S-wave time-shift can replace the gradient reflectivity equation, the inversion results may be very accurate. In cases where shear wave data might not be acquired, the approximation of the exact changes in this seismic attribute becomes more accurate if quadratic terms in relative changes of seismic properties are not neglected. The improved forward approximation in this attribute leads to inversion results characterized by weaker leakage and sharper discrimination between different fluid effects.

In a tight sand gas reservoir, due to its complicated geology and extremely low porosity and permeability, conventional inversion methods cannot always accomplish the task of characterizing reservoir distribution. In this paper, we apply the ray-impedance inversion on a tight-sand gas reservoir. We perform inversion on constant ray parameter profiles, with robust estimate of wavelet and stable result of reflectivity series to start with. Comparing the inverted ray impedance with the acoustic impedance, the elastic impedance and shear impedance from the commercial software, as well as these elastic parameters extracted from well logs, it shows that the frequency-dependent ray impedance is superior in identifying fracture zone and the characterization of reservoir distribution for this tight-sand gas reservoir.

AVO inversion is an essential and powerful interpretational tool. However, conventional AVO-inversion workflow in application to long-offset data will unlikely succeed, because it is limited to relatively plane interfaces, weak parameter contrasts and moderate incidence angles, where the offset dependence of the reflection amplitude can be described by linearized plane-wave reflection coefficients (PWRCs). It is also known that the PWRC is insensitive to the interface curvatures and breaks down at the near-critical offsets, where the head wave is generated, as well as at the post-critical offsets, where the reflected and head waves interfere. To avoid these inconsistencies, we exploit effective reflection coefficients (ERCs) that generalize PWRCs for curved interfaces and nonplane waves. Based on our previous results for plane interfaces, we generalize the improved approach to AVO inversion for the curved interfaces. Using a synthetic data example, we also show that the theoretical description of the actual reflection from a curved interface is directly applicable in AVO inversion.

This work presents a pilot study of the application of a seismic inversion technique for fracture density and fracture orientation in an area where available image log data revealed the presence of open fractures at a given well location. The inversion method is based on a singular value decomposition of azimuthal AVO data. This decomposition allows us to calculate seismic attributes which are linked to the fracture density of the fracture network through anisotropic rock physics modeling. The results reveal an area of high fracture density around the well where open fracture had been observed. The fracture orientations obtained are consistent with the interpreted image logs and outcrop studies in the area. The outcome of this pilot study is promising since the obtained results are consistent with existing image log data. Further testing and research on a larger area will be valuable to assess if the fracture characterization results are consistent with regional geological interpretations and other analysises of fracture networks in the reservoir.

Seismic acquisition is a trade-off between economical and quality considerations. Generally seismic data is recorded with large time delays between illuminating sources in order to avoid interference in time. The consequence is a poorly sampled shot domain. However, in the concept of blended acquisition, the time delay between sources is reduced significantly. Moreover, the sources may transmit encoded signals. Depending on the acquisition objectives, blended acquisition significantly improves the economics or quality or both by adding additional degrees of freedom in the acquisition design. By a deblending procedure, the individual source responses are retrieved. However, the deblended result contains residual noise due to the interference from other sources. The level of this noise depends on the choice of blending parameters. E.g., in the case of a simple code like linear phase encoding (i.e., applying time delays), it is larger than in the case of a more sophisticated code like transmitting sweeps as in vibroseis technology. In this paper an iterative approach is proposed for deblending, based on the estimation and subsequent adaptive subtraction of the interference noise. The type of coding that is applied is one of the factors that determines the required number of iterations.

Traditional data acquisition practice dictates the existence of sufficient time intervals between the firing of sequential impulsive sources in the field. However, much attention has been drawn recently to the possibility of shooting in an overlapping fashion. Numerous publications have addressed the issue from different scopes (de-noising, compressing, blind signal separation etc.) while others have defined the theoretical background. The term 'blending' has been introduced to describe this new trend in acquisition designs, the time-overlapping data acquisition. In turn, the term 'deblending' refers to an algorithm that recovers the data as if they were shot in the traditional way. Such an algorithm is presented in this paper, specially designed for the case of impulsive sources that fire with small time-delays. This algorithm is based on iterative interference estimation and subtraction. The key to signal extraction from blended data is the incoherency of the interference (as opposed to the coherency of the signal) accomplished by resorting the data into a different than the common source domain. The method is applied on a real marine dataset, where the blending process has been simulated numerically.