Second EAGE Conference on Seismic Inversion

Why is it that even when the seismic data we have are of good quality, Poisson ratio values derived from those data don’t necessarily tie the wells? This study addresses one of the factors.

Historically, using the exact Zoeppritz equations to invert reflection amplitudes was considered to be tedious. Instead, it was preferred to assume recorded amplitudes could be represented by linearized substitutes. This would enable the inversion process to rely upon standard matrix inversion routines.

The popular justification for this was to assume that the percentage change in elastic properties from layer to layer was small. In this study, fifty dual-layer shale/sand earth models were investigated. These models were based on bona fide downhole data. Exact forward modeling was used to generate seismic angle gathers. The gathers were then inverted for Poisson’s ratios of target sands.

Indeed, it was found that the “small percentage assumption” was a sufficient condition for success (when the percentage changes were “small enough”), but it was not always a necessary condition. This led to confusion concerning when inversion results could be trusted, and when they could not.

The situation screamed for a better inversion workflow. Indeed, nonlinear inversion consistently provided the correct Poisson ratio.

Simultaneous inversion combined with numerous classification techniques have been widely used since the methodology was firstly published. This paper presents a case study from the Otway Basin, in the South-east of Australia, where simultaneous inversion and subsequent petrophysical classification were applied to inform static modelling and well planning. We demonstrated the value of the workflow for well planning refinement by comparing results to a recently completed development well.

Seismic reservoir property estimation tools and processes output many things – some estimate elastic properties, some estimate reservoir properties and/or facies, and some output both. Some output 2D maps, some output 3D volumes. Some give associated uncertainty estimates, some provide multiple realizations, and some provide a single deterministic answer. All these products are created with the intent of informing seismic interpretation, well planning, reservoir modelling and resource estimation.

It is critical that seismic reservoir property estimation tools, and users of them and their products, are careful and accurate in the terminology they use to describe any output reservoir property estimates. As an example, terms such as net-to-gross are typically understood differently by different disciplines – the term “net” means something very specific to a petrophysicist but may be used as a loose or general descriptive term by a geologist. Misunderstandings around what any inversion product truly represents can ultimately lead to ill-informed business decisions.

Seismic reflectivity inversion using A-norm regularization produces sparse solutions by applying h-norm constraint. The fast iterative shrinkage-thresholding algorithm (FISTA) is one of the most effective method to solve /i-norm regularized inverse problem. A considerable number of iterations are commonly required in FISTA because its solution converges slowly towards nonzero sparse solution. To improve the convergence rate of FISTA, a modifying strategy was introduced into the traditional FISTA. When performing the soft-thresholding operator in FISTA, the thresholding value is adaptively adjusted by using the reciprocal of the solution in previous iteration as a weight to assign to the thresholding value of the current iteration. In this way, small variables in the solution produces big coefficients applied to the thresholding value, which causes small variables to quickly converge to zero. The adaptive FISTA has significantly improved accuracy and computational efficiency. It gained good results both in numerical and in field examples of /i-norm regularized seismic reflectivity inversion.

Marine reflection seismic data acquisition for inversion of near surface targets, which is restricted to the first 30–100m of the earth’s surface, is both a challenging and important task. Nevertheless, there are no comprehensive acquisition and quality control guidelines for this type of data set. In order to determine the requirements and boundary conditions for near surface seismic inversion, the soil property ranges of sediments in the North Sea and Baltic Sea have been established as a first step. Then the typically used seismic source signals have been analyzed. A range of synthetic models was generated and analyzed to develop optimum acquisition parameters. Source, receiver, positioning and recording parameters have been found to be most relevant. So, very exact corrections e.g. in terms of gun delay, positioning errors or source and receiver directivity have been identified to be substantial. Also, offsets greater than about three times the target depth and a channel spacing smaller than 2 m to avoid aliasing for high resolution signals have been found to be necessary. Further establishment of near surface seismic inversion in industrial and academic work flows has the great potential to reduce cost and enhance knowledge.

The preconditioning of input data into seismic inversion processes is extremely important. Without adequate preconditioning or with inappropriate preconditioning false positive and false negative inversion results can easily be produced.

The application of preconditioning to the seismic data is required for AVO and inversion work. Preconditioning is frequently needed regardless of the data type used (land/marine, single/multi component, 2D/3D/4D, recent/legacy) and the subsequent algorithm employed, although the precise workflow components may differ.

The authors note that, while modern and recently reprocessed seismic data is often vastly superior to legacy data, the requirement for data preconditioning remains if the user wishes to extract maximum value from the available data.

This case study uses examples for a recently acquired and processed, high quality, land seismic dataset, however, the observations and recommendations are applicable across the board.

In this case study the authors describe:

• the use of a number of preconditioning steps
• the purpose of the preconditioning
• common pitfalls in applying the preconditioning step
• QC methods for the preconditioning

Finally, the authors demonstrate the uplift of the preconditioning in the elastic domain using a deterministic model-based inversion algorithm. The authors consider this the best domain in which to QC the results of any preconditioning parameterization.

Note that the EAGE Inversion Classification table above refers to the case study example data to be used. The authors note that the methods shown are widely applicable with respect to data type and inversion algorithms

Basin model constrained Litho-Elastic (LE) amplitude versus angle (AVA) inversion is a new technology which integrates basin modelling with seismic inversion in frontier exploration area where little well control is available. Conventional LE inversion jointly solves for elastic parameters and lithology by integrating rock physics and lithology-dependent compaction trends as a function of depth. Basin model constrained LE inversion expands the LE inversion’s capabilities to handle multi-variant rock physics trends. For example, lithology dependent compaction trends of elastic parameters as a function of both effective stress and temperature. The method uses a 3D basin model as an input to the inversion to provide geologically consistent spatial constraints to the regional effective stress and temperature estimates. In this case study we review the additional capabilities we’ve incorporated in to the LE inversion of a frontier 3D seismic data set from offshore Nova Scotia.

We present a new target-enclosing full-waveform inversion method based on a new objective function. This objective function is based on the mismatch between wavefields reconstructed in the target domain with the convolution and correlation representation formulas, using data from the boundary of the target domain. The proposed method requires only kinematic knowledge of the subsurface model, particularly the overburden for redatuming, and no reconstruction of the model outside of the target area. In this sense it is truly local. We show a synthetic example with exactly redatumed wavefields and compare it with conventional FWI of surface data. The potential of the proposed method is demonstrated both in terms of the quality of target recovery and reduction in computational cost.

We present a case study for inversion of 4D seismic data on the Edvard Grieg field. We utilize a probabilistic framework which merges local interpretation into a structural context, and discuss the challenges met in a quantitative interpretation of 4D data in complex reservoirs. The Petro-elastic model (PEM) at Edvard Grieg reflects the complexity of the geology. Four distinctly different hydrocarbon bearing lithologies, yielding eleven relevant lithology fluid classes, and nine nonhealing hydrocarbon lithologies which influence the seismic response in the reservoir.

We argue that a model based on the same principles as 3D inversion but applied to 4D difference data provides a robust result which is easy to interpret. The approach respects constraints given by the petro-elastic response to saturation and pressure changes. A particular challenge in interpreting the 4D differences on Edvard Grieg is a strong elastic contrast directly above the reservoir which causes small alignment differences between base and monitor to produce amplitude errors of the same magnitude as the signal. The probabilistic methodology quantifies both the change and the uncertainty in the petro-elastic model and is suited for updating the reservoir model using ensemble-based data assimilation.

This paper describes a measure of comparing results derived from inversion methods that gives probabilities of different litho-fluid classes (LFCs) as output. The measure is used to evaluate how much the prior model is changed by the inversion process, to quantify ambiguity of classification in the inversion results, and to compare two different inversion runs having different LFCs in their prior models.

The method is applied to a North Sea dataset where well data was not available within the prospect. Accordingly, the LFCs were defined using data from wells about 40 km away. Given the degree of uncertainty this causes, inversion with modified LFCs were also run to investigate the sensitivity and robustness of conclusions drawn from the inversion.