Reservoir characterization

3D Close-the-Loop Method Based on Probabilistic Seismic Inversion

The paper discusses an approach for predicting the lateral variation of net to gross (NTG) by use of 3D probabilistic seismic inversion.

jpt-2017-03-techsyn18913-seismic-iptc-18913-ms-2.jpg
Shell

The paper discusses an approach for predicting the lateral variation of net to gross (NTG) by use of 3D probabilistic seismic inversion. The goal is to define and understand the distribution of sands and shales on the basis of seismic reflection data. The modeling and inversion are supported by the good quality of seismic data. This study underpins the benefits of seismically constrained reservoir modeling. The use of probabilistic inversion to map geological features is a new insight in the applicability of this methodology.

Introduction

The study field is located in the Carnarvon Basin offshore western Australia. The field was appraised with one well, which has added incremental volumes to existing nearby discoveries. Seismically constrained reservoir-­model building creates models that are constrained by, and matched to, geological concepts and seismic amplitudes and travel times. The complete paper shows an approach using 3D probabilistic seismic inversion to predict the lateral variation in NTG over a proven discovery. This, in turn, can be used to update the static-model properties, improving on the first-pass static-model build. Additionally, the inversion will aid in the understanding of some poorly understood seismic expressions, especially in the south of the field. In this area, the authors noticed dim amplitudes on the amplitude maps extracted from the seismic data, where the static model predicts the presence of the reservoir. Field The field is located in the Carnarvon Basin. The discovery was appraised by one well, which penetrated a thin sand layer of interdistributary bay reservoir facies (referred to as Sand 1) and a thicker distributary channel facies (referred to as Sand 2). The overburden comprises basinal shales and marls.

The main steps in the 3D close-the-loop work flow are

  1. The rock-property-trend models are derived using the available vertical wells in or near the field. A “rock model” is a set of equations that honors well data and links the petrophysical properties such as NTG and porosity with rock acoustic properties.
  2. A qualitative 3D check-the-loop (CTL) step is carried out. This involves the initial fit of the static model with the seismic data by forward modeling the synthetic seismic computed from the initial static model before inversion. The quality-control step that follows aims to detect mismatches between seismic and synthetic data.
  3. Finally, constrain the reservoir model by use of the seismic data by carrying out 3D probabilistic inversion of the key uncertain rock property.

Qualitative 3D CTL

The objective of the qualitative 3D CTL is to find obvious mismatches between the synthetic data generated from the model and the actual seismic. The sources of these mismatches have to be identified, and first-pass ideas for possible corrections to the model to obtain a better match need to be made. Because this is merely a qualitative check before inversion, the loop is not yet closed with any model updates; this will be achieved later in the quantitative CTL step by use of model-based seismic inversion.

A static model with an initial 3D NTG grid was available for the field. The first-pass static model designed to estimate the gas initially in place is a simple constant-average-properties model derived from the well. It is expected that the seismic will give better control on the properties away from the well. The regional rock models were used to predict the elastic properties in the reservoir zone, in the overburden, and in the underburden shales. Fluid models for each fluid type (gas and water) were built with the input from the pressure/volume/temperature report. Forward-modeled prior synthetics were generated with a deterministic wavelet (derived from the seismic) and compared with the given seismic.

Prior Synthetic vs. Seismic. Fig. 1 shows the comparison of the root-­mean-square (RMS) amplitude maps extracted at Top Sand 2 reservoir (in a window of -5 to +10 milliseconds) for the near and far stack seismic and near and far prior synthetics, respectively. The RMS amplitude levels of the two maps show significant differences. Hence, there is a need to update the reservoir model properties, mainly the NTG. This will be achieved with a probabilistic seismic inversion in order to close the loop.

jpt-2017-03-3dclosefig1.jpg
Fig. 1—Comparison of RMS amplitude map between the near and far stack seismic and prior near and far synthetic at Top Sand 2.

 

3D Probabilistic Seismic Inversion

The qualitative analysis described in the complete paper highlights some key differences between modeled and measured data in terms of amplitude differences. The current model predicts high amplitudes in the south of the fault block, which is not supported by the seismic amplitudes. In the next step, the reservoir model is constrained with seismic data by carrying out a 3D probabilistic seismic inversion for NTG with a proprietary stochastic model-based elastic inversion algorithm. This technology directly inverts for static-model parameters, which sets it apart from traditional inversions for impedances.

The inversion takes a geological model with the prior reservoir properties (NTG or porosity) and the given uncertainties for a specific rock with a set of defined fluid properties and perturbs it iteratively until a realization is found that describes a good fit with the given seismic. The forward-modeled posterior synthetics need to fit the actual seismic character well enough within the specified signal/noise ratio as defined. Any model that does not satisfy the specified noise level is rejected, and a new realization is tested. The average uncertain property from the final models can be accepted as the most likely solution that gives the best fit with the seismic.

The acoustic properties in the input static model are derived using the rock-property trends. Regional rock models were available that related the p-­velocity and density to the porosity and shear velocity to p-velocity in the field. The main property for uncertainty was NTG in Sand 1 and Sand 2 of the reservoir.

In the current static model, the overburden comprises a wedge comprised of marl, which shows significant thinning from the north to the south of the field. It is crucial to model the overburden correctly such that it does not influence the NTG estimation in the channel sands below. Promise inversion was carried out on multiple scenarios in order to model and capture the overburden response correctly and understand the effect on the reservoir sands below.

Amplitude maps are extracted in the reservoir zone from the posterior synthetics of the best-fit model and compared with the seismic. The amplitude match has improved significantly after inversion compared with the prior synthetic match, as shown in Fig. 1. The average NTG produced by the probabilistic inversion is used to update the reservoir model and finally close the loop. The results of the probabilistic inversion do provide a number of model realizations; the standard deviations of these models can be used as a measure to build confidence regarding the inverted results and how well the uncertain parameter has been constrained by seismic inversion.

3D Inversion Results

The inversion results clearly showed the distribution of good sands and reduced the reserves in both north and south, immediately affecting developmental decisions. Additionally, a channelized feature can be recognized from the inversion, which could not be interpreted from seismic amplitudes alone. This helped in additional volume bookings.

Conclusions

The use of probabilistic inversion to delineate geological features such as higher-NTG channels is a new insight in the use of this methodology. It is accomplished by using the probabilistic inversion to predict the main static reservoir property (NTG) in a geologically constrained manner and then updating the existing static model with this information. The results of the probabilistic inversion helped to interpret an additional channelized feature that was included in the volumetric estimation and helped in additional volume bookings.

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper IPTC 18913, “3D Close the Loop Using Probabilistic Seismic Inversion for a Gas Field in the Carnarvon Basin, Australia,” by Shilpi Srivastava, Jeroen Goudswaard, Sito Busman, and Justin Ugbo, Shell, prepared for the 2016 International Petroleum Technology Conference, Bangkok, Thailand, 14–16 November. The paper has not been peer reviewed. Copyright 2016 International Petroleum Technology Conference. Reproduced by permission.