Integrating Well and Seismic 2006 AESC Jarvis

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Integrating Well and Seismic Data for Reservoir Characterization: Risks and Rewards

Kevin Jarvis
Fugro-Jason Australia BV kjarvis@fugro-jason.com

SUMMARY
Developing accurate reservoir models is a key objective of oil companies worldwide. A properly constrained reservoir model can be used to quantify hydrocarbons in place and to optimize hydrocarbon production. The evaluation of the reservoirs is typically achieved using a combination of seismic and well data. Each of these data represents imperfect measurements with a certain level of error. The manner in which these errors are handled affects the integration of the two data types and determines the quality of the final reservoir model. In this paper, I discuss one of the ways that well and seismic data can be combined to form a reservoir model. A logical workflow is proposed that consists of a systematic progression from well data analysis, to rock physics analysis, to seismic inversion and finally reservoir property estimation. The workflow cannot be followed blindly and each reservoir characterization problem must be handled in a unique manner. One of the key elements to the seismic inversion component is the development of a realistic low frequency model. A sophisticated low frequency model has a significant impact on the final reservoir model. Using a series of simple examples and case histories I demonstrate that significant benefits can be obtained from following a workflow that is tailored to deal with the uncertainties that impact on the results. I also show some of the pitfalls that have the potential to create incorrect or ambiguous results. The ultimate goal is better reservoir definition, better reservoir management and higher drilling success. Key words: Inversion, Data Integration, Rock Physics.

The recognition of errors in both seismic and well data is not a straightforward process. Every geologic environment has well and seismic data of different vintages and from different contractors. Some rocks are more prone to washouts than others and some geologic environments are more difficult to obtain high-quality seismic data in. To properly integrate the data the interpreter must recognize the good data and make some type of correction or mitigation for the poor data. A technique to achieve integration is through seismic inversion. The inversion process uses both seismic and well data and generates results constrained by both data types.

METHODOLOGY
There are four main elements to a good inversion workflow: well log conditioning, seismic quality control and conditioning (if necessary), wavelet extraction and low frequency modelling. These elements are illustrated in Figure 1. Each of these elements will be discussed separately with particular attention paid to the role of the interpreter in making key decisions and parameter selection.

Figure 1. Inversion characterization Well Log Conditioning

workflow

for

reservoir

INTRODUCTION
There is an increasing need for reservoir characterization technology within the oil and gas industry. The need is mostly driven by economic realities: if reservoirs can be defined better using available technology then the end result is higher drilling success and fewer development wells. Reservoir characterization technology requires the integration of all available subsurface data. These data include well logs, seismic trace and velocity data. The data are the result of measurements carried out by sophisticated instrumentation and processed using highly developed software. This technology has advanced over the years and the accuracy has greatly improved. However, the results are still subject to error and these errors must be properly dealt with in order to use the data for reservoir characterization.

Well logs are a result of a physical measurement of the earths properties taken within the confined space of a borehole. The measurements are subject to borehole irregularities and to the elapse of time between drilling and logging of the hole. The primary goal in processing well log data is to obtain consistent and accurate logs from well-to-well. In addition, the logs must represent the rocks as seen by the seismic. Some type of dispersion correction should be applied to account for the differences in frequencies between logging tools and surface seismic reflection data and fluid substitutions should be performed if invasion effects are present. A technique that has proven to be very powerful for consistency checking is the development of a rock physics model. The rock physics model can take many forms from simple empirical relationships between properties to more sophisticated rock physics models such as inclusion-based

AESC2006, Melbourne, Australia.

models (e.g. Xu-White, 1995). The inclusion-based models break the rock down to fundamental components such as clay, quartz and calcite with pore space filled with brine, gas and oil and then reconstruct the rock based on the proportions of the components. These proportions can be estimated using petrophysical analysis techniques. Ideally the rock physics model will be robust enough that it can be used for fluid substitutions and shear log synthesis. The consistency checking from the rock physics model comes from the match of the modelled and measured data using the same model parameters for different wells. The interpreters should be aware of the type of rock physics model being used and the model parameters assigned. A standard quality control plot is to overlay the measured and modelled log curves and to generate statistics such as the cross correlation between modelled and measured and standard deviation of the differences. Model parameters such as pore aspect ratio and clay density should make sense for a given geologic environment. The goal of the rock physics modelling is to create a consistent set of logs. The final consistency of all well logs can be checked by a series of zonal-specific crossplots and histograms. An example of consistent data output from a rock physics model is shown in Figure 2. The lower crossplot shows clear lithology and fluid trends from a number of wells that are obscured in the uncorrected data due to invasion, production and washout effects. A final check of the conditioned well logs is the match of synthetics to the seismic data.

can be compensated by the wavelets and post-stack time alignment, but in general these are best dealt with by careful processing. Some errors in seismic data can be mathematically quantified (e.g. signal to noise ratio) and these should certainly be taken into consideration when using the data for inversion. Other errors (e.g. multiples) are not readily quantified and every effort should be made to remove them during processing and to minimize their impact on the inversion results. Wavelets The wavelets link the seismic and well data. They are often derived by inverting seismic data with the well data as the primary constraint. They are sensitive to errors in both the elastic well logs (sonic and density) and errors in the seismic data. When multiple well ties exist the wavelet extraction process can be carried out at each well individually or collectively to obtain a multi-well optimized wavelet. A common assumption for full-stack inversions is that the seismic is zero-offset. No multi-channel seismic dataset is truly zero-offset and angle-dependent effects are almost always present. The angle ranges represented by the acquisition geometry should be honoured when inverting seismic data for a wavelet. Do wavelets vary spatially? This assumption is often made because wavelets extracted at different wells do not have the same amplitude and/or phase spectra. An alternative assumption is that the wavelets vary little but the well data are inconsistent. If no obvious amplitude spectral variations occur within the seismic data (e.g. shallow attenuative channel deposits) the assumption of a constant wavelet derived using multiple wells is often the most reasonable. The interpreter should be aware of the wells that are used for wavelet estimation and quality control the synthetic-to-seismic ties using the extracted wavelets (both individual and multiwell). Wells that exhibit poor synthetic-to-seismic matches should not be used for wavelet extractions and there should be a reason for the poor synthetic tie (older log data, washed out borehole, etc). The ideal wavelet is compact with minimal side-lobe energy and wavelets extracted from multiple angle stacks should exhibit reasonable variations with increasing angle and not show unusually large changes in spectral amplitude or phase. The residuals between the seismic used for the wavelet and the resultant synthetic should also be low. How accurate must the wavelet phase be estimated? This is a difficult question to answer but a simple example as shown in Figure 3 may be illustrative. Figure 3 compares the simultaneous inversion of synthetic data (created using a known wavelet and low frequency model) with both the proper wavelets and wavelets shifted 30 degrees for the near and far stacks (the mid-stack wavelet is correct). The inversion outputs of P-impedance and Vp/Vs are very close and certainly the sand channel interpretation would not be that different on the two results. This example is one indication that wavelet phase does not have to be perfect to get usable inversion results. Low Frequency Modelling The low frequency model in a constrained sparse-spike inversion workflow (CSSI) represents the model, which fills the frequency gap from zero to the low end of the seismic

Figure 2. Crossplots of P-Velocity and S-Velocity showing well data before (top) and after (bottom) rock physics modelling. Seismic Quality Control No seismic dataset is perfect. Every attempt is made during acquisition and processing to choose optimum parameters. Relative amplitudes must be preserved during processing both vertically and spatially and stacking velocities must be optimized to properly image the geology. Complex geology requires proper migration, but so do amplitude anomalies from discontinuous bodies. Some elements of the seismic processing

bandwidth. This gap must be filled in with data which matches the well logs and is representative of the geological environment, in order to obtain the sharpest impedance image (with reduced inversion side-lobes) and for quantitative inversion results (e.g. reliable porosities or fluid prediction). The low frequency model is built using a combination of all available data, typically seismic stacking velocities, elastic well data and time horizons from seismic interpretation.

Detail is required in the definition of the lateral extent of discontinuous bodies. A discontinuous body of primary interest is the reservoir itself. Many reservoir rocks, particularly those with hydrocarbons, have distinct elastic properties, compared to non-reservoir rocks and this must be addressed in the low frequency model building process. One possible scenario is to interpret the first-pass inversion results as part of an iterative inversion workflow. A simple low frequency model can be used with the goal of outputting bandlimited inversion products (e.g. bandpass P-Impedance and bandpass Vp/Vs). Changes in both lithology and fluid can often be qualitatively interpreted on bandlimited products. The goal of the first-pass interpretation is to more precisely locate lithologic boundaries and to map out the extent of discontinuous bodies. This detailed interpretation then serves as the basis for properly interpolating the well data and creating an improved low frequency model for reliable full bandwidth inversion results. Figure 4 compares inversions, which use different low frequency models; an iterative inversion workflow with property substitution and simple well interpolation. The more sophisticated low frequency model gives results that closely match the input model, whereas the simple low frequency model gives a very poor subsurface image with potentially misleading results. The interpreter should be involved with all aspects of the low frequency model. The horizons are the most obvious inputs that are provided by the interpreter but the modification of horizons on bandlimited inversion results and the definition of potential reservoir bodies both require interpretive input. The final low frequency model should be quality controlled in both section and map view (time or horizon slices). Section view reveals the implied stratigraphic relationships and map view can be used to identify unusual features such as bullseyes around wells. The low frequency model can also be quality controlled after the completion of any full bandwidth inversion by simply filtering back the inversion output to remove all seismic frequencies. This is generally good practice in the quality control of inversions where statistical techniques are used to populate the model (often as part of the inversion process) and it may be the only way to identify potential bias in the low frequency model.

Figure 3. Outputs from simultaneous inversion of three synthetic stacks (with noise) using the correct wavelets on the left and an inversion with near and far stack wavelets shifted by plus and minus 30 degrees (respectively) on the right. P-impedance is on top and Vp/Vs below. The most obvious assumption is that the amplitude spectrum gap must be filled in. More importantly, the phase spectrum must also be addressed. Without addressing the phase spectral gap, the inversion results will be sub-optimum. The phase spectrum is defined by the lateral continuity of geologic horizons. This is often approximated by the interpretation of the seismic reflection events suitably tied to consistent geologic markers at the different well locations. Assuming different stratigraphic relationships (e.g. onlap, offlap or conformable) provide additional sophistication. The horizon interpretation represents the boundaries along which the well properties are interpolated and/or extrapolated. The end result must follow the geology. This does not necessarily follow the seismic. Noise that is present in all seismic data can introduce trace-to-trace variations that should not be honoured by the horizons. Rapid changes from one trace to another imply discontinuities that will result in phase spectrum variations. The goal is to replicate the local geology. Horizons should be smooth where the geology is expected to be slowly changing and should honour geologic discontinuities (e.g. faults) where they exist. The general horizon interpretation does not have to be detailed vertically as long as the stratigraphy can be reasonably approximated. The most important consideration is the consistency of the intersections of the horizons to the timeconverted well data. In some areas the geology that must be part of the low frequency model is not present in the wells either due to erosion, facies change or non-deposition. Careful consideration must be given as to how and with what data these zones are populated.

CONCLUSIONS
The application of seismic inversion technology has its rewards: better reservoir definition, improved resource estimates and better reservoir management. A well-planned workflow will address the quality control and conditioning of the input data, the extraction of reliable wavelets and the creation of a properly constrained low frequency model. The errors in the input data cannot be simply ignored and must be addressed as part of the workflow. The goal is to replicate the geological variations using the well control and to acknowledge the physics of the seismic experiment in the inversion processing.

REFERENCES
Xu, S., and White, R.E., 1995, A new velocity model for claysand mixtures.: Geophysical Prospecting 43, 91-118.

Figure 4. Synthetic model of P-impedance (top) and Vp/Vs (bottom) shown in left panels that are used to create three synthetic angle stacks with noise (0-35 degrees). The model is meant to represent gas sands isolated in shale. The input models are created at an accuracy of 1ms. The middle panels show the output of simultaneous inversion of the three stacks using an iterative inversion where the sand body properties are replaced after a first-pass simple trend (shale-only) inversion followed by substitution of gas-sand properties. The right panels show the simultaneous inversion (one-pass) outputs using a low frequency model created from one well (Synth2). The extent of the larger channel is not properly defined and the smaller channel does not show the correct properties. The only difference in the workflows is the sophistication of the low frequency model.

AESC2006, Melbourne, Australia.

You might also like