You are on page 1of 5

The quest for the low frequencies in pre-stack inversion

William R. PILLET* (TOTAL E&P CSTJF Pau), Emmanuelle BRECHET (TOTAL E&P Angola), Peter MESDAG (Fugro-Jason Netherlands BV), Marina FEROCI (Fugro-Jason UK)
Summary The pre-stack inversion process is often utilized to derive P impedance and Poisson ratio volumes used for reservoir characterization. When using high resolution (HR) seismic data, which are specifically designed for detailed reservoir delineation and characterization, the gap to bridge by the low frequency model is much greater (0-16Hz) than for conventional seismic acquisition (typically 0-7Hz). To overcome this problem linked to the use of HR data, postprocessing of the seismic data was not fully satisfactory. Hence, we designed and applied an iterative inversion workflow incorporating an additional conventional seismic data set. Introduction Initial inversion results For reservoir characterization purpose, TOTAL S.A often uses pre-stack inversion to derive absolute and relative values of P impedance and Poisson ratio to be correlated with petrophysical parameters. The absolute elastic parameters are the output from a full band width simultaneous inversion process and contain all frequencies from zero Hertz to the highest seismic frequency. The relative parameters are derived by low cut filtering the absolute volumes to remove the lowest frequencies coming from the well interpolation process (the low-frequency model). The quality of an inversion is dependent on the seismic data (its bandwidth, its signal/noise ratio, its processing), the wavelet and the low frequency model. Typically such a model is built by using well logs, interpreted seismic horizons and seismic velocities. On several occasions, this process applied to high resolution seismic data was affected by the lack of low frequencies of this kind of data set. During a recent project of inversion in West Africa, the lack of low frequencies was such that thick sands (25 m) could not be characterized properly. Hence we resorted to using a second conventional seismic data set and set up an iterative workflow to derive the best possible low frequency model for the HR inversion but also to bring most of the low frequencies from both data sets into the final result. The whole workflow was designed by using the FugroJason WORKBENCH which allows a great flexibility in integrating various datasets. Data used The main seismic data set consists of high resolution (HR) pre-stack time migrated data, divided into three angle sub The first pre-stack inversion run on this field in 2006 using only the HR sub-stacks and the four wells produced the relative P impedance and S impedance sections shown in Figure 1.
Relative IP A sand

stacks. The frequency bandwidth of each stack is respectively: 25-110Hz for the Near, 22-100Hz for the Mid, 17-80Hz for the Far. Good quality measured Vp, Vs and density logs from four vertical wells were used to derive the wavelets for each seismic stack. The same well logs were also used for the building of the low frequency models, guided by 4 regional horizons to make the structural background. The second seismic dataset was an older conventional seismic full stack, the bandwidth of which is between 6 and 80 Hz. We will present successively the initial pre-stack inversion results of the HR data, those of the post-processed HR and finally the results of the inversion combining the conventional and HR surveys.

Well 1 GR Relative IS A sand

Well 1 GR

Figure 1: Relative P impedance and S impedance sections at well 1

The absolute values constrained by the low frequency impedance logs from the nearby well properly show the thick A sand. Since the low frequency model is often poorly constrained (only 4 wells over few hundreds of km), the relative mode is preferably used for the computation of Vclay or lithological cubes. On the relative impedance sections, the thick sand is poorly imaged and side lobes are evident in these data.

SEG/San Antonio 2007 Annual Meeting

1810

The quest for low frequencies in pre-stack inversion


Post-processing In such a case, one option is the re-processing of the seismic data to enlarge the frequency band on the low side. In this case, the original processing was recent enough not to be questioned and we chose to only perform postprocessing. The three raw sub-stacks were fed through a deghosting process to improve the low frequency spectrum. The deghosting process consists in applying an operator converting the source/cable ghost signature into a source ghost signature only. A side effect of this process is a relative reduction of the higher frequencies. So spectral whitening was applied to slightly increase the high frequencies. Figure 2 displays a comparison of a section around well 2 from the full stack volume together with their respective amplitude spectrum.
HR Full stack

Correlation

Well 2 tie to HR seismic data

Well 2 tie to post-processed HR seismic data

Figure 3: Comparison of the Near seismic to well tie at well 2, from the original HR data and the post-processed one.

The relative P impedance sections resulting from this inversion are compared to the result of the initial inversion of the HR data in figure 4, around well-2
B sand
Relative IP from HR

Well 2 HR Post-processed Full stack


B sand

Thin sands

B sand
Well 2 Relative IP from post-processed HR

Well 2

Thin sands

Figure 2: Comparison of a full-stack section before (top) and after post-processing (bottom) and the corresponding amplitude spectra

B sand

Inversion of post-processed HR data


Well 2

The same pre-stack inversion workflow was used to invert the post-processed sub stacks. The inversion wavelets were re-extracted and their length was slightly increased to better capture the lower frequencies. The seismic to well-ties are still of very good quality, meaning that the post-processing did not increase noise at the cost of useful signal. Still from figure 3 one can see that some details dont match any better than in the initial well tie at well 2 The extracted wavelets also are reasonably similar to the original ones, except for their low frequency content. The inversion could be run with a merge frequency (frequency at which inverted impedances will be merged with the low frequency model) of 6Hz compared to the 16Hz in the initial HR inversion.

Figure 4: Relative P impedance sections from original HR and from post-processed HR data

The inverted sections show a dramatic increase in the low frequency content, even on the band-limited IP since its starts now around 6Hz. The thick B sand is much better imaged. But according to the reservoir geophysicist and geologist this has been achieved at the cost of the high frequencies. Indeed the thin sands indicated by an arrow are less well defined. The dual iterative inversion workflow The outcome of the test described in the previous section is positive but to a limited extent only. Hence we attempted to incorporate an older exploration 3D data set, of lower resolution which had been reprocessed in 2005. This data set was acquired with a different acquisition direction, a

SEG/San Antonio 2007 Annual Meeting

1811

The quest for low frequencies in pre-stack inversion


longer streamer and processed with a different binning. Since the aim here is to use only its low frequency in the whole process, we felt re-gridding was acceptable. In order to limit the cost of using a second data set, we decided to get the low frequencies from this seismic data through a post-stack acoustic inversion of a full stack. Additionally, we improved the structural aspect of the low frequency model. As mentioned earlier, the first inversion of the HR data low frequency model was built with 4 regional horizons. Such a process tends to distribute sand impedances found at the wells throughout the whole volume. This does not reflect the true geology of sand distributed in channel complexes. Hence an iterative workflow was applied, where a first acoustic inversion using the Invertrace+ module of the Workbench uses a shale only model as LF model (after a proper re-interpolation and a binning in acquisition direction of the old vintage survey) and a specific wavelet was extracted from this seismic dataset.
Conventional full stack 4 wells Shale only model 4 regional horizons

sand bodies. The final sand/shale model turns out to be much more realistic than the initial one based on regional horizons only. The succession of all the necessary operations is represented in the workflow described in figure 5. From these 2 post-stack inversions, an acoustic impedance volume is obtained in which the 6-16Hz band is now derived from seismic data. Once filtered by a high cut filter at 16Hz it will provide the LF model for the HR pre-stack inversion. The other important step is to derive Is and density models which are compatible. For this we have used two different IP to IS and IP to density, relationships for sands and shales. These relationships were derived from well log data and are shown in figure 6.
IP to IS mapping

Shale
Post-stack inversion (Invertrace +) Merge frequency 6Hz

Sand

IP to Density mapping
AI volume Sand detection by AI Threshold from Vsh / AI crossplot Shale only model Conventional full stack Sand only model Shale + sand bodies model Sand bodies outline

Shale
4 wells

Sand

Post-stack inversion (Invertrace +) Merge frequency 6Hz AI volume Sand detection by AI Threshold from Vsh / AI crossplot Shale only model Sand only model Final LF AI model Refined Sand bodies outline

Figure 6: Mapping IP into IS and density for sand and shales Cross-plots and relationships used

The low frequency models filtered at 16Hz are shown in figure 7 and can be compared to the initial model used for the first inversion
IP IS IP initial model IS initial model

Mapping of Acoustic impedance to S impedance and Density IP mod. IS mod. Dens. mod.

Binning & Interpolation to HR grid Pre-stack inversion (Rocktrace) Merge frequency 16Hz Final IP vol. Final IS vol. Dens. Vol.

IP final model

IS final model

Figure 5: Workflow of the iterative inversions

From this AI volume are extracted 3D sand bodies by using an AI threshold. The 3D bodies are then populated with sand impedances from the well log data and inserted within the shale background volume. A second post-stack inversion of the same seismic dataset, starting from this improved AI model allows refining the definition of the

Figure 7: Comparison of IP and IS LF models from post-stack inversion and from the very first pre-stack inversion model around well 1

SEG/San Antonio 2007 Annual Meeting

1812

The quest for low frequencies in pre-stack inversion

Final inversion results The pre-stack inversion of the HR sub-stacks using these new low frequency models leads to the sections observed in figure 8 and 9. These relative impedance sections features significantly different details compared to the initial sections. .
A sand R e la ti v e I P f r o m o r ig in a l p r e - s t a c k in v e r s io n

A true pre-stack inversion of the conventional data may theoretically lead to a better final product in terms of LF model but it would require much more work in terms of sub-stack alignment and wavelet extraction, hence more turn-around time. This will be tested shortly. Conclusions The use of two different seismic data sets (a conventional full stack and a set of high resolution sub-stacks) together with an iterative inversion sequence has lead to the extraction of the maximum frequency bandwidth, particularly on the low frequency side, from the seismic data. This has allowed a better imaging of thick sands without affecting the thin sand detection. This iterative inversion turned out to produce superior imaging capability compared to a less expensive post-processing of the HR data. Such a process can potentially give a second life to older seismic data set, as a complement to newer HR data used daily for fine reservoir interpretation. Acknowledgements We want to thank TOTAL SA for the permission to present this work.

W e ll 1 R e la t iv e IP f r o m n e w w o r k flo w

A sand

W e ll 1

Figure 8: Relative IP sections at well 1: comparison of initial inversion and final iterative inversion result

The A sand in particular is imaged better, even on relative impedances. Even if these differences may seem minor to the people not familiar with this specific geology, our geophysicist and geologist feel more comfortable with this new data set to determine new well locations to tap the thick reservoirs. On Is sections, differences are even more striking. Combining IP and IS volumes for lithological discrimination will be lead to more geological results and better prediction of thick sands at blind wells. .
A sand R e la tive IS fro m o rig in a l p re -s ta c k in ve rs io n

W e ll 1 R e la tive IS fro m n e w w o rk flo w

A sand

W e ll 1

Figure 9: Relative IS sections at well 1: comparison of initial inversion and final iterative inversion result

SEG/San Antonio 2007 Annual Meeting

1813

EDITED REFERENCES Note: This reference list is a copy-edited version of the reference list submitted by the author. Reference lists for the 2007 SEG Technical Program Expanded Abstracts have been copy edited so that references provided with the online metadata for each paper will achieve a high degree of linking to cited sources that appear on the Web. REFERENCES Dubucq, D., S. Busman, and P. Van Riel, 2001, Turbidite reservoir characterization: multi-offset stack inversion for reservoir delineation and porosity estimation: A Gulf of Guinea example: 71st Annual International Meeting, SEG, Expanded Abstracts, 609612. Mesdag, P. R., J. Fugro, F. Marina, L. Barens, J. F. Pau, P. H. Pau, and W. Pillet , 2007, Full bandwidth inversion for time lapse reservoir characterization on the Girassol Field: 69th Annual Conference and Exhibition, EAGE, Extended Abstracts, A032.

SEG/San Antonio 2007 Annual Meeting

1814

You might also like