Introducing Ji-Fi - Joint Impedance & Facies Inversion
M.A.C. Kemper (Ikon Science), K. Waters (Ikon Science), A. Somoza (Ikon Science), A. Pelham (Tullow Oil
Plc) & A. Kozhenkov* (Ikon Science)
Introduction
In this paper we introduce Ji-Fi, a new Joint Impedance & Facies Inversion system, which gives a
significant increase in quality over model-based Simultaneous Inversion, because it incorporates the
correct physics! We first review Simultaneous Inversion, then introduce Ji-Fi and compare one
against the other, first with a wedge model and then using a case study.
Simultaneous Inversion review:
There are many somewhat different ‘flavours’ of model-based Simultaneous Inversion, but at a high
level they all more or less follow the workflow outlined in Fig. 1.
Figure 1 High level model-based Simultaneous Inversion workflow.
Whilst this can be solved by simple least-squares optimization, two well-known shortcomings to
Simultaneous Inversion are:
The LFBM’s (AIo, SIo and ρo in Fig. 1) are difficult to determine accurately.
When you are away from seismic events the Impedances (AI, SI and ρ in this case) are forced
back to the LFBM values.
Note that seismic events are typically generated when you go from one Facies (a discrete quantity!) to
another. In other words, seismic inversion is a mixed discrete and continuous problem, which
Simultaneous Inversion attempts to solve only with continuous properties (i.e. Impedances).
Joint Impedance & Facies Inversion (Ji-Fi):
To overcome these issues we have developed Ji-Fi where the AIo, SIo and ρo LFBM’s are specified
individually for all Facies expected! This sounds like more work, but deriving LFBM’s for, say, Sand
and Shale individually is much easier than deriving AI, SI and Rho Depth Trends for Sand-and-Shale
combined as required in Simultaneous Inversion (see i. in the previous section).
6th Saint Petersburg International Conference & Exhibition – Geosciences – Investing in the Future
Saint Petersburg, Russia, 7-10 April 2014
By providing the LFBM’s for each expected Facies individually, we are able to invert for Facies and
for Impedances-per-Facies. So Ji-Fi is a mixed discrete and continuous inversion: the physics of the
inversion problem is captured correctly!
Wedge model
In Fig. 2a we present an Impedance sand wedge (“the truth”), with the (synthetic) seismic response
overlain. The purpose of seismic inversion is to recover the former from the latter as accurately as
possible.
Fig. 2b shows the result from Simultaneous Inversion. Where the sand wedge is thick we see that in
the middle the Impedance trends back to the LFBM value, one of the shortcomings mentioned earlier.
Also note that where the sand wedge is thin, the sand Impedance value is over-estimated.
In Fig. 2c the Ji-Fi result is presented. The impedance value in the middle of the thick sand wedge is
not some compromise LFBM Impedance value (as in Fig. 2b), but is the correct sand LFBM
Impedance value (and indeed outside of the sand wedge, we observe the correct shale LFBM
Impedance value). Where the sand wedge is thin, we do not observe any overshooting.
Figure 2 (a) Wedge AI and synthetic seismic superimposed (top); (b) Simultaneous Inversion AI; (c)
Ji-Fi AI.
Case study:
We applied both Simultaneous Inversion followed by Bayesian Classification and Ji-Fi to a field
offshore Western Australia.
In Fig. 3 we show a section and horizon slice through the Facies model obtained by Bayesian
Classification on the Impedances obtained by Simultaneous Inversion, a two-step method. In Fig. 4
we show the corresponding section and horizon slice through the Facies model obtained directly by Ji-
Fi.
6th Saint Petersburg International Conference & Exhibition – Geosciences – Investing in the Future
Saint Petersburg, Russia, 7-10 April 2014
Figure 3 A section (Left) and Horizon slice (Right) through the Facies cube obtained by Bayesian
Classification of Impedances from Simultaneous Inversion of a field offshore of Western Australian
(Shale is purple; Water-sand is blue; Oil-sand is green and Gas-sand is red).
Figure 4 A section (Left) and Horizon slice (Right) through the Facies cube of a Ji-Fi inversion of a
field offshore of Western Australian (Shale is purple; Water-sand is blue; Oil-sand is green and Gas-
sand is red).
We observe that overall the oil and gas bearing reservoirs are found in both inversion techniques, but
that the Simultaneous Inversion result is much noisier and discontinuous (best observed by comparing
the map views). Both techniques replicate what is observed in the wells except for the northernmost
one, where oil is predicted, but where water is encountered in the well (fault?). The fault pattern
comes out much more strongly in the Ji-Fi results.
Net Sand can be readily determined from the Facies volumes, by simply counting the water, oil and
gas bearing Facies, and the Simultaneous Inversion and Ji-Fi Net Sand results are compared in Fig. 5
below.
6th Saint Petersburg International Conference & Exhibition – Geosciences – Investing in the Future
Saint Petersburg, Russia, 7-10 April 2014
Figure 5 Net sand obtained from Simultaneous Inversion (Left) and Ji-Fi (Right) obtained from the
relevant Facies cube of a field offshore of Western Australian
Note that the Ji-Fi result is much more geological and detailed
Discussion:
In this abstract we have explained that seismic inversion is a mixed discrete/continuous problem,
which to date is solved continuously only in the great majority of algorithms. Ji-Fi uses the correct
physics, i.e. inverts jointly for Facies (discrete quantities) and for Impedances-per-Facies (continuous
quantities). The result is a hugely significant increase in inversion quality in our opinion.
Ji-Fi is actually much easier to run than Simultaneous Inversion, as you are not at the mercy of
endless LFBM building!
So Ji-Fi is better and easier; the price you pay is in increased run-times (but reduced set-up and QC
times re-pay that handsomely).
Ji-Fi is a new inversion technology, and we are rapidly applying it to various datasets, some of which
we will show in the paper (not enough space in this abstract). We anticipate a big uplift in
Unconventional plays, as Facies (ductile vs. brittle for Shale gas, tight vs. higher-porosity for Tight
Oil, cold vs. hot for Heavy Oil) can be inverted for more accurately than with conventional techniques
such as Simultaneous Inversion followed by Bayesian Classification.
Furthermore, a more reliable, Ji-Fi generated Facies image may have a positive impact on a number of
subsurface workflows, such a pressure prediction, de-risking of deep offshore prospects, more
accurate geological modelling in terms of flow units etc.
Acknowledgement:
We thank Tullow Oil Plc for sponsoring this Ji-Fi project, and CSIRO for their cooperation
6th Saint Petersburg International Conference & Exhibition – Geosciences – Investing in the Future
Saint Petersburg, Russia, 7-10 April 2014
