may be used to split elasto-facies associations out into their component facies, at
the meso scale.
As an example, assume a sequence of thin alternating hard sand/soft shale
layers. Although the sand and shale elastic properties are quite distinct, the
layer thicknesses are below seismic resolution. These two facies can therefore be
combined into a ‘thin-bedded’ elasto-facies association. The seismic inversion
described below will estimate where this ‘thin-bedded’ elasto-facies association
is present in the subsurface, and the corresponding impedances will inform on
the net-to-gross (if impedances are high, there will be more hard sand, which
equates to a larger net-to-gross) even though individual sand and shale layers
cannot be imaged.
In the rest of this article the term ‘facies’ is used, for brevity, even though it
may represent elasto-facies associations – the context should make it clear.
Todate:model-basedsimultaneous inversion
Model-based inversion starts with a so-called low frequency model (LFM) of
impedances (e.g. compressional velocity Vp, shear velocity Vs and density p),
to account for the fact that seismic lacks low frequencies. Note that the quality
of this LFM is much more important than the actual inversion algorithm used.
There are many such algorithms available, but in essence the seismic response
is synthesised from the LFM and compared to the actual seismic. While the misfit
is not minimised, the model is repeatedly updated, seismic re-synthesised, and
re-compared. Once the misfit is minimised the process stops; the last model of
impedances is then the inversion result. Note that facies are not used.
The problem with the above process is in the construction of the LFM, the
most important input to this process. Vertically this model should have a gently
varying (typically hardening, because of compaction) profile of sand impedance
values where sand is present, and same for other facies (shale, limestone, etc).
But prior to the inversion the location of the various facies is unknown, so what
impedance value to assign in the LFM? In practice this results in compromised
impedance values, degrading the inversion result.
This is what is often observed when interpolation of well impedance profiles
along seismic horizons, the most ubiquitous form of LFM construction, is used.
One well may have sand at a particular stratigraphic age, and in another well
may have shale at that same level. This means that in between these two wells
end up with compromise neither-sand-nor-shale impedance values in the LFM,
and the subsequent seismic inversion cannot correct this incorrect low frequency
behaviour for the simple reason that seismic lacks these low frequencies.
Anewapproach– facies-awaremodels-based inversion
To improve the construction of the all-important LFM, multiple simple LFMs are
entered, one for each of the facies expected (i.e. the low frequency information
is over specified), and then the new seismic inversion system derives the LFM
actually used as part of the seismic inversion process.
The new inversion derives models of first impedances (from the seismic) and
then facies (from the impedances) at each iteration of the optimisation loop.
The facies result depends on the last impedances results, but here the focus is
on how the impedances results (inverted from the seismic data) depend on the
last facies result (of the previous iteration): for this step a LFM of impedances is
required; this is re-constructed at each iteration from the various per-facies LFMs
as follows: Start with an empty LFM. Now where the last, most up-to-date facies
model says there is sand, copy in the sand LFM, partially populating the LFM.
Repeat for the other facies, until the LFM is entirely populated.
The final LFM (of the last iteration) is clearly not a static input (as in
model-based simultaneous inversion to date) but instead is a seismically-driven
output of the new inversion system, which evidently uses facies intimately. Note
that the mechanics of this new algorithm is described in Kemper & Gunning
(2014).
2
THIS IS
EXTREM
Lubrication oil is the lifeblood
of machinery and a key health
indicator. Continuous monitor-
ing helps to increase uptime
and extend machinery life.
ATEX Zone 1
Continuous monitoring
Built for extremes
The Metallic Wear
Debris Sensor
