Here - Delft Inversion

Th E106 08
Non-linear Full Wavefield Inversion Applied to
Carboniferous Reservoirs in the Wingate Gas
Field (SNS, Offshore UK)
A. Gisolf (Delft Inversion), R. Huis in't Veld (Argo Geological Consultants),
P. Haffinger (Delft Inversion), C. Hanitzsch (Wintershall Holding GmbH), P.
Doulgeris (Delft Inversion) & P.C.H. Veeken* (Wintershall Holding GmbH)
A high resolution full wavefield inversion (FWI-res) has been conducted over sub-cubes from the Wingate
3D seismic survey. Three wells provide calibration to the prestack elastic inversion of a deep-seated
Carboniferous reservoir section. The non-linear FWI-res technique uses higher order scattering (or
multiple energy) to constrain the output. The inversion is performed on migrated seismic data, including a
demigration step over the target interval. This object oriented approach is possible in the plane wave (Taup) domain, when assuming a layered medium immediately below the target location. The FWI-res
inversion scheme requires only limited user input: cleaned-up Common Image Gathers, top horizon to
demigrate the data and a velocity model. Data pre-conditioning is carried out by dip filtering in the timespace domain. This new trace-by-trace inversion is robust and satisfactory results are obtained even in
poor data zones.
76th EAGE Conference & Exhibition 2014
Amsterdam RAI, The Netherlands, 16-19 June 2014
A high resolution Full Wavefield Inversion (FWI-res) test over a limited target zone in the
Carboniferous Wingate gas field, located in the UK sector of the North Sea, is presented in this paper.
Purpose of the inversion was to reveal in greater detail elastic reservoir properties in small test areas
between and around three calibration wells. The fully elastic, non-linear, target-oriented inversion
scheme has been demonstrated effective on a synthetic data model constructed from Book Cliff
outcrops in the US (Tetyukhina et al. 2013). Here the same method is tested on real seismic data
within an operational context. Unlike conventional linear inversion processes, this FWI-res scheme
uses internal multiple scattering and non-linearity of reflection amplitudes. Therefore it has more
robustness against noise and arrivals not stemming from the target area.
Geological setting
The Wingate gas field (blocks 44/23, 44/24 and 44/19, UK offshore) was discovered in 2008 by well
44/24b-7 operated by Wintershall. The structure consists of several fault blocks containing northward
dipping fluvial reservoirs of Carboniferous age. The Late Westphalian A to Westphalian D sediments
are unconformably overlain by sealing Silverpit claystones of Late Permian age. The complex
Westphalian subcrop pattern, with intra-formational seals and intersecting faults, results a
compartmentalised field with several poorly to non-communicating segments. Sometimes different
free water levels are encountered in the fault blocks. Systematic well correlation using palynology,
chemostrat and seismic data, has resulted in a detailed sequence stratigraphic framework. With this
approach, the position of prolific reservoir intervals can be predicted with a reasonable degree of
confidence. The application of advanced inversion techniques is essential to improve the reservoir
model further, as it reduces the uncertainties in spatial distribution of reservoir properties and pore fill.
Non-linear inversion
The relationship between the measured seismic data and the property representation is non-linear.
This is evident as: 1) internal multiple scattering, 2) reflection amplitudes that are non-linear in the
property contrasts across interfaces and 3) travel times in the real medium that differ from the travel
times in the background medium. In the non-linear inversion algorithm all these effects are captured
by a mathematical formulation known as 'the scattering integral' (Fokkema & van der Berg, 1993).
Figure 1: Flowchart for non-linear inversion. The input data is redatumed, or demigrated, to a
target boundary at the top of the objective sequence (see also Figure 3).
The data-flow for the non-linear full wavefield inversion is shown in Figure 1. The mathematical
concept of multiple scattering is introduced in this scheme by the iterations in the outer loop, where a
higher order of scattering in the data is added in every extra iteration step, based on the current best
estimate of the properties (Gisolf & van den Berg 2010). This updating can go up to the 10th or 15th
order. The linear inversions invert every time the same data, while using a kernel that is based on the
current best estimate of the total wave field in the object under calculation. The properties are defined
as contrasts against their background values. In the FWI-res elastic inversion we invert for the
contrast in compressibility and the contrast in shear compliance, which is the inverse of the shear
76th EAGE Conference & Exhibition 2014
Amsterdam RAI, The Netherlands, 16-19 June 2014
modulus. In this application test the density has not been inverted for, but it is kept at the background
value to circumvent possible data quality issues (Figure 2). The background trends for the three
properties density, compressibility and shear compliance, are obtained from the well logs by simple
filtering. The vertical spatial bandwidth present in these backgrounds equates to a temporal bandwidth
of 0 - 3 Hz. Over this bandwidth properties can safely be interpolated between wells. From the full
logs a synthetic slowness gather was calculated with the ‘Kennett method’ (Kennet 1983). This gather
was inverted to see what can be retrieved from this type of data. A good fit between the actual well
properties and the inversion result from the well synthetics was obtained.
Well 44/23b-13
EPP synthetic data for 10 different horizontal slownesses
(ray-parameters), computed from the logs of the well
44/23b-13. Also shown are the elastic properties in
vertical two-way travel time. The wavelet used is a
standard 7-12-40-55 Hz bandpass filter. A positive (red)
loop means a hard kick.
Figure 2: Well logs for 44/23b-13 and the background trends obtained from them. A synthetic
slowness gather has been calculated from the logs.
Target oriented inversion
The inversion scheme (Figure 1) is not applied to surface data, but to data that is redatumed, or
demigrated, to a boundary at the top of the target interval. A mapped seismic horizon can serve as a
suitable boundary. The general workflow is shown in Figure 3.
Figure 3: Flow-chart for bringing the data down to a target location on the target boundary. At
every location A the 2D W / p CMP data is inverted for a 1D model. All locations are independently
processed on a trace-by-trace basis.
76th EAGE Conference & Exhibition 2014
Amsterdam RAI, The Netherlands, 16-19 June 2014
In the present study the data-set at the top of the objective sequence is created by demigrating the
migrated data over the target interval. This is possible when dealing with a plane wave W/ p domain,
assuming a layered medium immediately below the target location (A). Working in a trace-by-trace
mode ensures that the geological 3D dip is honoured by the data. Overburden transmission effects and
the wavelets at the target boundary are extracted from seismic-to-well matches. When more wells are
available then these wavelets can be interpolated between the boreholes.
Data pre-conditioning
The input data for this study consisted of pre-stack migrated offset cubes, with a maximum offset of
3000 m, which is very short given the depth of the target around 4000 m. Extensive pre-processing
was required to obtain a data-set at that depth, suitable for inversion to elastic parameters. Preprocessing steps for the 2D profile shown in this paper were: in-line velocity filtering of all common
offset lines, tight velocity filtering of all CIG’s, conversion of offset to ray-parameter using a stratified
overburden model and demigration of the migrated data to the target boundary using the seismic
velocity background model.
Full Wavefield Inversion
The inversion needs a very simple background model as input. The FWI-res method requires limited
user input, which makes it a more ‘hands-off’ approach. Benefits are that studies can be conducted
faster and results are more objective. The inputs are: 1) well logs for calibration and wavelet
extraction, 2) clean common image gathers, 3) top horizon of target zone and 4) a basic velocity
Figure 4: Inversion results for dimensionless compressibility contrasts at well 44/23b-13. Actuals
from the well are in red, predictions from seismic are in blue. The main reservoir sands are resolved,
but the retrieved values are relative. Middle figure in the upper row is the wave-number spectrum.
Upper right is the result, filtered in the wave-number domain back to seismic bandwidth.
Small cubes (1.5 by 1.5 km) around the calibration wells were inverted on a trace-by-trace basis.
Subsequently also 2D lines through the calibration wells were inverted. The generated output at the
intersections shows good stability of the procedure. The inverted section demonstrates presence of
fault blocks, that also have been mapped on the PSDM Wingate seismic dataset (Figure 5). The
section shown below displays the variation in compressibility (1/K). The compressibility correlates
with changes in P-velocity. Several reservoir packages are interpretable on the FWI-res output. Even
local brightening / dimming (tuning effects) at the gas-water contact are seen in some of the
Carboniferous reservoirs.
76th EAGE Conference & Exhibition 2014
Amsterdam RAI, The Netherlands, 16-19 June 2014
Figure 5: Full wavefield inversion (FWI-res) of 2D line across wells 44/23b-13 and 44/24-2. The
section shows compressibility variations, that are correlated with the Vp behaviour. Fault block in
well 44/24-2 on the right-hand side is located in a poor data area, but still the basic layering is
resolved by the inversion algorithm.
1) FWI-res provides a consistent trace-by-trace prestack inversion, tested here on a real seismic
dataset. It uses higher order scattering to constrain the inversion solution. The approach is
rather robust, even in poorer data zones of Carboniferous reservoirs in the Wingate PSDM
survey. No direct constraints are applied on the inverted properties.
2) A follow-up project is considered comprising a larger input area (182 km2) and that includes
three blind wells to check the reliability of the generated output.
3) Data conditioning in the x-direction and y-direction as well as in the offset direction before
demigration to the W/ p domain, creates better input for a more stable inversion.
4) The overburden background model only needs a suitable marker for the definition of the top of
the zone of interest. The input property models over the zone of interest should be very simple
and smooth. Smoothing and inter/extrapolation of well logs usually suffices to create such a
basic starting model.
5) The FWI-res method requires less user input than conventional inversion schemes (i.e. no
property constraining) and therefore represents a powerful ‘hands-off’ processing technique.
FWI-res is based on ideas developed by the Delphi Consortium (TU-Delft and TNO). The application
test has been performed by Delft Inversion subcontracted by Delphi Studio for Imaging on behalf of
Wintershall Holding GmbH. We thank our colleagues in Wintershall and the partners Gas-Union,
Gazprom International and XTO UK for permission to publish the results of this elastic inversion test.
Fokkema, J.T. and P.M. van den Berg, 1993. Seismic Applications of Acoustic Reciprocity. Elsevier,
Amsterdam, 350p.
Gisolf, A. and P.M. van den Berg, 2010. Target oriented non-linear inversion of seismic data. 72nd EAGE
Conference and Exhibition, Extended abstracts, Barcelona, 4p.
Kennett B.L.N., 1983, Seismic wave propagation in stratified media. Cambridge University Press, 352p.
Tetyukhina, D., S.M. Luthi and A. Gisolf, 2013. Acoustic non-linear full-waveform inversion on an outcropbased detailed geological and petrophysical model (Book Cliffs, Utah). AAPG Bulletin, Preliminary version
published online Ahead of Print 5 July.
76th EAGE Conference & Exhibition 2014
Amsterdam RAI, The Netherlands, 16-19 June 2014