P-390

Low frequency modeling and its impact on seismic inversion data

Nishant Kumar, Satyapal Singh Negi, RIL

Summary

Integration of low frequency model during seismic inversion enhances the qualitative and quantitative interpretation levels
of seismic inversion results. Information required to build low frequency model primarily comes from the well data. In some
geological environment it is possible to make a simple interpolation or extrapolation of the well data within a basic geological
framework to have a reasonable low frequency model. Often greater geological constraints are required which can be
incorporated in various ways.This paper summarizes the various methodologies for making low frequency model. An attempt has
also been made to have analytical look on the suitability of low frequency model. The impact of low frequency model of seismic
inversion data is also analyzed.

Introduction Seismic data is inverted for physical properties of rock like

acoustic impedance shear impedance, poisson ratio and
Rock and reservoir parameters computed from the inverted density etc. Since the seismic data is bandlimited so in the
rock properties can be inaccurate, leading to wrong drilling absence of low frequency model the inverted physical
and development decisions due to lack of low frequency properties are bandlimited or relative. In order to translate
information on conventional seismic data. (Whitecombe the inverted data for petrophysical properties for
and Hodgson 2007,Ozdemir et al 2007).It is important to quantitative interpretation like porosity or fluid content we
understand that missing low frequencies and their modeling require absolute impedances, density or vp/vs. To achieve
from well data can cause bias in inverted rock properties this absolute property we need low frequency modeling to
irrespective of any inversion algorithm. compensate for the missing band of the seismic data.

The seismic data is bandlimited, it lacks the low Low Frequency Modeling
frequencies that are essential for relative to absolute
property conversion. Bandlimited nature of seismic would Low frequency modeling is an important aspect of
also mean that it is relative with respect to some absolute obtaining absolute rock properties. Low frequency is the
trend. The seismic data is generally of the range from 6- background information of the data and the high
8Hz to 60-80Hz. In case of seismic acquired for high frequencies are the details of the data that adjust themselves
frequency purposes the lower limit even starts from 15Hz. with the background. The low frequency information can be
Though in case of our seismic data the lower limit stands at regarded as the mean trend of the data and the higher
12 Hz. So it was required to patch up this 0-12 Hz region frequencies are the deviation from that mean trend.
from some method.The extreme end of frequencies (0-2
Hz) can be obtained from seismic velocity data, the low We divide the low frequency modeling aspect into three
end (2-12 Hz) can be obtained from well data and the high parts
frequency end, to a certain extent, can be obtained from 1. Background trends
geostatistical procedures. 2. Intermediate trends
3. Anomalous signatures
We here use Q-marine data for our study. It is lacking in
low frequencies upto 12 Hz. Generally it is observed that all Generally shale acts as a background in most of the
high frequency seismic data lacks in low frequency content. reservoirs. So the shale trends are the Background trends.
The reason for this is a scope of a different paper.

Reliance Industries Ltd, Mumbai, India

[email protected]
 Low frequency modeling and its impact on seismic inversion data

Any deviation from this shale trend is an anomalous

signature. Removing the background trend from the data
would reveal out the anomalous geobodies.

We have introduced one more category as the intermediate

trend .This intermediate trend (2-12 Hz) separates out itself
from the compaction or depth influenced background trend Figure 1: The geological framework model depicting the layering
(0-2 Hz).This intermediate trend is the void that we have in scheme.this framework model was populated by wireline logs.
our data. We have seismic velocity data covering the entire Different colors represent different stratigraphic units.
exploratory region. This seismic velocity data can provide
the compaction or depth trend (0-2Hz).But for intermediate The layering scheme of framework model was populated by
trends we dont have any information spreading to the the well logs for the low frequency model building. The
entire areal extent. We have only sparse well control to feed layering scheme uses 2ms sampling interval.
for this area.
Velocity Calibration
The data used for the low frequency modeling are
1. Interpreted horizons Velocity calibration plays important role in low frequency
2. Seismic interval velocity data model building process. Seismic interval velocity data was
3. Well log data (high cut filtered) used. The velocity obtained from the well logs is of high
frequency of the order of KHz. So a high cut filter was
Two methods were used for building the low frequency applied at 1Hz with a slope of 1Hz on this well velocity. On
model and the data was inverted using both the the other side pseudo log from the seismic interval velocity
methodologies. volume was extracted at well locations. As the frequency of
Method 1: Low Frequency model using well data and the velocity data was predominantly up to 1.5 Hz so logs
horizons. from both the well velocity and seismic interval velocity
Method 2: Background Low Frequency model from 0-2 Hz were kept to this value using high cut filter of 1Hzwith 1Hz
using seismic interval velocity data and Intermediate Low slope.
Frequency model from 2-12 Hz using well data and
horizons The well velocity was divided by seismic interval velocity
(both in high cut form) within the gate of well log
For each of these two methods the following schemes were availability. This division provides us with calibration
used factor log.
Scheme1: Using well logs as it is
Scheme2: Shale trend modeling and Sand modeling As the seismic interval velocity was up to 7 seconds but the
Framework building, velocity calibration, shale model well velocity log was available in limited region so there
building, sand model building and finally merging into was the need to extend the well velocity log and calibration
complete low frequency model are the procedures those are log. This calibration factor log was extended beyond the
involved in low frequency model building. region of well log availability by extending it using a
constant value of unity. This unitary value was used to
Framework building retain the seismic velocity as it is where the well log is not
present.
Geological model framework was built to provide the
background trend using the seismic interpreted horizon.
This framework model defines the stratigraphic and
layering scheme. This model includes the entire regional
horizon.

2
 Low frequency modeling and its impact on seismic inversion data

The wireline P-velocity logs were crossplotted against the

P-impedance logs. The intention was to achieve the
trendline. The data points fall in a linear trend. Following
equation was obtained and the ultra low frequency model
for Pimpedance was constructed using this equation taking
Figure 2: P-velocity calibrated velocity model as an input.
wireline log (black) and (P-impedance) = M * (P-velocity) + B
seismic interval velocity
log (blue).
The correlation coefficient was 0.983284.

These calibration factor logs were then interpolated using

the framework. This gave us calibration factor volume.
This calibration factor volume was multiplied with the
seismic interval velocity volume to furnish calibrated
interval velocity volume.

Figure 6: Crossplot between wireline P-velocity and P- impedance

logs. The data falls into almost perfect linear trend.

After obtaining the ultra low frequency model for p-

impedance the next job was to construct ultra low
Figure 3: Calibration factor volume from a line near well. The red frequency model for density and Vp/Vs. Next crossplot
shows the region where wireline log information was absent.
between wireline P-impedance and density log was made.
Obtained equation was used to construct the ultra low
The red portion in the calibration factor volume represents
frequency density model.
the part where the well log velocity was not available and
the calibration factor logs were extended with the unity
The following relation was used. The correlation
value (i.e. 1). In this part seismic velocity will remain as it
coefficient was 0.868.
was.
Density = M *(P-impedance) + B

Figure 4: Calibrated velocity volume from the same line near well.
The velocity increase from violet to red color.

Figure 7: Crossplot between wireline P-velocity and Density

logs. This data has a linear trend.

Next crossplot between P-impedance and Vp/Vs was made.

Figure 5: Spectra of calibrated velocity volume from the same line
near well. The black line demarcates the cutoff (1.5Hz) above which
The trend was fitted for the shale region as the shale is
we dont have significant amplitude in the spectra. background. The shale trend was not fully parabolic so the
equation used was not parabolic instead we opted for fourth
Background low frequency models were built for various degree equation since it was fitting properly.
properties using the 0-1 Hz with 1Hz slope component
from the calibrated seismic interval velocity volume.

3
 Low frequency modeling and its impact on seismic inversion data

The equation used is following.

Vp/Vs = A*(P-impedance) 4- B*(P-impedance) 3 +
C*(P-impedance) 2 - D* (P-impedance) + E

Fig 12: Low frequency model generated from well log interpolation
and merged with velocity data for background trend.

Shale Trend model building

First the shale wells were constructed by chalking out the

Figure 8: Crossplot between wireline P-velocity and Vp/Vs logs. The sand zones and replacing them with the background shale
trend is polynomial for shales. line.

Then these logs (p-impdenace, density and vp/vs) were

extrapolated using the equations derived from the
crossplots made in ultra low frequency modeling.

Figure 9: Ultra low frequency models of P-Impedance, Vp/Vs and Figure 13: The shale logs overlain by sand logs for P- Impedance,
density (from up to down). Vp/Vs and density.

Then these shale well logs were interpolated using the

geological framework using the nearest neighbour
interpolation technique with a sampling of 2msec.

Two categories of shale models were made one with the

Figure 10: Spectra of the ultra low frequency model.
intention of providing the low frequency model using the
wells alone and the other with the motive of mixing the
Simplistic models using the well log information
ultra low frequency models with the shale models that start
only from 1Hz with 1Hz slope called as method 1 and 2.

We started with simplistic models to test their impact on

inverted data. These models were the simple interpolation
of well log information within the geological framework.
These wireline logs were high cut filtered at 12Hz.

Fig 11: Low frequency model generated from well log interpolation.

4
 Low frequency modeling and its impact on seismic inversion data

Low Frequency Shale Trend model using only was to be done. Sand low frequency trend model building
wells was attempted after running the first pass of the seismic
inversion inverting the seismic for P-impedance, S-
impedance, density and Vp/Vs in a bandlimited manner.

A crossplot was made between bandlimited P-impedance

and S-impedance logs colored with water saturation log to
identify the possible gas sand prone bodies and to further
reflect these regions in the bandlimited volume (reflected by
simultaneous occurrence value of p-impedance and s-
impedance that indicate gas sand presence from
bandlimited logs and then using these values in bandlimited
volumes as an indicator of possible gas sand presence).

Figure 14: Low frequency shale trend models of P- Impedance,

Vp/Vs and density overlain by the shale logs.

Low Frequency Shale trend model using wells and

seismic velocity

Fig 17; Crossplot between bandlimited p-impedance and s-

impedance log colored with water saturation log. The data values
within the polygon were used for sand body rendering.

A Geobody check was made using bandlimited P-

impedance and bandlimited Simpedance to locate the
portions of possible gas sand regions. The skeleton or the
geometry of these geobodies was saved. The bandlimited
Figure 15: Low frequency shale trend models using wells and volumes of pimpedance, Vp/Vs and density within this
velocity for P-Impedance, Vp/Vs and density overlain by the shale saved frame were separated out and were termed as our
logs.(velocity contribution was used from ultra low frequency model. sand trend models.

We only show here P-impedance models. Models for

Vp/Vs and density can be similarly achieved.

Fig 16: Amplitude spectra of background trend and intermediate trend

merged P-impedance shale model. The spectra show the spectral band
that is missing in seismic data and is to be compensated.

Sand modeling
Fig 18: Sand model for P-Impedance.

Having built the shale trend models sand trend modeling

5
 Low frequency modeling and its impact on seismic inversion data

Complete Low frequency trend model

These sand models are then merged with the shale models
to provide the complete low frequency trend models. These
final trend models are then used in full bandwidth inversion
job.

Fig 19: Complete low frequency model (shale + sand) for P-

Impedance.

Fig 20: Complete low frequency model (shale + sand) for P-

Impedance for intermediate trend. Seismic velocity was used for
background trend. Fig 21: Inverted data from low frequency model a) simple well log b)
well log and velocity c) shale sand model d) shale sand model and
Full band Inversion velocity.

These low frequency models a) simplistic model using well Impact of low frequency on Inverted data
information only b) simplistic model using velocity
information for background trend and well information for Generally for the intermediate trends portion we only have
intermediate trend c) shale model sand model d) using well log information. This well log information is sparse. It
velocity information for background trend and shale sand is often said that the interpolation of these well logs within
information for intermediate trend were used for inverting the geological framework will not do justice for
to absolute data. We found that the results obtained from intermediate trend modeling since sparseness of well
shalesand modeling were better .Even in shale information requires geostatistical interpolation scheme and
sandmodeling the results from without usage of velocity different schemes can give different results for the
were better than usage of velocity. This could be due intermediate trend.
to calibration errors and localized velocity problems.

6
 Low frequency modeling and its impact on seismic inversion data

that the shale sand modeling can deliver good results.

Generally shale trends dont change significantly laterally
and could be good indicator of compaction trends. We also
used sand signatures so as to incorporate it in the complete
trend model. The lack of intermediate trends also doesnt
deteriorate the result to a significant degree. We found that
there were not very significant differences in the inverted
data results for various gaps in intermediate trends.

References

Ozdemir,H., 2008, Unbiased deterministic seismic

inversion; 70 EAGE Conference and Exhibition Extended
abstracts, P353.

Ozdemir, H.; Hansent,J.W. and Tyler, E. 2006, Rock and

reservoir parameters from pre stack inversion of surface
Fig 22: Inverted data from low frequency model with a frequency gap seismic data. First Break 2(10), 83-87.
of a) 2-12 Hz b) 4-12 Hz c) 6-12 Hz d) 8-12 Hz e) 10-12 Hz shale
sand model and velocity. Whitcombe D. and Hodgson L., 2007 Stablising the low
frequencies, the Leading Edge 26, 66-72.
So we tested the inverted results by keeping the gap in
intermediate trends. We inverted the data by keeping the Acknowledgement
gaps 2-12Hz, 4-12Hz, 6-12Hz, 8-12Hz, and 10-12Hz.
The author gratefully acknowledges Reliance Industries
We found that there were not much differences between the
Ltd. (E &P Business), Mumbai, India for support and
inverted data results because of gap in the intermediate
permission to publish the work. In particularly thank to Mr.
trends. Though the differences increase slightly with depth.
Bhagaban Das, RMG for his support during the project.
The difference between wireline and inverted results were
relatively more than the differences among inverted results
having gap of various intermediate trends.

Fig 23: Wireline P-impedance data and pseudo logs from the inverted
data utilizing low frequency model with the frequency gaps of 2-12
Hz, 4-12 Hz, 6-12 Hz, 8-12 Hz, 10-12 Hz .On right side spectra of the
same logs.

Conclusions

We have shown here various schemes for low frequency

modeling and their impact on full band inversion. We find