Professional Documents
Culture Documents
A Tutorial: On The Effect of Seismic Noise
A Tutorial: On The Effect of Seismic Noise
A Tutorial: On The Effect of Seismic Noise
A tutorial
on the effect
of seismic
noise.
(Requires lots
of toggling to
really see how
noise interacts
with signal).
Toggle guide
After
A tutorial
on the effect
of seismic
noise.
(Requires lots
of toggling to
really see how
noise interacts
with signal).
The noise has
been removed
by Paiges non
linear logic.
I start with two sets of before
& after gathers to emphasize
the fairly amazing success of
the noise removal algorithm.
after showing some results I
take you on to a discussion of
the noise itself, using more of
the examples as prompts.
So toggle back
and forth.
Toggle guide
Before
I am putting these
2 before and afters
up front to get you
used to toggling. Ill
then show some of
my inverted and
integrated results
before continuing
with a long series
of
further
examples.
Toggle guide
After
Some results
I am putting these
2 before and afters
up front to get you
used to toggling. Ill
then show some of
my inverted and
integrated results
before continuing
with a long series
of
further
examples.
Whie you can see
the logic got rid of a
lot of multiples and
refractions,
the
data is still noisy,
so be kind in
judging the results.
And remember you
are lookiing at the
simulation of sonic
logs.
Toggle guide
Introducing my
inversion/integration
module
Its purpose is to simulate sonic logs from
seismic data. Since reflections are primarily
caused by abrupt changes in velocity, this is
our best hope to actually model lithology. The
ability to do this is vital to long range seismic
correlations.
At the left is an in-line with a superimposed
sonic log. Because the stratigraphy is quite
regular in this study zone, the sonic log is
almost generic, and no great care was taken
to place it exactly (although the location is
close). The purpose was to make sure my
logic knew what it was doing when it indicated
the presence of thick beds. As you can see,
there is no problem with the well match here.
The thrust of this show is noise removal
Hopefully you will spend a good bit of time
toggling between the input and the de-noised
output. However, to prove to myself that the
logic was stable for the entire volume, I ran
every 20 in-lines, and two cross lines. After
showing the latter, I take you back to the main
toggling theme. At the end of that series I will
show the in-line results.
One finger on the left arrow and one on the
right leads to easy toggling keep them
there and it gets to be fun.
Before
are
a series of factoids. That
form the basis of the noise
removal logic.They are all
either self evident, or at least
easily provable. Most have
been
ignored
by
the
processors
Noise removal
To begin let us
look at basic gather
principles. Each gather trace
is made up of a complex overlap
of primary reflections coming
from individual reflectors, each traveling
independently . The mix does not occur
until the receiver forces it. The final spacing
between these primaries is dependent on the
horizontal travel path component. Thus its likely
that the resulting mixtures from successive offsets
will display different character.
Continuity within the gather will depend on
the separation between the involved primaries.
is not
100% but I think you will
agree it is a big step in the
right direction.
Toggle guide
After
Next toggle pair
are
a series of factoids. That
form the basis of the noise
removal logic.They are all
either self evident, or at least
easily provable. Most have
been
ignored
by
the
processors
Noise removal
To begin let us
look at basic gather
principles. Each gather trace
is made up of a complex overlap
of primary reflections coming
from individual reflectors, each traveling
independently . The mix does not occur
until the receiver forces it. The final spacing
between these primaries is dependent on the
horizontal travel path component. Thus its likely
that the resulting mixtures from successive offsets
will display different character.
Continuity within the gather will depend on
the separation between the involved primaries.
is not
100% but I think you will
agree it is a big step in the
right direction.
Toggle guide
Before
Since
As we move through
these comparison sets youll
see dozens of individual events
on both sides of the tuning question.
After liftoff, some continue with just a
broadening, and some completely break
up. This illustration of seismic geometry is
worth a lot of study. I try to help by pointing
to some of the better examples.
the
gathers
Toggle guide
After
Next toggle pair
Since
As we move through
these comparison sets youll
see dozens of individual events
on both sides of the tuning question.
After liftoff, some continue with just a
broadening, and some completely break
up. This illustration of seismic geometry is
worth a lot of study. I try to help by pointing
to some of the better examples.
the
gathers
Toggle guide
Before
Watch
the
arrows
Toggle guide
After
Next toggle pair
Watch
the
arrows
Toggle guide
Before
Toggle guide
After
Next toggle pair
Toggle guide
Before
Consistency
is a logical
test, so keep
on toggling.
Toggle guide
After
Next toggle pair
Consistency
is a logical
test, so keep
on toggling.
Toggle guide
Before
Noise intro
The system is seeing two
types. The first are strong
inter-bed multiples and the
second are equally strong
refractions (discussed later).
The best mode for attacking
both is via essentially raw
data, where the curvature
patterns of the two are more
distinct from the reflections.
Here we have to deal with
velocity error, and that is not
so easy.
On
the
slides, I ask
input
you to pay
attention to the zone starting
at 800 ms.. You should be
able to see a multiple pattern
that begins to encroach on
the the target zone around
one second. As I mentioned,
you have to get used to the
fact that when two strong
patterns mix, you only see
smatterings of each. This is
problem the system faces.
Toggle guide
After
Next toggle pair
Noise intro
The system is seeing two
types. The first are strong
inter-bed multiples and the
second are equally strong
refractions (discussed later).
The best mode for attacking
both is via essentially raw
data, where the curvature
patterns of the two are more
distinct from the reflections.
Here we have to deal with
velocity error, and that is not
so easy.
On
the
slides, I ask
input
you to pay
attention to the zone starting
at 800 ms.. You should be
able to see a multiple pattern
that begins to encroach on
the the target zone around
one second. As I mentioned,
you have to get used to the
fact that when two strong
patterns mix, uou only see
smatterings of each. This is
problem the system faces.
Toggle guide
Before
Toggle guide
After
Next toggle pair
Toggle guide
Before
Toggle guide
After
Next toggle pair
Toggle guide
Before
Worth a
good look.
Toggle guide
After
Next toggle pair
Worth a
good look.
Toggle guide
Before
Toggle guide
After
Next toggle pair
Toggle guide
Before
Toggle guide
After
Next toggle pair
Toggle guide
Before
Non-linear coding
gives us the ability to
direct the scanning
logic to places it can
detect noise patterns.
Toggle guide
After
Next toggle pair
Non-linear coding
gives us the ability to
direct the scanning
logic to places it can
detect noise patterns.
Toggle guide
Before
Keep
toggling!
Frequency sensitive
filtering is destructive in its
nature. It discriminates against
slopes it does not like. It is far
better to predict noise patterns then
gently lift them off, bringing out the
energy which lies below. The success
of this noise removal algorithm depends
on this virtue. Unfortunately this data had
been subjected to traditional band pass
filtering before stack.
Toggle guide
After
Next toggle pair
Keep
toggling!
Frequency sensitive
filtering is destructive in its
nature. It discriminates against
slopes it does not like. It is far
better to predict noise patterns then
gently lift them off, bringing out the
energy which lies below. The success
of this noise removal algorithm depends
on this virtue. Unfortunately this data had
been subjected to traditional band pass
filtering before stack.
Toggle guide
Before
Toggle guide
After
Next toggle pair
Toggle guide
Before
Toggle guide
After
Next toggle pair
Toggle guide
Before
Toggle guide
After
Next toggle pair
Toggle guide
Before
Toggle guide
After
Next toggle pair
Toggle guide
Before
Toggle guide
After
Toggle guide
The section on the left is a stack of the de-noised gathers. The one on the right is the inverted and integrated version.
The pre-processors seem to have done everything they could to make our task more difficult. As we move through this series
I will point out a few of these bad processing steps, I hasten to add that what they did was pretty much industry standard.
I start with the use of pre-stack migration. Obviously there are no strong dips here to warrant such logic. The fact is that the
heavy mixing inherent in the process improves the appearance. Howeve it also muddles fault breaks and wrecks havoc with noise.
Continuing the pre-stack migration objection If you study these in-lines carefully, you should see indications of faulting. I point
out one or them here. Muddling by the mixing action completely obliterates much of these clues. On the integrated side, breaks in
character are often our best clues., In any case this is one of the areas calling for intense visual study.
From the noise removal point of view, the most serious damage done by the pre-stack migration was the garbage it produced
as a result of not being able to handle multiples and refractions in its data re-arrangement.
The noise removal logic uses pattern recognition to search out events having non-reflection lineups. Because we were not
able to get raw data with no NMO correction, we had to work with velocity error, which is not so good when looking for refractions.
Another fault hint again the pre-stack migtation has pulled data across what should be breaks.
Getting a little raunchy again the noise removal is pretty good but far from perfect.
We begin to see increase amplitude on what I have tentitavely identified as the Morrow. Remember that integration has
made the amplitudes more meaningful, so this could be of interest.
Again, even though there is still a lot of noise, the detail could be important.
Excuse me!
This is the Of course we might pay attention to the rest of the section, but end.