[Users] Re: About Channel detection

[Friso Brouwer]

Dear Mr. Hou,

Channel detection is frequently done in OpendTect. There are several
methods for detecting channels. Depending on there characteristics of
the geology, the seismic data quality and the focus of your project some
methods might work better than other. The best way is to test and then
select one of the following methods:

If your channels are at different levels in the seismic (3D
distribution), from simple to advanced, the following methodologies are
available:

-1- Use a single attribute that characterizes the channel. Examples:
a) Energy (works well if you have a good contrast in acoustic properties
between channels and embedding, e.g. shallow sandy channels, incised in
a homogeneous shaly background). In the volume viewer the extend and
geometry of the channels can be assessed by making all values
transparent , except the anomalous values associated with the channels
(right click on color bar, select edit).
b) Similarity (detects the edges of the channels), this works if the
channels have fairly steep edges, such that there is a sharp lateral
break in seismic character. Also, it is convenient for interpretation if
the geology is fairly flat, or good horizons cutting through the
channels are available (typically I interpret a horizon on a continuous
event near the interval with channels and shift that up an/or down to
display the edges of the channels).
c) Curvature: in cases when there is an anomalous bedding direction
within a channel, or deformation due to differential compaction, the
curvature attribute may highlight channels. Note that a good quality
steering cube is a pre-requisite for this.

-2- Use the spectral decomposition attribute. Different parts of the
channel may show up at different frequencies, this indicates where the
total thickness or internal geometry of the channel changes. See also
this documention
<http://www2.dgb-group.com/images/stories/PDF/tu-05-01-sd.pdf>.

-3- Unsupervised NN (neural network): This NN will divide the seismic in
a number of classes with each its different set of seismic
characteristics. Start by creating and storing a random pickset in the
part of the survey under investigation (click picksets in the Opendtect
tree, create new, generate random pick in volume/between horizons, after
creating the pickset do not forget to store). Select a number of
attributes that characterize your channels (see -1- and -2-, also use
average frequency and instantaneous attributes)- note that in the 3D
application of unsupervised NNs the attributes should be insensitive to
the phase of the reflection, including enough length in the time windows
used (> 1/2 dominant period wavelet), otherwise a blurred "copy" of the
input seismic will be created. After creating and applying the
unsupervised NN (described here
<http://www2.dgb-group.com/rel/html/dgb/c645.htm#AEN676>), you will need
to interpreted the classes. In the volume viewer the 3D extend and
geometry of the classes characteristic for channels can be evaluated by
setting all other values in the colorbar to transparent.

-4- Supervised NN: You will pre-interpreted a number of points as being
channel and no-channel (2 manually created pick sets). Then you will
train the NN to discriminate channel from no-channel, using a similar
set of attributes as in -3-. Apply the channel node of the NN and
highlight the channels by making lower values transparent. Find more
documentation here <http://www2.dgb-group.com/nn_doc.html>.The general
workflow for supervised detection is described in this presentation
<http://www2.dgb-group.com/component/option,com_docman/task,doc_download/gid,29/>.

-4a- Fingerprint attribute: a simpler variation of the supervised
method, not using a NN and only using positive picks. This works best if
you have one typical example of the channel and you want to infer the
extend of that channel, or areas in the survey with the same
characteristics, see detailed description
<http://www2.opendtect.org/rel/html/r2599.htm>.

If your channel(s) is confined to one or a few horizon levels, with
horizon interpretation available.

Methods 1, 2, 4 and 4a will still work as in the 3D situation.

For method 3 we can make a major improvement, but a good quality horizon
that accurately follows a one chrono-stratigraphic level needs to be
available - if the horizon is inaccurate, part of the variation in this
attributes will be due the errors in the horizon. The difference with
the 3D situation is that we now can use phase-sensitive attributes to
the NN, we do this by using the complete waveform around the horizon.

Workflow: Open the default attribute set called "unsupervised
segmentation 2D". Create a random pickset on the horizon under
investigation (click "PickSet" in the OpendTect tree, create new,
generate random picks on horizon, after creation do not forget to
store). Determine the length of the reflection signature of the event
under investigation and use the shifted seismic samples accordingly in
the NN. Train the NN. When the NN is train the class centers can be
viewed by clicking on "info" in the NN management window and click
display. Finally apply the NN on the horizon and interpret the classes
using morphology and well ties.
Notes:
a) The results can be stored as surface data (no need to re-apply the NN
to the horizon). Right click on the appropriate NN layer in the
OpendTect tree and select "save attribute".
b) Sometimes certain seismic classes should be merged to fit one
geological object, e.g. a channel with thicker and thinner parts might
be split in 2 classes.
c) This is in my eyes the most sensitive method, picking up subtle
changes related to lithology and deposition. However, a pittfall is
inprint of variations in top and bottom layers - to interpreted the
results lateral constant top and bottom layers are assumed, such that
all variation in reflection signature is attributed to the variations in
the target formation. This is a problem for almost all methods
mentioned. However, because the greater sensitivity of this method, the
danger is greater here.

Regards,

-- 
Friso GC Brouwer
Geoscientist

dGB-USA
One Sugar Creek Center Boulevard
Suite 935
Sugar Land, TX, 77478

Tel 281 240-3939 (main)
Tel 281 240-6957 (direct)
Fax 281 240-3944
e-mail friso.brouwer at dgb-group.com
http://www.dgb-group.com


Jianyong Hou wrote:
> Dear Sirs
> I have 3D seismic data and want to detect Channel from it by
> OpendTect. Which seismic attributes are often useful for this channel
> detection?
> Thank you
> JGI Jianyong Hou
> ------------------------------------------------------------------------
>
> _______________________________________________
> Users mailing list
> Users at opendtect.org
> http://lists.opendtect.org/mailman/listinfo/users
>   


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opendtect.org/pipermail/users/attachments/20070618/cda37378/attachment.html>