IEEE Signal Processing - March 2018 - 85

younger sediment. Channels, as well as the associated depositional features such as overbanks, lobs, and fans, are important
targets of seismic interpretation because river/stream flows often
carry sands of high porosity and permeability that are superior
traps for hydrocarbon accumulation. Such rock difference gives
channels a unique, distinct texture, similar to salt domes; meanwhile, considering its meandering spatial distribution, interpreting such structures can be treated as a curve boundary detection
and tracking in digital image processing.

Gas chimneys
A gas chimney is defined as the leakage of gas from a poorly
sealed hydrocarbon accumulation in the subsurface. Gas chimneys
are often used as a robust hydrocarbon indicator, which implies
present or previous vertical migration of hydrocarbons or fluids
containing hydrocarbons between different geologic sequences.
In a seismic profile, a gas chimney is visible as a vertical zone
of poor data quality or push-downs. Therefore, a gas chimney is
often featured with a distinct texture. Correspondingly, similar to
salt domes, interpreting such structures can be treated as a texture
segmentation problem in digital image processing.

Common geophysical data sets
High-quality seismic data are essential for accurate subsurface interpretation with broad applications to both industrial
applications (e.g., oil and gas exploration) and environmental
studies (e.g., earthquake monitoring and CO2 storage). Driven
by the oil and gas industry, 3-D seismic data have been widely collected around the world in the past few decades, particularly in the areas with high hydrocarbon potential, such
as the North Sea, the Middle East, and the Gulf of Mexico.
Such data were often confidential during the initial stage of
reservoir exploration and production but are then released
for public use when the production goal has been achieved.
Table 1 lists seven seismic data sets that are commonly used
for developing and testing interpretation algorithms. The
table includes the geographical location of the surveys, the

dominant subsurface structures, the data size, as well as the
spatial dimensions. The first six data sets are available in the
typical SEG-Y format, which can be loaded directly using any
geophysical exploration software (e.g., OpendTect and Petrel)
or popular programming packages such as MATLAB and
Python. Large North Sea Data Set of Migrated Aggregated
Seismic Structures (LANDMASS) is available in the MATLAB MAT-File format and provides a large number of image
patches generated from the F3 block to facilitate structurebased seismic interpretation and machine-learning studies.
All links to these data sets and other resources can be found
at https://ghassanalregib.com/datasets-and-resources-forcomputational-seismic-interpretation/.

Subsurface event detection and tracking
A successful reservoir exploration requires a reliable identification of indicative subsurface structures as introduced in the previous section. Consequently, the majority of existing interpretation
algorithms focus on detecting such structures or events in geophysical terms. We have recently proposed an additional task that
tracks such structures by creating a semiautomated interpretation
workflow. In this section, we review these methods for each of
the four key structures, respectively. A depiction of all such algorithms is shown in Figure 3.

Fault detection
In a two-dimensional (2-D) seismic section, faults indicate displacements along fractures, as observed in Figure 2. Because of
the geological constraints that are associated with the formation
process, faults have two specific features. One is the geological
feature, which is the discontinuity along horizons. The other is
the geometric feature, i.e., line-like or curved shapes in 2-D seismic sections, which appear as curved surfaces in a 3-D seismic
volume. Computational fault detection methods are commonly
developed based on these two features. The discontinuity of
faults can be characterized by several seismic attributes such as
entropy [2], curvature [3], and coherence [4]-[6]. Among them,

Table 1. List of commonly used data sets for interpretation [1].
Data set and Location

Dominant Structures

Size

Domain

Dimensions

F3 block in The
Netherlands, North Sea

Salt domes (~1,500 ms); faults (~1,200
ms); gas chimneys (~500 ms)

494 MB

Time

Inline: 651, Crossline: 951, Samples/trace: 463

Stratton field in south
Texas, United States

Channels (~845 ms); faults (~2,000 ms)

122 MB

Time

Inline: 100, Crossline: 200, Samples/trace: 151

Teapot Dome in
Wyoming, United States

Faults (~5,500 ft)

421 MB

Depth

Inline: 345, Crossline: 188, Samples/trace: 1,601

Great South Basin in
New Zealand

Faults (~2,000 ms)

38.1 GB

Time

Inline: 1241, Crossline: 2,780, Samples/trace:
751

Waka Basin in New
Zealand

Channels (~1,800 ms)

23.2 GB

Time

Inline: 801, Crossline: 5,756, Samples/trace:
1500

SEAM synthetic data set

Salt domes (~2,000 ms)

4.2 GB

Time and
depth

Inline: 1169, Crossline: 1,002, Samples/trace:
851

LANDMASS data set from
F3 block

Horizons, chaotic reflections, faults and
salt domes

0.95 GB

Time

Horizon patches: 9,385, chaotic patches: 5,140,
fault patches: 1,251, salt-dome patches: 1,891

IEEE Signal Processing Magazine

March 2018

https://www.ghassanalregib.com/datasets-and-resources-for-computational-seismic-interpretation/ https://www.ghassanalregib.com/datasets-and-resources-for-computational-seismic-interpretation/

Table of Contents for the Digital Edition of IEEE Signal Processing - March 2018