Seismic data on the workbench

“My goal is to equip explorationists with a toolbox containing all relevant algorithms to efficiently create optimal seismic volumes for interpretation and stratigraphical analysis,” says Anders, who hails from Sweden but has been part of the Norwegian generation of successful entrepreneurs in the energy service sector over the past decades. “We want to simplify complex geo­physics by making seismic post-stack data improvement and optimalization easily accessible to all interpreters.”

Anders continues to explain that the majority of seismic volumes have undergone mainstream pre-process­ing. “The seismic character and qual­ity vary and in many cases are sub-op­timal for structural and stratigraphic interpretation,” he says.

“We are therefore building an AI interactive and stepwise process where­in the interpreter experiments with various seismic attributes, visual blends or physical blends, until a combination that gives the best result for efficient seismic interpretation. We will intro­duce a multi-attribute physical blend function, accompanied by a copilot designed to guide and assist the inter­preter in determining the most suitable seismic volume,” continues Anders.

“Presently, many interpreters ac­cept the provided seismic volume without scrutinizing its data quality and usability, mostly due to limited project time. We provide the inter­preter with all the necessary post-pro­cessing imaging tools, enabling users to interactively, easily and efficiently extract all available deterministic in­formation from the valuable seismic volume,” adds Brit.

Extracting geologically meaningful structures

Anders and Brit have observed that de­spite all the conversations around AI and machine learning, which certainly dominates much of the subsurface ge­oscience narrative, challenges remain when applying deep learning to large-scale seismic interpretation.

“Neural networks are typically con­strained by memory limitations, requir­ing seismic 3D volumes to be subdivid­ed into smaller sub-cubes for training and inference,” says Anders. “These sub-volumes must then be stitched to­gether, often introducing discontinu­ities and mismatches. More critically, this subdivision can obscure important geological context, especially for large-scale stratigraphic patterns that span the full extent of a volume.”

“These limitations have moti­vated us to rethink our approach to multi-horizon and multi-zone inter­pretation. When interpreters examine a seismic section, their eyes are naturally drawn to broad, low-frequency features – coherent reflections and first-order stratigraphic units that represent dep­ositional cycles bounded by maximum flooding surfaces, unconformities, or other major geological events. These features are intuitive for humans to trace, yet they remain difficult for auto­mated methods to extract consistently.

To address this, we introduce a framework designed to automatically extract first-order sequences from a seismic volume. Our approach begins with spectral decomposition, empha­sizing low-frequency components to enhance the visibility of large-scale stratigraphic units. These enhanced features are then interpreted through the reflection dip field, which we treat like a fluid velocity field, gen­erating seismic flowlines that follow the underlying structural geometry of the seismic volumes.

Flowlines capture the geometry of seismic reflectors by integrating along the local dip, yielding dense networks of trajectories. However, not all flow­lines correspond to significant geologi­cal features. To isolate those of interest, we apply a targeted “smart selection” strategy based on seismic attribute analysis along the flowlines themselves.

This workflow begins by comput­ing seismic attributes along each flow­line in a 2D seismic section. These at­tributes capture changes in reflection character and, when evaluated along geologically meaningful paths, reveal lateral continuity and context that is often missed in traditional, seismic amplitude analysis.

We then identify breaks or con­trasts in these attribute profiles to select the most geologically relevant flowlines. These selected trajectories serve as seed lines for extension into 3D, allowing us to automatically gen­erate a multi-zone probability volume that reflects the underlying first-order seismic stratigraphy. Each zone can be visualized independently, colored, or filtered to support focused interpreta­tion of specific intervals (Figure 1).

With this approach, we emphasize the goal of identifying and extracting geologically meaningful structures, ad­vancing toward a more consistent and automated interpretation framework that also can be used for meaningful seis­mic attribute extraction and analysis.”

The workbench

Another thing the GeoMind team has in the pipeline relates to further en­hancement of seismic imaging quality. “Post-stack seismic volumes often con­tain valuable geological information that remains obscured due to noise, amplitude imbalance, or limited fre­quency content – making image en­hancement an important step for max­imizing interpretability,” says Anders. “One-click AI functions such as Ajax, are already available in OpenMind and prove to be very successful.”

“We are currently working on a so-called processing workbench de­signed to support both expert users and non-experts to simplify and streamline seismic image enhancement and mul­ti-attribute volume creation, ultimately improving geological interpretation.”

To ensure scalability and perfor­mance, the system supports batch job scheduling in the background, allowing users to launch and queue processes without interrupting inter­pretation work.

“A key capability of the work­bench will be multi-attribute volume blending, where several attributes such as relative acoustic impedance, edge sharpening, frequency rebalancing, and fault likelihood, combined into a single enhanced output,” adds Brit. “We can build these volumes today through a bit of a cumbersome process in the OpenMind trace calculator, whilst we aim to streamline this func­tionality. This will produce volumes that better highlight stratigraphic and structural features than conventional full-stack displays.”

Figure 3 demonstrates how such blending enhances interpretation: Clearer reflector terminations, im­proved fault visibility, and more bal­anced amplitude distributions. While this specific attribute combination produced optimal results for one pro­ject, every seismic volume is unique, and the ability to experiment and tai­lor blends is essential. The final output of the workbench is a high-quality ZGY volume, fully compatible with platforms like Petrel.”

This enables interpreters to seam­lessly transition from enhancement to interpretation, leveraging the im­proved data quality directly within their existing workflows.