🤖 AI

Unsupervised Electrofacies Classification and Porosity Characterization in the Offshore Keta Basin Using Wireline Logs

This study demonstrates that an unsupervised K-means clustering workflow applied to wireline logs can effectively characterize electrofacies and porosity in the core-scarce offshore Keta Basin, providing a robust framework for early-stage formation evaluation in frontier basins.

Original authors: Hamdiya Adams, Theophilus Ansah-Narh, Daniel Kwadwo Asiedu, Bruce Kofi Banoeng-Yakubo, Marcellin Atemkeng, Thomas Armah, Richmond Opoku-Sarkodie, Rebecca Davis, Ezekiel Nii Noye Nortey

Published 2026-05-01

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Hamdiya Adams, Theophilus Ansah-Narh, Daniel Kwadwo Asiedu, Bruce Kofi Banoeng-Yakubo, Marcellin Atemkeng, Thomas Armah, Richmond Opoku-Sarkodie, Rebecca Davis, Ezekiel Nii Noye Nortey

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to understand the inside of a giant, dark cake that you can't cut open. You can't see the layers of sponge, fruit, or cream, but you have a long, magical stick that you can push all the way through the cake. As you push it down, the stick measures things like how hard the cake is, how much sugar is in it, and how much air is trapped inside.

This is exactly what the researchers in this paper did, but instead of a cake, they were looking at the offshore Keta Basin in Ghana—a giant underwater layer of rock where oil and gas might be hiding. The problem? They didn't have any "cake samples" (core data) to look at directly because drilling for samples is too expensive and difficult in this area. They only had the "stick" (wireline logs) to tell them what was down there.

Here is how they solved the mystery, broken down into simple steps:

1. The "Magic Stick" (Wireline Logs)

The researchers used a standard set of six measurements taken from a single well (Well C). Think of these as six different senses the stick has:

Gamma Ray: Tells them how "muddy" the rock is (like checking for dirt).
Density & Neutron: Tell them how heavy the rock is and how much space (porosity) is inside it.
Sonic: Measures how fast sound travels through the rock (like tapping a wall to see if it's hollow).
Resistivity & Photoelectric Factor: Tell them about the rock's electrical properties and what minerals it's made of.

2. The "Grouping Game" (Unsupervised Clustering)

Since they didn't have a textbook to tell them what the rocks were, they used a computer trick called K-means clustering.

Imagine you have a huge bag of mixed-up LEGO bricks of different colors and shapes, but you can't see the colors clearly. You ask a smart robot to sort them into piles based on how similar they feel to each other. The robot doesn't know what a "red 2x4 brick" is called; it just groups the bricks that feel the same together.

In this study, the computer looked at all the data from the "magic stick" and grouped the rock layers into four distinct piles (clusters) based on how similar their measurements were.

3. Checking the Work (The "Silhouette" Test)

How do you know the robot didn't just make random piles? The researchers used a mathematical test called a Silhouette score.

Think of it like a popularity contest. If a LEGO brick is in the right pile, it should feel very comfortable there and very far away from the other piles.
The computer gave them a score of 0.50. In the world of data, this is like getting a "B" grade. It's not perfect, but it's strong enough to say, "Yes, these groups are real and meaningful," not just random noise.

4. What Did They Find? (The "Rock Continuum")

Once the computer sorted the rocks, the researchers looked at the piles and gave them names based on what the measurements suggested. They found a smooth transition, like a gradient, rather than sharp, sudden changes:

Pile 1 (The Muddy One): High clay content, lots of "dirt," and lower porosity (less space for oil).
Pile 2 & 3 (The Mix): A blend of sand and clay.
Pile 4 (The Clean One): Mostly clean sandstone, with more space (porosity) and less clay.

They also noticed that as they went deeper, the rock generally got tighter and less porous (like a sponge getting squeezed), but there were specific layers where the rock was "fluffier" and had more space, which is where oil or gas might hide.

The Big Takeaway

The main point of this paper is that you don't need to dig up a physical sample to understand the rock.

By using a smart computer program to sort the "magic stick" data into groups, they created a reliable map of the underground layers. This is a "frontier" method, meaning it's perfect for new, unexplored areas where you can't afford to drill for samples. It turns a confusing mess of numbers into a clear, organized picture of what the ground looks like, helping scientists decide where to look for energy resources next.

In short: They taught a computer to sort underwater rocks by their "personality" (measurements) instead of their "name" (samples), and it worked well enough to give them a clear map of the Keta Basin's underground layers.

1. Problem Statement

The study addresses the challenge of subsurface characterization in frontier offshore basins, specifically the Offshore Keta Basin in Ghana. These regions face significant limitations:

Data Scarcity: Core data is sparse or non-existent due to high drilling costs and logistical challenges.
High Uncertainty: Exploration decisions must be made with limited geological control.
Reliance on Logs: Wireline logs are often the only continuous source of high-resolution subsurface information.
Subjectivity: Traditional interpretation relies heavily on manual, subjective log analysis, which lacks reproducibility in data-scarce environments.

The core problem is how to derive robust, geologically meaningful lithological units (electrofacies) and porosity estimates using only standard wireline logs without labeled training data or core calibration.

2. Methodology

The authors developed a fully reproducible, unsupervised machine learning workflow applied to Well C in the Keta Basin. The process involved the following steps:

Data Acquisition & Preprocessing:
- Dataset: ~11,195 depth-indexed samples from Well C (depth interval: 1,358.34 m to 3,064.31 m).
- Input Logs: Gamma Ray (GR), Bulk Density (RHOZ), Neutron Porosity (NPHI), Compressional Sonic (DT), Photoelectric Factor (PEFZ), and Deep Resistivity (AHT60).
- Quality Control: Removal of borehole washout intervals, outlier screening ( $\pm3\sigma$ ), and Z-score normalization to ensure equal weighting of variables.
Porosity Estimation:
- Calculated Density Porosity ( $\phi_D$ ) and Neutron Porosity ( $\phi_N$ ) using standard limestone matrix assumptions ( $\rho_{ma} = 2.71$ g/cm $^3$ ).
- Computed an Average Porosity ( $\phi_{avg}$ ) to serve as a smoothed petrophysical indicator for validation, rather than a direct prediction target.
Unsupervised Clustering (K-Means):
- Applied K-means clustering in the multivariate log space.
- Cluster Selection ( $k$ ): Determined using the Elbow Method (analyzing inertia reduction) and Silhouette Analysis (measuring cluster cohesion and separation).
- Optimal $k$ : Selected as 4 clusters based on a sharp drop in inertia and an average silhouette coefficient of ~0.50.
Geological Interpretation:
- Translated statistical clusters into Electrofacies based on log-response patterns (e.g., high GR + low density = shale; low GR + high density = clean sand).
- Mapped clusters to a lithological continuum: Shale $\rightarrow$ Sandy Shale $\rightarrow$ Shaly Sandstone $\rightarrow$ Clean Sandstone.

3. Key Contributions

The paper makes four primary contributions to the field of geoscientific data analysis:

First Application: Provides the first documented electrofacies characterization of the Offshore Keta Basin using unsupervised machine learning.
Integrated Framework: Successfully integrates density-neutron porosity computation with unsupervised clustering, creating a unified framework for joint petrophysical and lithological interpretation.
Hybrid Approach: Combines quantitative clustering diagnostics (inertia, silhouette) with geological reasoning to guide the selection and interpretation of clusters, ensuring the results are physically meaningful.
Reproducible Workflow: Establishes a log-only, unsupervised workflow specifically designed for frontier basins where core data is unavailable, offering a transparent and cost-effective alternative to traditional methods.

4. Results

Clustering Performance: The K-means algorithm successfully identified four distinct electrofacies. The average silhouette coefficient of 0.50 indicated moderate but meaningful separation, confirming that the clusters are statistically coherent while acknowledging the natural transitional nature of geological formations.
Lithological Continuum: The resulting electrofacies formed a systematic, depth-continuous pattern representing a geological continuum from shale-dominated units to cleaner sandstone-dominated units.
Porosity Correlation:
- Porosity generally decreased with depth due to mechanical compaction.
- High-porosity intervals correlated strongly with electrofacies characterized by lower Gamma Ray (cleaner sands) and shorter sonic transit times.
- Low-porosity intervals aligned with high Gamma Ray and high neutron responses (shale-rich intervals).
Visual Validation: Crossplots (GR vs. RHOZ, NPHI vs. DT, PEFZ vs. GR) showed distinct, non-overlapping clusters with kernel density envelopes, confirming that the clustering captured dominant modes in the data rather than random noise.

5. Significance and Impact

Frontier Basin Exploration: The study demonstrates that reliable subsurface characterization is possible in data-scarce environments without expensive core data. This is critical for early-stage appraisal in emerging hydrocarbon provinces like the Keta Basin.
Reduced Subjectivity: By relying on quantitative metrics and unsupervised algorithms, the workflow minimizes interpreter bias, leading to more reproducible results.
Scalability: The methodology is transferable to other frontier offshore basins globally where similar wireline log suites are available but core data is lacking.
Future Potential: The authors propose this workflow as a foundation for future integrated studies, including multi-well analysis and the integration of seismic attributes to map lateral electrofacies continuity.

In conclusion, the paper validates that unsupervised machine learning is a robust tool for extracting geological structure from wireline logs, providing a practical, cost-effective solution for formation evaluation in challenging exploration settings.