SIGMA: A Physics-Based Benchmark for Gas Chimney Understanding in Seismic Images

Bao Truong; Quang Nguyen; Baoru Huang; Jinpei Han; Van Nguyen; Ngan Le; Minh-Tan Pham; Doan Huy Hien

cs.CVMar 24, 2026

SIGMA: A Physics-Based Benchmark for Gas Chimney Understanding in Seismic Images

Authors: Bao Truong, Quang Nguyen, Baoru Huang, Jinpei Han, Van Nguyen, Ngan Le, Minh-Tan Pham, Doan Huy Hien,

Organizations: FPT Software AI Center · University of Liverpool · Imperial College London · University of Arkansas ·

Abstract

Seismic images reconstruct subsurface reflectivity from field recordings, guiding exploration and reservoir monitoring. Gas chimneys are vertical anomalies caused by subsurface fluid migration. Understanding these phenomena is crucial for assessing hydrocarbon potential and avoiding drilling hazards. However, accurate detection is challenging due to strong seismic attenuation and scattering. Traditional physics-based methods are computationally expensive and sensitive to model errors, while deep learning offers efficient alternatives, yet lacks labeled datasets. In this work, we introduce SIGMA, a new physics-based dataset for gas chimney understanding in seismic images, featuring (i) pixel-level gas-chimney mask for detection and (ii) paired degraded and ground-truth image for enhancement. We employed physics-based methods that cover a wide range of geological settings and data acquisition conditions. Comprehensive experiments demonstrate that SIGMA serves as a challenging benchmark for gas chimney interpretation and benefits general seismic understanding.

Figures & tables

Figure 2: An example seismic data with gas chimney areas.

Figure 3: Seismic image construction. (a) An illustration of seismic image acquisition with gas chimney effect. The sound waves become distorted when they pass through the gas chimney area. (b) The final seismic image after migration, showing that the gas chimney area leads to poor seismic attribute quality.

Figure 4: Dataset creation pipeline. (a) Overview of seismic image construction framework with gas chimney. (b) Step-by-step dataset generation process: starting from an original velocity model, a random fracture network is generated, followed by gas saturation simulation through a physically grounded modeling stage. The gas-affected velocity model is then constructed, and final seismic images are synthesized using reverse time migration.

Figure 5: Data samples. We present samples from our SIGMA, each sample consists of original velocity model, gas saturation map, gas-affected velocity model, gas-affected seismic image and clean seismic image. The yellow bounding boxes highlight the differences between normal and gas-affected regions.

Category	Parameter	Value	Metric
Gas chimney modeling	Gas bulk modulus ( $K_{g}$ )	$8\times 10^{8}$	Pa
	Shale bulk modulus ( $K_{s}$ )	$0.045\times 10^{9}$	Pa
	Shear modulus ( $G$ )	$3211$	Pa
	Initial gas density ( $\rho_{g}$ )	$1900$	$\text{kg/m}^{3}$
	Shale density ( $\rho_{s}$ )	$1900$	$\text{kg/m}^{3}$
	Diffusion time ( $t$ )	$750$	$year$

Table 1: Simulation and modeling parameters. Physical constants, hyperparameters, and numerical configurations used in our SIGMA dataset generation framework.

Figure 6: Effect of different physical conditions. Visualization of variations in the gas chimney effect $\mathbf{V}_{p}(\mathbf{x},t)$ and their impact on the results under different physical constant settings.

Figure 7: User study for assessing realism of synthetic seismic images. We conducted a perceptual test where each participant was shown ten unlabeled seismic images (five real and five synthetic) in randomized order and asked to judge whether each image looks like a seismic image or not . The chart summarizes the overall response distribution across ground-truth categories, reflecting how realistic the synthetic images appear compared to real ones.

Model	Year	IoU $\uparrow$	Dice $\uparrow$
FaultSeg [ 87 ]	2019	0.76	0.86
DualUnet [ 83 ]	2023	0.84	0.91
FaultFormer [ 77 ]	2025	0.80	0.87
FaultViT [ 44 ]	2025	0.75	0.86

Table 2: Gas chimney detection results.

Figure 8: Qualitative on enhancement task. We compare gas chimney enhancement outputs from different methods. The yellow bounding boxes highlight degraded gas chimney regions where models are expected to enhance and reconstruct seismic features.

Model	Year	SSIM $\uparrow$	PSNR $\uparrow$	Corr $\uparrow$	SNR $\uparrow$
ConditionGAN [ 94 ]	2022	0.52	20.02	0.81	3.33
SeisDDPM [ 23 ]	2023	0.41	16.14	0.61	0.68
SeisGAN [ 50 ]	2023	0.30	15.66	0.46	2.08
SeisResoDiff [ 95 ]	2024	0.30	16.05	0.70	0.59
SIST [ 22 ]	2024	0.65	19.00	0.77	3.51

Table 3: Gas chimney enhancement results.

Seismic Imaging

See all 15 papers

Domain-Guided Prompting of the Segment Anything Model for Seismic Interpretation: The Role of Attributes, Visualization, and Hybrid Prompts
Jun 14, 2026Aniq Ahmad, Heather Bedle, Ahmad MustafaPromptable Image Segmentation Vision Foundation Model Adaptation
School of Geosciences, University of Oklahoma, Norman, Oklahoma, USA · King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia
Reservoir property image slices from the Groningen gas field for image translation and segmentation
May 5, 2026Abdulrahman Al-Fakih, Nabil Sariah, Ardiansyah Koeshidayatullah +1Image-to-Image Translation
Department of Geosciences, King Fahd University of Petroleum and Minerals, Dhahran, 31261, Saudi Arabia. · Center for Integrative Petroleum Research, King Fahd University of Petroleum and Minerals, Dhahran, 31261, Saudi Arabia
OpenSeisML: Open Large-Scale Real Seismic and well-log Dataset for Generative AI
May 19, 2026Ipsita Bhar, Huseyin Tuna Erdinc, Thales Souza +2Synthetic Data Generation Training Data Curation
Georgia Institute of Technology, Atlanta, USA · Osokey Ltd, Henley-on-Thames, UK

Jun 14, 2026J/K move · Enter open · S save

Domain-Guided Prompting of the Segment Anything Model for Seismic Interpretation: The Role of Attributes, Visualization, and Hybrid Prompts

Aniq Ahmad, Heather Bedle, Ahmad Mustafa

School of Geosciences, University of Oklahoma, Norman, Oklahoma, USA · King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia

Promptable Image Segmentation Vision Foundation Model Adaptation Zero-Shot Referring Image Segmentation Semantic Segmentation Prompt Engineering

Open paper AskPDF arXiv

The advent of large pretrained foundation models for computer vision has significantly improved the efficiency of visual data interpretation. The Segment Anything Model (SAM), in particular, offers powerful zero shot segmentation capabilities through prompt based interaction, thus making it a promising tool for seismic interpretation. However, most existing applications of SAM rely on fine tuning for specific geological targets, which requires extensive labeled data, incurs high computational cost, and often compromises the model's generalization capability. In this study, we introduce a principled framework for zero shot adaptation of foundation models to seismic data. The framework is built on two key components: (1) aligning seismic attributes and visualization choices (e.g., colormaps) with the geological target of interest, and (2) employing a hybrid prompting strategy that combines sparse user defined point prompts with dense mask prompts derived from SAM's internal feature activations. We systematically evaluate this framework across multiple geological targets, datasets, prompt configurations, and seismic attribute representations. Our results demonstrate that geologic target aware selection of seismic attributes and colormaps, combined with hybrid prompting, enhances the separability of geological features and improves boundary delineation and segmentation accuracy relative to point based prompting alone. Our findings show that, when these components are jointly applied, SAM can achieve competitive segmentation performance in a fully zero shot setting, thereby eliminating the need to retrain SAM for each geologic feature. This work establishes a practical and scalable pathway to leverage foundation models in seismic interpretation, reducing reliance on labeled data while preserving model generality.

Abstract

Seismic Imaging

Explore similar work

Domain-Guided Prompting of the Segment Anything Model for Seismic Interpretation: The Role of Attributes, Visualization, and Hybrid Prompts

Reservoir property image slices from the Groningen gas field for image translation and segmentation

OpenSeisML: Open Large-Scale Real Seismic and well-log Dataset for Generative AI