Geological Structure Embeddings from a Pretrained Diffusion Representation Model Applied to the Sleipner 4D CO2 Seismic Benchmark

Author(s)	Mr. Sunil Manikani, Dr. Murad Yaghi
Country	India
Abstract	We investigate whether a pretrained diffusion representation model (PDRM) trained purely to denoise seismic amplitude images acquires an embedding space that is more discriminative than that of a structurally matched autoencoder. Using the publicly available Sleipner CO2 seismic benchmark (249 inline cross-sections), we train a compact UNet-based PDRM and compare its embedding quality against an equivalent autoencoder capacity machine (EACM) with an identical 128-dimensional bottleneck. PDRM embeddings extracted at noise timestep t=100 yield substantially richer representations: Silhouette 0.363 vs 0.103, Davies-Bouldin 1.316 vs 2.952, and 52.6% vs 15.2% of total PCA variance in two components. An elbow analysis over k=2-8 and a timestep sweep over t in {50,100,200,400,800} show the PDRM advantage is consistent across cluster counts and robust in the range t=50-200. Critically, identified clusters are treated as data-driven groupings of seismic image character; correspondence to geological units is not established and requires independent validation against well data.
Keywords	diffusion, deep learning, ccs
Field	Computer > Artificial Intelligence / Simulation / Virtual Reality
Published In	Volume 7, Issue 4, July-August 2026
Published On	2026-08-01

About AIJFR Fees & Payment Current Issue Publication Archive	Submit Research Paper Track Submission Status Publication Guidelines Publication Ethics Peer Review & Plagiarism	Join as a Reviewer Editors & Reviewers Reviewer Referral Program Get Reviewer Membership Certi.	Website/Journal Policies Usage Policy Content Policies Privacy Policy

Contact Us		+91-9898-79-39-59	editor@aijfr.com

Advanced International Journal for Research