33 Advances in Subsurface Characterization for Energy Exploration and Utilization

Advances in the characterization of the subsurface for applications in both oil and gas, and geothermal energy (conventional, AGS, and EGS), carbon capture and storage (CCS), energy storage, and other low-carbon technologies.

Wednesday, 20 May 2026
8:00 AM - 10:20 AM
Alvarado A

Estimation of Seismic Velocity and Geomechanical Properties from Triple Combo Logs Using Machine Learning: Application in the San Juan Basin for CCS

Paper number: 33-4
Program: Technical Sessions
Time: 9:05 AM – 9:25 AM

Authors

Abstract

Accurate subsurface velocity models are critical for reservoir characterization and seismic interpretation, however, direct sonic log measurements are frequently incomplete due to operational limitations, cost constraints, and partial wireline coverage. This study presents a robust data-driven framework for predicting compressional-wave velocity (Vp) using widely available triple-combo logs and advanced machine learning algorithms across the San Juan Basin. Characterized by thick Mesozoic sedimentary sequences and extensive depleted hydrocarbon fields, the basin requires high-resolution subsurface models to support emerging carbon capture and storage (CCS) initiatives and broader energy-transition applications. A comprehensive multiwell dataset penetrating the Dakota through Wingate formations was assembled, incorporating gamma ray, bulk density, and resistivity logs alongside measured compressional-wave velocities for supervised training. To rigorously evaluate model generalizability, selected wells were withheld as blind-test cases. The predictive workflow included systematic preprocessing steps such as outlier filtering using statistical thresholds, feature normalization to ensure scale consistency, logarithmic transformation of resistivity (Rt) to address skewness, and detailed statistical assessment of feature distributions to enhance model stability. Multiple machine learning algorithms were benchmarked, including Random Forest, Artificial, Neural Networks, Support Vector Machines, Decision Trees, Boosted Trees, and Linear, Regression. Model performance was evaluated using the coefficient of determination (R²) and root-mean-square error (RMSE), with hyperparameter optimization performed through cross validation to minimize overfitting and improve predictive reliability. Among the evaluated models, the Random Forest algorithm consistently delivered superior performance, achieving the highest R² values and lowest RMSE across both validation and blind-test wells. The optimized model was subsequently applied to generate basin-wide compressional velocity predictions, enabling the reconstruction of high-resolution velocity profiles in intervals lacking measured sonic data. The results demonstrate that machine learning-enhanced prediction using conventional triple-combo logs provides a scalable, accurate, and cost-effective solution for velocity modeling in data-limited subsurface environments. The estimated parameters and properties are vital for the effective implementation of CCS operations, to ensure caprock integrity, assess fault reactivation risk, guide safe injection strategies for long term CO₂ containment, and to improve CO₂ plume imaging and monitoring in the San Juan Basin.  

Geological Society of America Abstracts with Programs. Vol. 58, No. 4

doi: 10.1130/abs/2026RM-410705

© Copyright 2026 The Geological Society of America (GSA), all rights reserved. Permission is hereby granted to the author(s) of this abstract to reproduce and distribute it freely, for noncommercial purposes. Permission is hereby granted to any individual scientist to download a single copy of this electronic file and reproduce up to 20 paper copies for noncommercial purposes advancing science and education, including classroom use, providing all reproductions include the complete content shown here, including the author information. All other forms of reproduction and/or transmittal are prohibited without written permission from GSA Copyright Permissions.