High-resolution 4D time-lapse seismic monitoring of CO2 sequestration sites using machine learning for improved reservoir characterization

Project Abstract

High-resolution 4D time-lapse seismic monitoring combines repeated 3D seismic surveys with advanced machine learning techniques to reveal dynamic subsurface changes associated with CO2 injection and migration in sequestration sites. This study integrates multi-physics data, including seismic amplitude variations, ?p-wave and ?s-wave attributes, and rock physics models, to quantify CO2 plume geometry, saturation, pressure evolution, and rock property alterations with unprecedented temporal accuracy. We develop a robust workflow that includes careful survey design optimization for repeatability, pre-stack and post-stack data conditioning, and noise attenuation strategies tailored to carbonate and clastic reservoir environments commonly encountered in sequestration projects. A data-driven inversion framework leveraging deep neural networks and physics-informed learning is employed to infer spatially distributed changes in porosity, permeability, and fluid thresholds from 4D seismic responses, while mitigating non-CO2 related variability such as temperature drift, brine migration, and mechanical compaction. The methodology emphasizes transfer learning across vintages and sites, enabling rapid generalization to new sequestration locations with limited labeled data. We introduce a hierarchical Bayesian approach to quantify uncertainties in plume extent, saturation, and property changes, providing decision-makers with probabilistic risk assessments for containment integrity and long-term monitoring strategies. Validation is conducted against synthetic benchmarks generated from high-fidelity reservoir simulations and cross-validated with independent monitoring datasets, including electromagnetic, gravity, and borehole log measurements. The research investigates the sensitivity of seismic signatures to CO2 phase state transitions, residual trapping, and cap rock integrity, elucidating the relationships between impedance contrasts, fluid substitute properties, and poroelastic responses. A case study at a mature geological storage site demonstrates real-time detection of early breakthrough, plume pinching, and regional leakage pathways, accompanied by quantitative estimates of CO2–brine displacement efficiency and pressure buildup dynamics. The study also evaluates the impact of time-lapse acquisition geometry, vendor-imposed processing workflows, and incomplete vertical coverage on interpretability, providing practical guidelines for cost-effective long-term monitoring programs. The outputs include (i) a scalable 4D seismic workflow optimized for CO2 sequestration settings, (ii) machine learning models capable of robust CO2 saturation and property estimation with quantified uncertainty, and (iii) decision-support dashboards that translate geophysical changes into actionable reservoir management insights. By bridging geophysical measurement fidelity with data-driven inference, the research advances reservoir characterization under dynamic geofluid processes, enhances confidence in sequestration performance assessments, and supports regulatory compliance through transparent, traceable monitoring frameworks.

Project Overview

What This Project Is About

---

A plain-language overview of the topic and what the project investigates.

---

---

The Problem It Addresses

---

What problem or gap this project tackles and why it matters to the field or society.

---

---

Objectives of the Project

---

1. Understand how 4D seismic data can reveal changes in a storage site over time.
2. Explore how machine learning can help interpret seismic signals faster and more accurately.
3. Develop a simple workflow to monitor CO2 injection and plume movement.
4. Assess uncertainties in the monitoring results and suggest improvements.

---

---

What You Will Do Step by Step

---

1. Learn basic concepts of seismic data and CO2 sequestration in simple terms.
2. Collect or simulate a small set of 4D seismic data from a CO2 site.
3. Preprocess data to remove noise and align time steps.
4. Apply basic machine learning methods to detect changes between surveys.
5. Visualize results to show where CO2 may have moved.
6. Evaluate the accuracy of the interpretations with straightforward checks.

---

---

Expected Outcome

A clear, easy-to-read monitoring workflow and a set of maps or plots showing CO2 movement over time, along with notes on reliability and limitations.