Integrating advanced frequency-domain signal processing with machine learning for accurate leak detection in subsurface CO2 storage - Article Abstract

Aug 5, 2026

Ensuring the integrity of geological CO2 storage is essential for long-term CCS success, yet rapid and accurate detection of potential leakage remains difficult under limited monitoring data. This proof-of-concept framework integrates frequency-domain signal processing with machine learning using only pressure data from a monitoring well. Pressure signals are transformed via FFT to extract physically meaningful features sensitive to leakage effects, enabling a two-stage pipeline: leak/no-leak classification and leak localisation regression. Results show strong detection performance with Naive Bayes and Random Forest, and high localisation accuracy using KNN and Gradient Boosting, with improved sensitivity and spatial inference versus time-domain methods.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

AccessFree to download

Introduction

CCS importance and leakage risks

Leakage pathways: natural and anthropogenic

Proposed framework

Frequency-domain feature extraction via FFT

Two-stage machine learning pipeline

Results and performance

Leak detection classification

Leak localisation regression

Discussion and implications

Sensitivity improvement vs time-domain methods

Computational efficiency and real-time integration

145 (2026) 205798 Contents lists available at ScienceDirect Gas Science and Engineering journal [homepage:](homepage: www.journals.elsevier.com/gas-science-and-engineering)[ www.journals.elsevier.com/gas-science-and-engineering](homepage: www.journals.elsevier.com/gas-science-and-engineering) | Integrating advanced frequency-domain signal processing with machine learning for accurate leak detection in subsurface CO2 storage
Saeed Haratia , Sina Rezaei Gomaria,* , Mohammad Azizur Rahman b , Rashid Hassan c, Ibrahim Hassand, Ahmad K. Sleitie, Matthew Hamilton f
a School of Computing, Engineering, and Digital Technologies, Teesside University, Middlesbrough, TS1 3BX, UK b College of Science and Engineering, Hamad Bin Khalifa University, Education City, Doha, Qatar c Department Petroleum Engineering, Texas A&M University, Texas, College Station, TX, 77843, USA d Department Mechanical Engineering, Texas A&M University at Qatar, Doha, 23874, Qatar
e Department Mechanical Engineering, Qatar University, Doha, 2713, Qatar
f Department of Computer Science, Memorial University of Newfoundland, St John’s, NL, A1C 5S7, Canada | | | | | --- | --- | --- | --- | | A R T I C L E I N F O | | A B S T R A C T | | | Keywords:
Subsurface CO2 storage Leak detection
Fast fourier transform Machine learning Pressure signal analysis | | Ensuring the integrity of geological CO2 storage is critical for the long-term success of carbon capture and storage (CCS) technologies. The detection and localisation of potential leakage events rapidly and accurately remains a key challenge, particularly under constraints of limited monitoring data. This study presents a proof-of-concept framework that integrates advanced frequency-domain signal processing with machine learning to address this challenge using only pressure data from a monitoring well in a CO2 storage site. Here, pressure signals are translated into the frequency domain using Fast Fourier Transform (FFT) in order to extract physically meaningful features that are highly sensitive to leakage phenomena. These features capture subtle variations in signal behaviour that are often missed in time-domain analysis. A two-stage machine learning pipeline is also developed, involving a classification stage to distinguish leak versus no-leak conditions, followed by leak localisationin a regression stage.
The results showed that in the leak detection stage, ensemble and probabilistic classifiers, particularly Naive Bayes (test accuracy = 0.9873, F1 = 0.9788) and Random Forest (test accuracy = 0.9823, F1 = 0.9016), outperformed linear models by a substantial margin. In the localisation stage, the K-Nearest Neighbours Regressor (test R2 = 0.9899, MAE ≈ 6.8 m) and Gradient Boosting Regressor (test R2 = 0.9790, MAE ≈ 9.5 m) achieved the highest spatial prediction accuracy. Additionally, the findings demonstrate that frequency-domain feature engineering substantially enhances leak-detection sensitivity and spatial inference accuracy compared to timedomain methods. The proposed framework is computationally efficient, requiring only sparse pressure data, and can be integrated into real-time monitoring systems. | | 1. Introduction Carbon capture and storage (CCS) is a critical technology in the mitigation of greenhouse gas emissions to combat climate change. By capturing CO2 emissions from industrial sources and securely storing them in geological formationsthe potential to significantly reduce atmospheric carbon levels while supporting t, CCS has he transition to a low-carbon economy (Bui et al., 2018). The success of such projects begins with rigorous site selection, involving multi-criteria decision--making approaches to identify suitable geological formations that can ensure long-term containment (Harati et al., 2024c). However, even with optimal site selection, ensuring the long-term containment of injected CO2 is paramount, since leakage through faults, abandoned wells, or comp

FAQs

What problem does the study address in subsurface CO2 storage?

The study targets rapid and accurate detection and localisation of potential CO2 leakage events, especially when monitoring data are limited.

How are pressure signals processed before machine learning?

Pressure data are converted to the frequency domain using Fast Fourier Transform (FFT) to obtain physically meaningful features sensitive to leakage-related signal variations.

How is the machine learning pipeline structured?

It uses a two-stage approach: first a classification model to distinguish leak vs no-leak, then a regression model to localise the leak.

Aug 5, 2026

Upload by Clementine