Enhanced CO2 leak detection in soil: High-fidelity digital colorimetry with machine learning and ACES AP0

Abstract

The importance of effective carbon capture and storage (CCS) in addressing climate change issues highlights the need for robust CO₂ leak monitoring systems. Limitations of conventional methods have prompted interest in alternative approaches, such as optical CO₂ sensors, which offer non-invasive and continuous monitoring. Here, we present a novel methodology for high-fidelity digital colorimetry to enhance CO₂ leak detection in soil, integrating machine learning algorithms with the ACES AP0 color space. Optical CO₂ sensors, utilizing a cresol red-based detection solution, were calibrated and validated in a controlled environment chamber designed to simulate CO₂ leakage. Digital images of the sensor's colorimetric response to varying CO₂ levels were analyzed in five color spaces. The ACES AP0 color space, renowned for its expansive color gamut and perceptual uniformity, exhibited optimal performance in discerning subtle color variations induced by changes in CO₂ concentration. Ten machine learning regression models were evaluated, and Multivariate Polynomial Regression (MPR) emerged as the most effective in converting ACES AP0 color data into precise CO₂ concentration estimates, achieving a Mean Absolute Percentage Error (MAPE) of 2.9 % and a Root Mean Square Error (RMSE) of 0.0731. Field validation at a carbon capture and storage (CCS) facility corroborated the robustness and accuracy of this method, showcasing its potential for real-world applications in CCS and environmental monitoring.