Physics-informed AI for real-time quantification of methane emissions using optical gas imaging

Accurate quantification of methane emissions is essential for generating reliable emissions inventories, fundamental for regulatory compliance and effective climate action. In most cases, gaining a deeper understanding of emissions and linking them to the activity factors of equipment and assets provides highly valuable insights for optimizing their operational management, delivering far greater value than mere environmental compliance. Estimation-based approaches often lack precision, making it necessary to adopt direct measurement methods that enable continuous, real-time monitoring across industrial assets. This study presents a hybrid methodology that combines advanced InfraRed (IR) image processing with a physicsinformed artificial intelligence (AI) model to estimate methane emission rates from gas IR imagery, as well known as Optical Gas Imaging (OGI). The goal is to enable 24/7 autonomous monitoring with high sensitivity and accuracy. One of the main challenges in this task lies in the inherently low radiometric contrast between methane and the background. Often, the gas plume signal is at or below the noise level of the IR sensor, making it extremely difficult to extract reliable and consistent features. As such, the approach focuses on refining algorithms to isolate only the most physically meaningful parameters — such as plume geometry, thermal contrast, and motion— while filtering out background noise and irrelevant variations. These extracted features are passed through a radiometric model that translates them into quantitative descriptors of the plume’s thermodynamic behavior. This physical data serves as input to an AI model designed to integrate empirical learning with the physics of gas dispersion, ensuring robustness and generalization across different scenarios. The system is validated on both experimental and real-world datasets, showing improved consistency and accuracy compared to traditional methods. Its real-time processing capability removes the need for offline analysis, enhancing operational responsiveness. By combining physics-based modeling with advanced AI, this work addresses the key challenges of low-contrast detection. It supports the development of scalable, autonomous systems for continuous methane emissions quantification.