Answers are grounded in this paper and cite their sources.
Demonstrates AI effectively predicts helium concentrations, indicating potential for resource exploration.
This study presents a robust artificial intelligence (AI)–based framework for predicting subsurface concentrations of hydrogen (H 2 ) and helium (He), with a focus on the Devonian aged Duvernay Formation in Alberta, Canada. An initial assessment of over 641,000 wells across 10 geological formations revealed that He, owing to its near‐Gaussian distribution and moderate correlation with production variables, was more amenable to modeling than H 2 . In contrast, H 2 exhibited a highly skewed distribution, weak feature correlations, and spatial alignment with geological structures along the Rocky Mountain Fold and Thrust Belt, limiting its predictability. Among the formations, the Duvernay Formation was selected as the primary modeling target because of its data quality, spatial coherence, and geological relevance. A stacked ensemble learning approach that integrates tree‐based, kernel‐based, linear, and neural network models was applied using production‐related features. The He prediction model achieved strong performance with a coefficient of determination ( R 2 ) exceeding 0.9. In contrast, the H 2 model demonstrated limited accuracy, underscoring the need for additional geologic and geochemical inputs. Unsupervised clustering via t ‐distributed stochastic neighbor embedding ( t ‐SNE) further revealed distinct patterns associated with elevated He zones, offering geospatial insights beyond direct measurement. These results highlight the feasibility of AI–based prediction of subsurface He using routinely collected production data, while emphasizing the limitations of similar approaches for H 2 . The framework offers a scalable, interpretable, and cost‐effective tool to guide early‐stage exploration for noble and clean energy gases in sedimentary basins.
No takes yet. Share an insight, caveat, or question.
Kim et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: