2026 GeoGulf Transactions, Vol. 75
TITLE
Curated AI Data Workflows for Low Carbon Energy
AUTHOR(S)
Jess B. Kozman and Jack Bashian
ABSTRACT
Access to high-quality and well-curated data for AI/ML training and consumption requires legacy databases that reflect modern data management best practices. This enables critical business decisions, interoperating with digital data to support low-carbon energy projects like geothermal, mineral extraction, and carbon sequestration, using high quality, well-curated, and easily accessible data. Recent experiences with generative AI/ML have highlighted geotechnical data sets for low-carbon projects that can lack integration, governance, and data deliverability biases, and gaps. Published government datasets may have as little as 20% of geoscience data that is fully Findable, Accessible, Interoperable, and Reusable (FAIR), while spatial and temporal biases in geotechnical datasets can misrepresent the true distribution of earth properties critical to supporting business decisions. Value added aggregation, standardization, and curation of these datasets can ensure that they are measurably more fit for purpose for AI/ML workflows. This has the potential of saving operators from the frequently cited first experience with AI/ML where the results indicate that the data is not well curated or standardized enough to support the data driven queries submitted.
Mitigating the impact of fragmented data sources, inconsistent data quality and insufficient data interoperability will improve AI/ML deployments for re-use in low carbon energy projects using aggregated geotechnical data sets. Accuracy of responses from AI Augmented SQL queries can be increased by automating metadata extraction and supporting the use of open-source industry standard data models. Data quality and ingestion workflows allow analysis of metadata patterns and data set usage behaviors to improve deliverability.
Subsurface energy data aggregators and publishers can deploy optimum industry accepted data management standards to support AI/ML. Deployments can be evaluated for data availability and deliverability, measuring how delivering FAIR data can improve AI metrics. Key data types available on platforms that deliver FAIR data can increase the amount of value-added time available on projects. This can be leveraged to increase the volumes of data that are fit-for-purpose for subsurface multi-resource data-driven models.
PAGE(S)
447-468
DOI
https://doi.org/10.62371/VQLX8654
Purchase Link via the Bureau of Economic Geology Bookstore is not yet available. Check back later for update. |