Long-term agricultural field experiments (LTEs) provide essential data spanning decades for crop modeling and understanding agricultural processes under global change. Despite their scientific value, broader reuse of LTE data is limited by two factors: each experiment focuses on a narrow set of experimental factors, and many fail to follow FAIR data principles. The LTE‑Overview Map (https://bonares.de/ltfe/) has made significant progress by assembling comprehensive metadata for roughly 680 long‑term experiments worldwide. This resource now enables synthesis studies that combine multiple LTEs. However, the metadata collection remains incomplete and relies on time‑consuming manual curation. To overcome these bottlenecks, we developed an automated pipeline that combines large language models (LLMs) with rigorously defined Pydantic data schemas. Implemented as a Python workflow, the system parses the scientific literature associated with each LTE, extracts relevant experimental descriptors, and translates them into standardized, interoperable metadata records. We also extract the research context and specific objectives for which LTE data are used in each scientific paper. This information serves as an initial indicator for evaluating data fitness‑for‑purpose and tracking LTE data reuse. Beyond the structural checks imposed by the Pydantic schemas, we are currently implementing semantic verification and LLM-as-judge checks with additional human supervision. We will present results of these approaches, which enable standardized evaluation of the results. This LLM-driven approach towards metadata management not only accelerates metadata curation and enhances data discoverability but also provides the basis for a more comprehensive understanding of how LTE data are used across the agricultural research community.
Lachmuth et al. (Mon,) studied this question.