Bridging literature and models: a workflow for creating agricultural datasets for model application using AI

Xinxin Chen (Leibniz Centre for Agricultural Landscape Research, ZALF), Benjamin Leroy (Technical University of Munich, TUM), Jeffrey W. White (Agricultural and Biological Engineering Dept., University of Florida), Hans-Jörg Vogel (Helmholtz-Zentrum für Umweltforschung GmbH, UFZ), Gerrit Hoogenboom (Agricultural and Biological Engineering Dept., University of Florida), Senthold Asseng (Technical University of Munich, TUM), Frank Ewert (Leibniz Centre for Agricultural Landscape Research, ZALF), Heidi Webber (Leibniz Centre for Agricultural Landscape Research, ZALF) Bridging literature and models: a workflow for creating agricultural datasets for model application using AI Agricultural simulation models can provide insights into complex crop-soil interactions, potentially informing decisions around sustainable cropping system management. However, their use is limited by insufficient datasets for model testing. Published research describing experiments and presenting results are a valuable source of information for model development and calibration. •experimental information in papers is largely in unstructured prose, various terms and units, summary tables and figures •modelling applications require well-structured, machine-readable data with consistent vocabularies Background Challenges Main objectives 3 •develop an automated workflow to extract and harmonize data from published literature, generating datasets suitable for model development and calibration •explore LLMs-based structured data extraction in agriculture 2 1 Workflow 4