Crop phenology data extraction from research papers using a large language model

To extract high-quality crop phenology observations reported in field experiments and research papers, we propose a semi-automated extraction method using the LLM-based online tool ChatPDF combined with a Python program. From 164 studies, phenological data for maize, soybean, wheat, and rice were extracted and evaluated against manual extraction. The method substantially reduces extraction workload, while issues remain such as errors in estimating latitude/longitude and harvest years, and omissions caused by ambiguous phenology terminology. We discuss these limitations and potential improvements toward large-scale implementation. Author contribution note :Toshichika Iizumi and Yohei Ono contributed equally as co-first authors.Takahiro Takimoto and CHAOGEJILATU contributed equally as co-second authors.