Database of tropical weeds functional traits based on the species list of the Wiktrop portal and built using an automatic extraction with a conversational large language model from online botanical desctriptions of species

This database have been compiled based on the weed species list available on the Wiktrop online portal (https://wiktrop.org/fr, gathering 1049 species when the database was created). For each species, we used a conversational Large Language Model (LLM) to automatically extract functional traits values from different online sources (https://wiktrop.org/fr, https://www.worldfloraonline.org/, https://www.feedipedia.org/, https://tropical.theferns.info/, https://www.wikipedia.org/). The collected data has been cured and filtered to remove traits with a fuzzy definition and non realistic trait values while keeping only traits jugged as ecologically relevant (however, as the extraction method is based on an automatic process involving a LLM, and as the entire list of trait values are not human-verified, the dataset may partially contain spurious or erroneous values). For detailed methods, please refer to the related document (https://agritrop.cirad.fr/614130/).