Abstract
This paper presents the first continuous, spatially-explicit reconstructions of the palaeodistributions of 65 plant species found regularly in association with early agricultural archaeological sites in West Asia, including the progenitors of the first crops. We used machine learning to train an ecological niche model of each species based on its present-day distribution in relation to climate and environmental variables. Predictions of the potential niches of these species at key stages of the Pleistocene–Holocene transition could then be derived from these models using downsampled data from palaeoclimate simulations. Our models performed well against independent contemporary test data, but their ability to predict the occurrence of specific species at archaeological sites was much more variable, probably reflecting a tendency of the method to underestimate the species’ fundamental niche. Nevertheless, the majority of species are predicted to have had more restricted geographic distributions under past climate conditions compared to today. Crop progenitors and several wild food species are modelled to have been concentrated in the Levant and, to a lesser extent, Cyprus and Western Anatolia, suggesting these regions may have served as glacial refugia. The average size of species’ niche shrunk by an average of c. 25% from the terminal Pleistocene to the Early Holocene, indicating that economically significant plants were adapted to cryo-arid conditions and did not, as often assumed, initially respond positively to the ‘ameliorated’ climate of the Holocene.