3.1 Ecological niche modelling
Evaluation of model performance based on both training and test sample data indicated that the models had high predictive power (AUC = 0.9976 and 0.9966, respectively). Results yielded a continuous geographical distribution of C. chuniana across several mountain ranges in southern China during the LIG (Figure 2a). In contrast, during the LGM the geographical distribution contracted into three fragmented areas (Figure 2b). We compared these data to a vegetation map of the LGM (http://intarch.ac.uk/journal/issue11/2/map/download_page_js.htm), based on which the three fragmented distribution areas were located mainly inside forest steppe (number 7) and partially in semi-arid temperate woodland or scrub (number 3) (Ray & Adams, 2001). The geographical distribution of C. chuniana was inferred to have expanded widely during the MH, occupying most of the Chinese subtropical region. No changes in the geographical distribution were evident between MH and current. Precipitation in April, May and June was revealed to be the most influential climate factor for suitable distribution range ofC. chuniana (Table S1).