Ecological niche modeling (ENM)

ENM was carried out in MaxEnt 3.4.1 [140, 141] to predict the potential distribution range of F. hispida in the Last Interglacial (LIG: ca.120–140 kya before present), Last Glacial Maximum (LGM: ca. 21 kya before present), Mid-Holocene (MIH: ca. 5 kya before present), present (1970–2000) and future (2070). We recently performed ENM analysis of F. heterostyla [62]. In total, 396 occurrence records of F. hispida across Indo-Burma were obtained from the GBIF (Global Biodiversity Information Facility, https://www.gbif.org), the CVH (Chinese Virtual Herbarium, https://www.cvh.ac.cn) and our field expeditions. The 19 bioclimatic variables corresponding to the five focal periods were downloaded from the WorldClim database (http://www.worldclim.org). Pairwise correlations of the 19 variables were tested to avoid variable multicollinearity and model overfitting. Variables with Pearson correlation coefficients of |r| ≤ 0.8 were used for subsequent analyses. 75% of the occurrence records were used as training data, and 25% were used as test data in 10 replications. MaxEnt outputs represent the habitat suitability, ranging from 0.0 to 1.0, in Cloglog format. The accuracy of each model prediction was evaluated statistically using the area under the receiver operating characteristic (ROC) curve [142]. The AUC ranges from 0 to 1, where a score above 0.7 is considered an indicator of good model performance [143]. The importance of each climatic variable for explaining the potential distribution was determined by jackknife resampling of the training and test gains. In addition, we matched the potential distribution to the ecoregions to determine which ecoregions were suitable for F. hispida. The map of ecoregions was derived principally from patterns of rainfall, temperature, geological history, broad vegetative patterns and expert opinion on community distributions [144, 145].