Past, present and future distributions of Oriental beech (Fagus orientalis) under climate change projections

Results Model evaluation and environmental parameters relevant for Fagus orientalis growth in Asia Minor The results are indicating that models were good in distinguishing presence data from the background based on their AUC values varying from 0.79 to 0.99 (all > 0.7, Table 3, [57]). According to this, the best-fit algorithm was RF and the worst-fit was BIOCLIM. In the follow- ing, we will focus on the RF algorithm provided with ‘biomod2’, which fitted best for F. orien- talis. Successful model results from the present climate could be used to identify and evaluate optimal environmental conditions for F. orientalis growth. The projections for the past, MH and LGM, and for the future, 2050 and 2070, gave different distributions with different climate change scenarios and global climate models (S1–S10 Figs). Past projections Projections for the LGM at 21 ka and the MH at 6 ka from two different global climate models gave consistent results. Past projection outputs of RF are shown with the pollen records col- lected from published palynological studies of Fagus during LGM and MH (Fig 2). The model outputs show that the possible distribution of Fagus was mainly from the mid-Black Sea region to Caucasia and northern Iran to southern Turkmenistan during the LGM (Fig 2A). Addition- ally, RF predicted a refuge area in the Amanos region during the LGM. For the MH, the popu- lations ranging from the mid-Black Sea region to Caucasia remained in these regions, whereas, the Iranian/Turkmen region almost lost its population, except some patches. During the MH,

species distribution shifted towards the west, especially towards Europe and western Anatolia (Fig 2B). Outputs arising from other algorithms are shown in S1–S10 Figs. Late Quaternary palynological records of Fagus in Asia Minor A general feature of the 30 pollen profiles containing Fagus is the sparse occurrence or absence of this genus during the LGM and its more or less continuous presence since about 10 ka (Fig 2 and Table 4). LGM values are usually low indicating long distance dispersal (LDD). One Table 2. Bioclimatic variables (precomputed in WorldClim dataset) used as environmental input in the models. Abbreviations Bioclimatic variables Unit BIO1 Annual Mean Temperature ˚C BIO2 Mean Diurnal Range ˚C BIO3 Isothermality - BIO4 Temperature Seasonality ˚C BIO8 Mean Temperature of Wettest Quarter ˚C BIO9 Mean Temperature of Driest Quarter ˚C BIO12 Annual Precipitation mm BIO15 Precipitation Seasonality mm BIO19 Precipitation of Coldest Quarter mm https://doi.org/10.1371/journal.pone.0242280.t002 Table 3. AUCtest values of all the algorithms, bioclim, maximum entropy (MaxEnt), generalized additive model (GAM), generalized linear model (GLM), and random forest (RF) performed with present climate conditions (1960–1990). Algorithm BIOCLIM MaxEnt GAM GLM RF AUCtest 0.79 0.94 0.96 0.93 0.99 https://doi.org/10.1371/journal.pone.0242280.t003 PLOS ONE Species distribution modeling of Oriental beech (Fagus orientalis) PLOS ONE | https://doi.org/10.1371/journal.pone.0242280 November 17, 2020 6 / 19