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

Discussion Model evaluation for different SDM algorithms Modelling the spatial distribution of F. orientalis provided details of the species responses to different climate change scenarios and identified possible past refugia by means of projected past distributions and palynological data. A key assumption of SDM with bioclimatic variables is that the ranges of the modelled species are in equilibrium with climate [25, 29, 83]. In gen- eral, the modelled present distributions fitted reasonably well with the observed distribution data, which allowed us to use the trained model for past and future simulations. As mentioned before, all models ran successfully according to their AUC values (Table 3), and the combina- tion of regression and classification approaches, as well as the in-algorithm verification process provided more accurate outcomes in RF and made this algorithm the most successful one [84]. On the other hand, according to the AUC values (Table 3) the least successful algorithm was BIOCLIM. Due to the advanced characteristics and calculations of other algorithms, it was expected that they would give a better fit than BIOCLIM. One problematic algorithm was GAM (S5 and S6 Figs) with a very wide distribution of Oriental beech projected for LGM and MH. Although GAM has the second-best AUC value, we interpreted this as an overfitted dis- tribution. It is known that this algorithm is highly sensitive to large sample size since the fitted functions are not constrained to any functional form when sample size increases [85]. Further,

it is noteworthy that while there are substantial differences between modelling performances from different algorithms, the outcomes are consistent, which is a hint for repeatability [86]. The present distribution of Oriental beech The present distribution of Fagus orientalis in the RF algorithm/MIROC-ESM was largely con- gruent with the observed presence records (Figs 1 and 3A), which is also illustrated by its high AUC value. A range expansion in the European region (Fig 3A) most probably reflects similar present growing conditions of F. orientalis and its sister species F. sylvatica. Hence, it is not surprising that models recognize habitats of F. sylvatica as suitable for F. orientalis [8, 17]. Although there is a close match between observed and modelled distribution, it is interesting to note that the model predicted range expansion to the south of the Pontic Mountains (Kuzey Anadolu Dağları). Under present conditions, this is unlikely as the rain shadow south of the main ridge of the Pontic Mountains creates a sharp change from fully humid Cf climates (see [16, 87]) to Mediterranean Cs and continental, arid BS climates. Whereas, Fagus is replaced by Pinaceae at higher elevations, it is replaced by Quercus species (e.g. Quercus macranthera Fisch. et Mey.) both at higher elevations and in drier areas beyond the main ridge [88]. Table 5. Spatial cell numbers of contraction, overlap, and expansion shown in Fig 3 are quantified in this table. Period Future Scenario Expected Contraction Expected Contraction in model Overlap Expected Expansion Overlap with observed

occurrence data (%) Present – 115 – 10,372 9.444 98.85 2070 RCP 2.6 7,944 7919 3,954 1.558 37.68 RCP 4.5 9,009 8226 2,582 927 24.61 RCP 8.5 9,816 8914 1,087 446 10.36 Overlap of the species occurrence between the observed data and model outputs are given as percentages. In future projections, from optimistic to pessimistic scenario, contractions are increasing, overlap and expansion are decreasing. https://doi.org/10.1371/journal.pone.0242280.t005 PLOS ONE Species distribution modeling of Oriental beech (Fagus orientalis) PLOS ONE | https://doi.org/10.1371/journal.pone.0242280 November 17, 2020 9 / 19