Climate change jointly with migration ability affect future range shifts of dominant fir species in Southwest China

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2019 The Authors. Diversity and Distributions published by John Wiley & Sons Ltd. 1CAS Key Laboratory of Mountain Ecological Restoration and Bioresource Utilization & Ecological Restoration Biodiversity Conservation Key Laboratory of Sichuan Province, Chengdu Institute of Biology, Chinese Academy of Sciences, Chengdu, China 2University of Chinese Academy of Sciences, Beijing, China 3Swiss Federal Research Institute WSL, Birmensdorf, Switzerland 4Laboratory for Climate and OceanAtmosphere Studies, Department of Atmospheric and Oceanic Sciences, School of Physics, Peking University, Beijing, China 5Department of Forest Sciences, University of Helsinki, Helsinki, Finland

For drawing some enlightenment into how future ACC will affect species distributions, conservationists have tried to figure out the effects of recent as well as historical climate fluctuations on species distributions through field observation (e.g. Kelly & Goulden, 2008;Zhu, Woodall, & Clark, 2012), palaeobotany (Jackson, Betancourt, Lyford, Gray, & Rylander, 2005;Liepelt et al., 2009) or molecular phylogeographical methods (e.g. McLachlan, Clark, & Manos, 2005;Naciri & Gaudeul, 2007). However, these approaches tend to concentrate on specific species at a small scale because of the costly and/or burdensome collection efforts.
Moreover, the methods mentioned above are not sufficient for predicting the potential changes in species distributions under future ACC.
Species distribution models (SDMs) (also referred to as ecological niche models, ENMs, see Franklin, 2009;Peterson et al., 2011) quantify the relationship between existing occurrence records and environmental factors through multiple algorithms and then generate species habitat suitability (Guisan & Zimmermann, 2000).
In the face of upcoming rapid climate change, it is unrealistic for plants to be able to evolve in correlation with physiological adaptation strategies in a short period, so the migration or dispersal ability is a prerequisite for their survival (Corlett & Westcott, 2013). However, since migration ability was not effectively taken into account in initial modelling (Guisan & Thuiller, 2005;Pearson & Dawson, 2003), the majority of SDM studies prefer to apply either full-or no-migration scenarios in predicting the species' potential distribution under future ACC (Araújo, Cabeza, Thuiller, Hannah, & Williams, 2004;Bateman, Murphy, Reside, Mokany, & VanDerWal, 2013). In fact, the actual future distributions of species are likely to lie somewhere between the no-and full-migration scenarios (Berg et al., 2010). The forecasts from SDMs without considering a realistic migration ability of species may hinder our accurate judgement of the real risks to species, by either over-or underestimating them (Alexander et al., 2018). Recently, more researchers have become aware of this problem and have tried to use various modelling approaches to incorporate more realistic migration (i.e. partial-migration scenarios) into SDMs, for example, future projections for plants or animals in South Africa (Midgley, Hughes, Thuiller, & Rebelo, 2006;Uribe-Rivera et al., 2017), Australia (Fitzpatrick, Gove, Sanders, & Dunn, 2008;Reside, Vanderwal, & Kutt, 2012), and Europe (Meier, Lischke, Schmatz, & Zimmermann, 2012;Saltre, Duputie, Gaucherel, & Chuine, 2015).
To some extent, in comparison to full-migration and no-migration scenarios, these predictions have improved the accuracy of modelling results.
Since the Quaternary period, the fluctuation in temperature between glacial and inter-/post-glacial periods induced a dramatic change in species distributions at the global scale (Hewitt, 2000). Fortunately, the mountains of Southwest China (MSWC) has served as refugia and enabled the survival of species during globally decreasing temperatures due to the relative stability of its local climate. The distribution of many species, for example Juniperus przewalskii (Zhang, Chiang, George, Liu, & Abbott, 2005), Picea crassifolia (Meng et al., 2007) and Abies species (Song et al., 2012) on the Qinghai-Tibet Plateau (QTP) is also closely related to the retreat or advancement of glaciers and the existence of refugia in MSWC. Additionally, mountains with a substantial elevation range offer short-distance corridors for the migration of species at a vertical gradient, and these species can recolonize when the temperature becomes more suitable again (Qiu, Fu, & Comes, 2011).
Currently, the MSWC is experiencing an unprecedented warming trend, which is much higher than the global average warming trend over the past half-century, posing a severe challenge to the

K E Y W O R D S
Abies forest, climate change, migration ability, mountains of Southwest China, potential distribution, SDMs survival of mountain plant communities (Alexander et al., 2018;Shi et al., 2015). Given the potential risks faced by species in the MSWC, it is necessary to simulate the distribution dynamics caused by ACC in advance.
To our knowledge, a few studies have compared current and future suitable habitats of firs in China at the genus level by SDMs, and they suggest that the ranges of firs will shrink and shift northward under future ACC (Liu, Wang, He, & Zhang, 2018;Shao, Zhang, Phan, & Xiang, 2017). Regrettably, when these SDMs were conducted, the influence of migration ability was ignored, and differences among species were masked by modelling the whole genus. Meanwhile, it is argued that plants in different regions will show individualized responses to ACC, depending on species-specific physiological tolerance (Lindner et al., 2014). Because of the strong inter-regional differences in topography, climate, vegetation and physical barriers, the feedback mechanism of plants to ACC in the MSWC is complex (Bellard et al., 2014). Therefore, there is an urgent need for a thoughtful future assessment of the multiple fir species of the MSWC, taking the interaction between migration limitation and climate change into account.
In the present study, we aimed to simulate climatically suitable areas (CSAs) and quantify the magnitude and direction of the changes in projected distributions of six fir species in the MSWC from the current period to 2061-2080 under different climate × migration scenarios. Due to the complex topography and climate conditions of the MSWC, we formulated three hypotheses: (I) assuming full migration, driven by different climatic factors, not all firs of this region will face a reduction in CSAs, and some may benefit from future climate change and obtain larger CSAs; (II) assuming partial migration, the predicted areas of newly colonized habitats of some fir species may be significantly reduced; and (III) not all species are expected to migrate northward in response to rapid future climate change. Considering these three hypotheses, we representatively selected six dominant fir species from different parts of the MSWC to predict current and project future habitat suitability using the Maximum Entropy (Maxent) model (Phillips, Anderson, & Schapire, 2006). Taking into account the uncertainty of future ACC, two emission scenarios, RCP 4.5 and RCP 8.5 (Meinshausen et al., 2011), were selected. In addition, we applied three migration scenarios for each fir's future prediction: full-, no-and partial-migration scenarios (Bateman et al., 2013).

| Study area and species occurrence data
The MSWC extend from the western Himalayas and the Yarlung Zangbo Canyon to the Hengduan Mountains and the western Sichuan Plateau (Royden, Burchfiel, & van der Hilst, 2008;Zhao, 1990). Here, vast ridges and valleys spread from the Qinghai-Tibetan Plateau to the western rim of the Sichuan Basin with altitudes ranging from a few hundred metres to more than 7,000 metres (Royden et al., 2008). In the "three parallel river-running areas" (i.e. Lancangjiang, Nujiang and Jinshajiang Rivers), most of the mountain ridges are oriented generally in a north-south direction, for example Boxoilaling-Gaoligong Shan, Taniantawen Shan-Nu Shan and Mangkang Shan-YunLing (Royden et al., 2008).
Their pronounced difference in elevation, heterogeneous geography and varied climate result in a high diversity of different vegetation types with high endemic species richness (~3,500 endemic vascular plant species) making the MSWC as one of the most important global hotspots of biodiversity (Li, 2018;Wu, 1980). For this study, the MSWC were identified as an area within the geographical coordinates of 20.6°-35.8°N and 85.4°-105.8°E  (Table S1).
However, some species do not have sufficient distribution data for successful modelling (Stockwell & Peterson, 2002). According to their current geographical distributions (Fan, 2006), six dominant and endemic fir species were selected and divided into three groups: "North"-Abies recurvata and Abies faxoniana, which mainly occur in the upper Minjiang River region; "Middle"-Abies  Table 1 and Appendix S1 in the supplemental files). Since some of these occurrences lacked geographical coordinates, Google Earth (http://ditu.google.cn/) served to complement the latitude and longitude information.
Moreover, any species occurrences that were based on introduction and cultivation were excluded. Given that most SDM methods require input data to be spatially independent for the model to perform well (Naimi, Skidmore, Groen, & Hamm, 2011), SDMToolbox (Brown, 2014; http://sdmto olbox.org/) was used to ensuring that only one occurrence record per species was used within each grid cell at a resolution of 30 arc-seconds (~ 1 km at the equator).
To avoid multicollinearity of variables (Graham, 2003), we examined the cross-correlation of the 22 variables using the "cor" function in R (R-Core-Team, 2015) and eliminated the highly correlated (|Pearson r| ≥ .8) climatic variables (Blach-Overgaard, Svenning, Dransfield, Greve, & Balslev, 2010). Because extreme temperature and humidity are often considered the most critical limiting factors affecting tree growth in alpine regions ), such variables were given priority in our study (see Table S5 for more details). Finally, out of the total 22 variables, only nine were selected as predictors, including BIO2 (mean diurnal range), BIO3 (isothermality), BIO4 (temperature seasonality), BIO5 (max temperature of warmest month), BIO11 (mean temperature of coldest quarter), BIO14 (precipitation of driest month), BIO15 [precipitation seasonality (coefficient of variation)], SR (solar radiation) and WS (wind speed) (Table S5).  Minjiang River D a d u h e R i v e r Y a lo n g ji a n g R iv e r J i n s h a j i a n g R i v e r L a n c a n g ji a n g R iv e r N u j i a n g R i v e r

Yalung Tsangpo
For global climate models (GCMs), we used BCC-CSM1-1 (Beijing Climate Centre, China Meteorological Administration), which is considered one of the more suitable GCMs for climate change research in China (Yang, Jiang, & Li, 2016). Two representative concentration pathways, RCP 4.5 (moderate emission scenario) and RCP 8.5 (pessimistic emission scenario), released by the IPCC Fifth Assessment Report (AR5) (IPCC, 2013;Meinshausen et al., 2011), were selected to represent the possible future climate scenarios. All future climatic layers were based on WorldClim v1.4 (Hijmans et al., 2005) at a 30 arc-second spatial resolution. For periods of 2041-2060 and 2061-2080 (Tables S3 and S4), the predictor layers were downloaded from WorldClim v1.4. Besides, we downloaded corresponding layers of the period 2021-2040 from CGIAR web portal (http://www.ccafs-clima te.org) for the subsequent migration analysis. Like other studies (Chakraborty, Joshi, & Sachdeva, 2016;Zhang, Yao, Meng, & Tao, 2018), we assumed SR and WS to remain unchanged when projected into the future. Finally, all raster data were extracted to the regional extent of the study area with ArcGIS 10.3 (Esri).

| Evaluation of current and future habitat suitability
We used the Maximum Entropy approach (Maxent version 3.3.3 k; Phillips et al., 2006) to calibrate and to predict the CSAs of each species for the current period and projected CSAs for future periods (2040, 2060 and 2080) based on the nine selected climatic predictors. To reduce uncertainty caused by sampling artefacts, we randomly divided distribution data into training data (75%) and validation data (25%). To validate the robustness of the models, replications of 20 times were carried out using the subsampling method, in which the presence points are repeatedly split into random training and testing subsets, and the results were finally averaged. The maximum number of background points was set to 10,000, and the maximum iterations were set to 5000 times for seeking the optimal solution, while we used for the remaining parameters default values (Morales, Fernández, Baca-González, & Yoccoz, 2017;Radosavljevic & Anderson, 2014). The generated suitability maps F I G U R E 2 Occurrence records (left side of each secondary figure) and predicted distribution of current CSAs (right side of each secondary figure) for the six studied fir species

| Incorporating migration ability
To simulate the effects of migration ability on the accessibility of future suitable habitat areas, three different migration rates were assigned to all species: full migration (unlimited m/year), partial migration (200 m/year, Cheddadi et al., 2014;Xu, 1998; see Appendix S2 in the supplemental files for more details) and no migration (0 m/ year). The full-migration (FM) scenario was obtained directly from the Maxent model default output applying the species-specific MTSS threshold, and it was the most optimistic assumption that species could colonize all suitable habitats under climate changes (Franklin, 2010). In contrast, the no-migration (NM) scenario was the most pessimistic scenario in that it assumed species could not migrate at all and only lose suitable areas as the climate changes. This scenario was achieved by restricting 2080 projections to the suitable pixels of the current predictive map (Franklin, 2010

| Quantify the magnitude and direction of range shifts
To quantify the magnitude of change in the projected distributions of each species across the two climate scenarios and the three mi-  TA B L E 1 Overview of distribution data and SDMs for each species investigated (e.g. Thurm et al., 2018;Zhang et al., 2018). Here, we calculated the centroids for both current and future species distributions and used these centroids to project a vector arrow to indicate the magnitude and direction of range shifts using the SDMToolbox (Brown, 2014).

| Important climatic factors and predicted current CSAs
The AUC values of all Maxent models were higher than 0.9 (Table S6 mean 0.988 ± 0.004), indicating that our SDMs had an excellent overall prediction ability. Among all variables, the three top-ranked factors were isothermality (BIO3), temperature seasonality (BIO4) and solar radiation (SR), whose cumulative relative importance to all species exceeded 65% ( Table 2). The response curves of all variables can be found in the supplemental files ( Figures S2 and S3). However, these three climatic variables varied considerably in their contribution rate among the species groups ( Figure 3; Table 2). In particular, there was a remarkable difference in the contribution of BIO4 and SR between the "South" and "North" groups of species. Besides, for the "Middle" group of species, the three variables (BIO3, BIO4, and SR) had an almost equal contribution in shaping their distributions ( Figure 3).
Overall, the species showed individualistic differences in their current CSAs (Figure 2). Among the six species, A. recurvata had the smallest CSAs (ca. 1.23 × 10 5 km 2 ) and its highly suitable areas were predicted primarily in Aba prefecture of Sichuan Province (Figure 2a; Table S7). Compared to the CSAs for A. recurvata, the current CSAs  Table S7).

| Projected future change in species distributions
Generally, our projections for 2080 based on the full-migration scenario indicated that all species differ in their CSAs changes among the RCP 4.5 and RCP 8.5, with some species dramatically expanding or contracting their CSAs. Under both RCP 4.5 and RCP 8.5 scenarios, by 2080, the CSAs of A. recurvata were predicted to expand in southern Gansu, western Sichuan and eastern Tibet and contract in the southern and eastern parts of its current range (Figure 4a). For A. recurvata, a sizeable net expansion in CSAs from the current period to 2080 was projected to occur in both emission scenarios, with a 50.1% expansion under the RCP 4.5 scenario and a 38.8% expansion under the RCP 8.5 scenario (Figure 5a; Table S8). In contrast, the projected CSAs of A. faxoniana underwent a less pronounced contraction, with a decline of 4.3% under the RCP 4.5 scenario and 3.7% under the RCP 8.5 scenario (Figures 4b and 5a; Table S8). The enormous inconsistencies in species' CSAs changes among the two emission scenarios  Table   S8). Simultaneously, we found that by 2080, currently highly suitable areas of most species will become less suitable regardless of the assigned scenario, especially for the "South" group of species (Figure 4; Table S7).   (Figure 5c; Table S8). In comparison to the no-migration scenario, the most significant change simulated in the partial-migration scenario occurred in A. recurvata, with an average 46.6% of newly colonized CSAs under RCP 4.5 and RCP 8.5 (Figure 5b; Table S8). At the same time, under the RCP 4.5 scenario, A. georgei expected to undergo an overall decrease (13.2%)

| Migration scenario analysis
in the partial-migration model rather than an increase (22.2%) in the full-migration model because approximately more than 80% of the colonizable CSAs were not accessible (Figure 5b; Table S8).

| Direction of future range shift
The vectors between the present and the future centroids indicated that the magnitudes and directions of the range shifts of all the species differed under both RCPs as well as the full-and partial-migration scenarios ( Figure 6; Table S9; Figure S4 for full-migration result in the supplemental files). In the partial-migration scenario, all species differed in the directions of their centroids shifts from the current period to 2080 ( Figure 6) (Figure 6).

| Key factors shaping species distributions
The importance of temperature-related variables and solar radiation on the current distributions of all fir species exceeded those of water-related variables ( Table 2). The unique geography and weather conditions of the MSWC explained this discrepancy.
The species we studied mainly inhabit the high mountain areas, where the annual precipitation is approximately > 900 mm due to the East Asian and southwestern monsoons, thus making it a humid region that enables these species to not experience drought stress (Farjon, 2001;Ye & Gao, 1979). The importance of the climatic key drivers, however, appears to be strongly correlated with the geographical location of each species (Figure 3).
Generally, isothermality is regarded as the most decisive factor for firs in the MSWC (Chhetri et al., 2018;Liu et al., 2018).
However, accounting for additional factors, apparent differences exist in the importance of solar radiation and temperature seasonality between our "North" and "South" species ( Figure 3). Note: The variables highlighted in bold are the three top-ranked importance factors for each species. BIO2 = mean diurnal range (°C), BIO3 = isothermality (°C), BIO4 = temperature seasonality (°C), BIO5 = max temperature of warmest month (°C), BIO11 = mean temperature of coldest quarter (°C), BIO14 = precipitation of driest month (mm), BIO15 = precipitation seasonality (coefficient of variation) (%), SR = solar radiation (kJ m -2 day -1 ) and WS = wind speed (m/s). and distribution of ecosystems through photosynthesis (Piedallu & Gégout, 2008). We emphasized the importance of solar radiation on the distribution of the "North" group of species which is consistent with findings of Zhang et al. (2016) and underline the necessity of incorporating solar radiation in species distribution modelling.

| Distribution dynamics under the fullmigration scenario
In our first hypothesis, we stated that some fir species may benefit from future ACC and may have a broader potential distribution under the condition of unlimited migration. According to our fullmigration assumption, this hypothesis was indeed confirmed; future climate change was suggested to be detrimental to the persistence of the current CSAs for most firs in the MSWC, but A. recurvata was expected to expand its CSAs by an average of 44.5% under both emissions scenarios (Figures 4a and 5a).
Compared with other full-migration projections, our optimistic forecast for A. recurvata was not consistent with the overall declining   trend of the genus Abies as highlighted by Liu et al. (2018) andShao et al. (2017). Benefiting from future climate change, a large number of currently unoccupied areas in southern and western Sichuan will be potentially suitable for A. recurvata (Figure 4a). On the other hand, Liu et al. (2018) found at the genus level that the reduction in distribution area under the RCP 8.5 scenario was significantly higher than that under the RCP4.5 scenario. Although our projections for most species were consistent with this general finding, for A. squamata, the net loss of CSAs under the RCP 4.5 scenario was predicted to be twice as large as that under the RCP 8.5 scenario (Figure 5a).
Obviously, the generalizations of these previous studies masked individual changes that were contrary to the overall trend, and our study identified such specific responses of fir species to climate change.
Similarly, on the QTP, Chhetri et al. (2018) also predicted that the suitable area of Abies spectabilis would increase significantly in the future. At the same time, our full-migration scenario for A. georgei was more in line with that of Kou, Li, and Liu. (2011), and we both emphasized a less optimistic estimate of the future existence of this species under the more severe climate change scenario. However, A. faxoniana was the species least affected by ACC in our prediction; this projection was different from that of Kou et al. (2011), who predicted that its suitable areas would expand dramatically at the end of this century, even in southwestern Tibet. The prediction of Kou et al. (2011) was based on a fuzzy distribution data set and only three environmental variables. In comparison, the more accurate distribution data and comprehensive predictors used by our SDMs made our prediction of A. faxoniana (Figure 2b) closer to the actual distribution described by Fan (2006). Nevertheless, unlimited, long-distance migration seems unrealistic for mountain plants (Engler et al., 2009) except for the most vagile invasive species (Hellmann, Byers, Bierwagen, & Dukes, 2008).

| Migration constraints
Full-migration scenarios predicted potentially colonizable areas, which are inaccessible in the no-migration scenarios (Franklin, 2010). Similarly, under partial-migration scenarios, all the species studied were expected to decrease their suitable areas to a certain degree ( Figure 5b). However, the partial-migration scenarios split all potentially colonizable areas into accessible and inaccessible areas.
According to our results, the potentially colonized areas, especially for species of the "South" group, are significantly reduced in the partial-migration scenarios in comparison with the full-migration scenarios (Figure 5a,b). Obviously, this result reinforced our second hypothesis, and the most significant change occurred in A. georgei under the RCP 4.5 climate scenario, with more than 80% of its northern colonizable area is inaccessible in the partial-migration scenario ( Figure 5b, Table S8). Our projections for A. recurvata suggested that its new future CSAs were mostly in the neighbourhood of its current CSAs and could be colonized relatively easily (Figure 4a). However, A re c, R C P 4 .5 A re c, R C P 8 .5 A fa x, R C P 4 .5 A fa x, R C P 8 .5 A g eo , R C P 4 .5 A sq u, R C P 4 .5 A sq u, R C P 8 .5 A er n, R C P 4 .5 A er n, R C P 8 .5 A fo r, R C P 4 .5 A fo r, R C P 8 .5 A g eo , R C P 8 .5 A re c, R C P 4 .5 A re c, R C P 8 .5 A fa x, R C P 4 .5 A fa x, R C P 8 .5 A g eo , R C P 4 .5 A sq u, R C P 4 .5 A sq u, R C P 8 .5 A er n, R C P 4 .5 A er n, R C P 8 .5 A fo r, R C P 4 .5 A fo r, R C P 8 .5 A g eo , R C P 8 .5 A re c, R C P 4 .5 A re c, R C P 8 .5 A fa x, R C P 4 .5 A fa x, R C P 8 .5 A g eo , R C P 4 .5 A sq u, R C P 4 .5 A sq u, R C P 8 .5 A er n, R C P 4 .5 A er n, R C P 8 .5 A fo r, R C P 4 .5 A fo r, R C P 8 .5 A g eo , R C P 8 .5 Expansion Contraction Unchanged

(a) (b) (c)
Another uncertainty concerns the migration abilities of species, which vary greatly among taxonomic groups (Bateman et al., 2013). McLachlan et al. (2005) found temperate broad-leaf species, such as the genus Acer and Fagus, migrate at approximately 172 m/yr and 214 m/yr, respectively, to track rapid ACC. Fitzpatrick et al. (2008) found that the simulated migration rate of c. 500 m/yr might be ideal for simulating the potential distribution of some broad-leaf species (Banksia) in Western Australia. In contrast, La Sorte and Jetz (2010) indicated a general statement that it was better to apply the no-migration scenario to future predictions of mountainous coniferous species. Based on a no-migration scenario, Dyderski et al.
(2018) emphasized the threatened areas in the projections of Abies alba, rather than its large potential CSAs in Northern Europe, as the authors thought A. alba has a low migration ability. Reports of precise migration rates are, however, rare (Meier et al., 2012); therefore, it is difficult to determine which migration scenario is suitable for predicting the studied fir species. Interestingly, our study found a clear contrast in range shifts, especially in the "South" group of species when migration constraints were considered. Although the partial-migration rate utilized in this study may not accurately reflect the actual migration abilities of each fir species, the three migration scenarios together provided a more complete picture of the potential future changes in the distributions of the studied fir species.

| Direction of future range shifts
In this section, we discuss the third hypothesis that not all of the   (Xu, Tao, & Sun, 1973), which also reinforced our prediction. According to our projections, the studied firs will probably once again northward and westward migrate to the hinterland of QTP in response to future climate change.
Species responses to climate change vary with regions (Lindner et al., 2014). The direction in which species migrate with climate change depends on their specific physiological tolerance.
Generally, species tend to migrate to higher latitudes, but there might be exceptions (Camille & Hanley, 2015). Shafer, Bartlein, and Thompson. (2001) Figure 6). However, analysing these subtle differences can support some assisted migration planning to mitigate the effects of future ACC on forests (Hällfors et al., 2016). This holds especially for species, such as, A. forrestii and A. georgei in our study, which very likely face difficulties tracking changes in their CSAs (Figure 4e,f; Figure 6).

| Risk to species in climate-sensitive areas
The junction of southern Sichuan and northern Yunnan was identified as a climate-sensitive area for species of the "Middle" and "South" groups, an area where firs were projected to lose much of their current CSAs ( Figure S6). The alteration in temperature is a limiting factor for the growth of firs ( Figures S1-S3). Besides, this region has a long history of intense human activities, especially near the city of Panzhihua, and the high level of industrial resource development makes its forests more vulnerable to human disturbance ( Figure S7). More importantly, biogeographic barriers such as mountain ranges may limit the migration of these species (Wason, Bevilacqua, & Dovciak, 2017). Unfortunately, as these fir species of "South" group have reached almost the maximum height of the mountains, there may be limited opportunity to escape global warming through changes in elevation (Table S1, Fan, 2006). Hence, long-distance migration is the only option.
We speculate that the barriers of mountains and rivers in the "three parallel river-running areas" could prevent a large number of seeds from reaching suitable habitats in time as the climate changes, resulting in an expansion rate failing to catch up with habitat loss.

| CON CLUS I ON S AND RECOMMENDATI ON S FOR FUTURE RE S E ARCH
Our climate niche models identified solar radiation as a key factor shaping the distribution of the "North" group of species, while temperature seasonality was identified as a key factor affecting the "South" group of species. We found that species in the southern part of the MSWC seemed to be more threatened by climate change, and this threat was amplified after limited migration was added as well as under the severe climate change scenario.
Consequently, these species may face a higher risk of severe habitat loss in the future; hence, conservation assessment and planning are urgent priorities for these species that are endemic, dominant and considered keystone species of the fir forests in Southwest China. Although the projections of species distributions were developed under unavoidable simplified assumptions and uncertainties, they indicate potential challenges for Abies conservation and underscore the importance of incorporating migration ability into climate change effect assessments.

ACK N OWLED G EM ENTS
We would appreciate the help of Ms. Zhang Fengying, Mr.

CO N FLI C T O F I NTE R E S T
None of the authors has any conflict of interest to declare.

DATA AVA I L A B I L I T Y S TAT E M E N T
The used online database of all species distribution data and en-