Based on the Chinese and Latin literary names of E. sibiricus, the distribution points in China were comprehensively compiled from the existing literature, field investigation, and public online database. Four online databases (GBIF, Ecoengine, iNaturalist, and iDigBio) related to plant distribution information were queried by the R package spocc v1.2.2 (https://cran.R-project.org/package=spocc), and the search ended on 1 June 2024. The species name and its classification were queried and corrected through the NCBI Taxonomy database using the R package taxize v0.9.1 (https://CRAN.R-project.org/package=taxizedb). The distribution data were checked for the common spatial and temporal errors using the R package CoordinateCleaner v3.0.1 (Zizka et al., 2019). To reduce spatial autocorrelation and sampling bias, the Spatially Rarefy Occurrence Data for SDMs tool in SDM Toolbox v2.4 (Brown et al., 2017) was used to rarefy the species distribution data to ensure that only one distribution point was contained in each environmental unit at 5 arc-min resolution (~10 km).

To effectively explore the potential spatiotemporal distribution pattern of E. sibiricus, this study obtained environmental variables ( Table 1 ) that widely affect the plant growth and distribution, including bioclimatic, climatic, soil, and topographic data. Bioclimatic and climatic data were sourced from Paleoclim website (http://www.paleoclim.org). Current bioclimatic and climatic data (CP, 1979–2013) were obtained using the Anthropocene v1.2b dataset, while future bioclimatic and climatic data (FP, 2071–2100) were obtained using the CMIP6 dataset under the MPI-ESM1-2-HR climate model (Jägermeyr et al., 2021) and the ssp375 economic sharing pathway (Riahi et al., 2017). These bioclimatic and climatic variables mainly represent long-term annual trends, seasonality, and extreme situations of temperature and precipitation. The biomass of grass roots was mainly distributed in 0–30 cm soil layer to absorb soil nutrients, and the fibrous root system in 0–10 cm soil layer was extremely important for growth and development (Li et al., 2011). Soil layer data (i.e., 0–0.045 m, 0.045–0.091 m) containing nutrients necessary for plant growth were obtained from the World Soil Database (http://globalchange.bnu.edu.cn/research/soilw) (Shangguan et al., 2014), and the average value of different soil layer data was calculated by ArcGIS v10.8 (https://desktop.arcgis.com/). Elevation indirectly affects the adaptability of species to the environment by affecting temperature, and the SRTM elevation data was selected from the WorldClim website (https://www.worldclim.org). All environmental variables were unified into GCS_WGS_1984, resampled to 5 arc-min resolution and masked to China region by ArcGIS software. The China region was created from the map with review number GS (2019) 1822, and the base map was not modified. In order to avoid the multicollinearity of environmental variables leading to overfitting of the prediction model, the Correlations and Summary Stats tool in SDM Toolbox was used to detect the correlation among environmental variables, retaining only environmental variables with |r| <0.80 and significant ecological significance. The reserved environment variables were normalized through Circuitscape tool v1.0.2 (http://www.circuitscape.org) as a reference with elevation data.

To ensure the consistency of environmental data and the effectiveness of model predictions, this study constructed three types of environmental datasets: the full environmental dataset under the current period (CP-full dataset), including bioclimatic, climatic, soil, and topographic data; the simplified environmental dataset under the current or future period (CP or FP-simp dataset), including bioclimatic, soil, and topographic data. The CP-full dataset is used to finely predict the potential spatial distribution pattern of E. sibiricus in the current period; the CP or FP-simp dataset is used to analyze the temporal pattern dynamic changes of the potential distribution of E. sibiricus from the current to future periods. The potential spatial distribution pattern of E. sibiricus was modeled and predicted by R package ssdm v0.2.9 (Schmitt et al., 2017), which integrated different environmental datasets and species distribution points. To obtain more robust prediction results, this study selected nine modeling algorithms, including MAXENT, GLM, generalized additive model (GAM), multiple adaptive regression splines (MARS), generalized boosting model (GBM), classification tree analysis (CTA), random forest (RF), and ANN and support vector machine (SVM).

During the modeling process, the pseudo-occurrence points of the species of E. sibiricus were automatically generated, and 75% distribution data were randomly divided into training datasets for model construction, and the remaining were used as testing datasets to evaluate model performance. The independence between training and testing datasets was verified by 10 repeated crossover methods based on holdout strategy. The area under the curve of receiver operating characteristics (AUC) threshold is 0.75 as the standard for ensemble modeling algorithms, and Pearson correlation tests were performed among algorithms. To coordinate the inherent advantages and disadvantages of each algorithm, an integrated scoring system of true skills statistics (TSS), Kappa, and AUC was used to evaluate the predictive performance of the ensemble model (Chau et al., 2023). The values of TSS, Kappa, and AUC ranged from 0 to 1, 0.85 ≤ TSS | Kappa ≤ 1.0 or 0.9 ≤ AUC ≤ 1.0 considered that the model was accurate; 0.70 ≤ TSS | Kappa < 0.85 or 0.8 ≤ AUC < 0.9, the model predicts well; 0.55 ≤ TSS | Kappa < 0.70 or 0.6 ≤ AUC < 0.7, the model predicts general; 0.40 ≤ TSS | Kappa < 0.55 or 0.5 ≤ AUC < 0.6, the model predicts poorly; if TSS | Kappa < 0.40 or AUC < 0.50, the model predicts abnormal. The relative importance of environmental variables was estimated by R package ssdm. Finally, the R package ssdm generates continuity distribution probability with a threshold accuracy of 100.

Based on the natural break classification method, the distribution probability was divided into three levels of suitability (high, low, and unsuitable regions) by using the Reclassify tool in ArcGIS software, and the area of different suitable levels in each administrative region was counted by using the Region analysis tool in ArcGIS software. To obtain the most suitable survival environment conditions for E. sibiricus in the current period, the species survival curve was calculated by MaxEnt v3.4.4 (Phillips et al., 2006). In the parameter settings of MaxEnt software, enable the options of Create response curves, Do jackknife, Random seed, and Write plot data; set Random test percentage to 25%, Regularization coefficient to 1.1, and Max number of background points to 10,000. The operation model was repeated 10 times based on the Bootstrap strategy and defaulted to other parameters.

The change and migration of species to suitable regions were a response to environmental changes over a long period of time. To explore the temporal distribution pattern of E. sibiricus from the current (1979–2013) to future (2071–2100) period, Wilcoxon test was conducted on the area of suitable levels at different periods using SPSS v25.0 (https://www.ibm.com/spss). The geometric centroid of suitable regions is used as a general description of the distribution characteristics of species in each period, and the migration of the geometric centroid can more intuitively display the changes of suitable regions of species in different periods. To clarify the dynamic migration path of the suitable region of E. sibiricus, the distribution probability was used, the geometric centroid positions of E. sibiricus distribution in different periods were located by the region analysis tool in ArcGIS software and R package foreign v0.8 (https://cran.r-project.org/package=foreign), and the geographical distance between different centroids was calculated by R package geosphere v1.5 (https://cran.r-project.org/package=geosphere).

The potential planting pattern of E. sibiricus in the current period was predicted by Marxan v4.0.5 (Game et al., 2013), which involves the setting of planning unit, target, cost, and parameter optimization. Considering the continuity of the planting region and the simulation ability of Marxan software, a square planning unit (grid) with a side length of about 5 km covering the land of China was created by using the Fishnet tool in ArcGIS software. According to the protection standard of the Guidelines for Management Planning of Protected Areas issued by the International Union for Conservation of Nature (IUCN, https://www.iucn.org/), 10% planting coverage rate was selected as the planting target, which does not affect the ecological environment and maintains the basic ecological functions of species. Due to Chinese policy of “arable land minimum” (Chen et al., 2019), this study only selects the probability of refined distribution in the current period and the consensus rate of herbaceous vegetation coverage as equally weighted planting cost. The consensus rate of herbaceous vegetation coverage was selected from Herbaceous Vegetation Reduced v1.0 dataset (https://www.earthenv.org/landcover). The parameters such as boundary length (BLM), species penalty factor (SPF), and iteration times were optimized by ArcGIS plug-in ArcMarxan Toolbox v2.0.2 (https://aproposinfosystems.com/). Based on the natural break classification method, the potential planting regions were divided into three levels of planting (high, low, and unselected regions) by using the Reclassify tool in ArcGIS software, and the area of different planting levels in each administrative region was counted by using the region analysis tool in ArcGIS software.

In this study, two cultivars [E. sibiricus cv. “Tongde” (S1) and “Chuancao No. 2” (S2)] and six germplasms [LM01 (S3), LM02 (S4), LM03 (S5), LM04 (S6), LM05 (S7), and LM06 (S8)] of E. sibiricus were selected to carry out field experiments in the forage breeding base of the Grassland Science Research Institute, Hongyuan County, Sichuan Province (32°47’N, 102°32’E, 3460 m) for 3 years. The experiment adopted a randomized block design with three replicates, with a sample plot area of 15 m² (3 m × 5 m) and a spacing of 0.5 m between plots. Each sample plot was evenly seeded with 33 g of seeds and the row spacing of 0.3 m. To effectively evaluate the growth performance of E. sibiricus, 23 agronomic traits ( Table 2 ) for 8 tested materials were measured (n = 10) from 2021 to 2023 according to “Description Standard and Data Standard of Elymus sibiricus Germplasm Resources” (Li et al., 2023). The values of all agronomic traits were normalized by R package vegan v2.6 (https://CRAN.R-project.org/package=vegan) to unify dimensions. Principal components analysis (PCA) and phenotypic contribution analysis of all tested materials were carried out using R package FactoMineR v2.11 (https://CRAN.R-project.org/package=FactoMineR). The membership function analysis among tested materials was analyzed by SPSSPRO software (https://www.spsspro.com/) to evaluate the comprehensive properties of different materials. Based on the potential planting pattern of E. sibiricus, the top 25 municipal administrative regions were selected according to the decrease of potential planting area with high selected regions. The suitability probability was extracted by using the extraction analysis tool in ArcGIS software under constraints of different municipal administrative regions. The distribution probability was used as the geographical environment standard to measure the suitable distribution of E. sibiricus, and the agronomic traits were weighted by the phenotypic contribution degree to measure the adaptability level of different tested materials. The optimal dissimilarity index was screened by R package vegan, and a Mantel test between the top 25 municipal administrative regions and weighted agronomic trait values of six germplasms was carried out by ChiPlot online software (https://www.chiplot.online/) to predict the introduction adaptability of different germplasms.

A total of 538 distribution data of E. sibiricus were compiled in this study. After data quality control and rarefaction, there were 421 effective distribution points ( Figure 1A ), which were widely distributed in Qilian Mountains (northeast of QTP), Taihang Mountains (junction of Loess Plateau and Inner Mongolia Plateau), and Tianshan Mountains. The correlation of environmental variables showed that the correlation ranges of 46 environmental factors in the CP-full dataset were −0.98 to 0.99 ( Figure 1B ), and the correlation ranges of 25 environmental factors in the CP and FP-simp dataset were −0.98 to 0.99 and −0. 99 to 0.99 ( Supplementary Figure S1A ), respectively. Since 17 environmental factors in bioclimatic variables are derived from mean annual air temperature (bio01) and annual precipitation amount (bio12), bioclimatic variables are screened based on bio01 and bio12. After correlation analysis of all environmental factors, 24 environmental factors with low correlation and obvious ecological significance were screened out from the CP-full dataset, including Elev, AK, EAl, AP, ECa, EH, EMg, ENa, PH, TK, TN, TP, Bio01, Bio12, Bio15, Bio02, Bio03, Bio07, CMI_range, GSP, HURS_mean, HURS_range, PET_mean, and PET_range. The same eight environmental factors (Elev, Bio01, Bio12, Bio15, Bio02, Bio03, Bio07, and GSP) were screened from the CP or FP-simp dataset. The correlation analysis of nine modeling algorithms showed that the correlation range based on the CP-full dataset was 0.33 to 0.88 ( Figure 1C ); the correlation ranges based on the CP and FP-simp datasets were 0.54 to 0.93 ( Supplementary Figure S1B ) and 0.60 to 0.93 ( Supplementary Figure S1C ), respectively, indicating that there are significant deviations and uncertainties among different algorithms. The evaluation results of the integrated scoring system for the model prediction performance showed that RF algorithm had the best modeling effect on the spatial distribution of E. sibiricus under different environmental datasets, while MAXENT algorithm was the worst. It is worth noting that GLM algorithm cannot effectively model on CP or FP-simp dataset. Overall, the AUC, Kappa, and TSS based on the CP-full dataset were 0.82, 0.70, and 0.75, respectively (Figure 1D); the AUC, Kappa, and TSS based on the CP and FP-simp dataset were 0.89, 0.67, 0.72 and 0.89, 0.68, 0.72, respectively. The evaluation results of the integrated scoring system indicated that the ensemble model constructed by different algorithms can effectively simulate the potential spatial distribution pattern of E. sibiricus.

Relative contribution can be used to evaluate the influence of environmental variables upon the spatial distribution of E. sibiricus. 24 environmental factors from the CP-full dataset were used to refine and predict the spatial distribution pattern of E. sibiricus in the current period. The relative contribution analysis showed that the relative contribution ranges of 24 environmental factors ranged from 1.57% to 11.36% ( Figure 2A ), among which the total relative contribution of bioclimatic, climatic, soil, and topographic variables was 23.62%, 39.23%, 31.93%, and 5.21%, respectively. Specifically, the cumulative relative contribution of 11 environmental factors reached 69.03%, including the annual range of monthly near-surface relative humidity (HURS_range, 11.36%), mean annual air temperature (Bio01, 9.67%), annual range of monthly potential evapotranspiration (PET_range, 8.52%), mean monthly potential evapotranspiration (PET_mean, 7.53%), annual range of monthly climate moisture index (CMI_range, 6.48%), elevation (Elev, 5.21%), exchangeable Ca²⁺ (ECa, 4.94%), available P (AP, 4.27%), mean monthly near-surface relative humidity (HURS_mean, 3.78%), exchangeable H⁺ (EH, 3.71%), and annual precipitation amount (Bio12, 3.57%). Overall, the spatial distribution pattern of E. sibiricus is obviously responsive to hydrothermal environmental conditions, especially humidity, air temperature, and potential evapotranspiration, and the elevation and soil calcium and phosphorus nutrients also play an important role in its distribution. In addition, based on the survival response curve, the optimum survival environment value under the optimum survival probability of E. sibiricus was extracted ( Table 3 ). The ranges of optimum survival probability and its coefficient of variation under different environmental factors were 0.08–0.93 and 1.85%–125.47%, respectively, among which the optimum survival probability and coefficient of variation with GSP were extremely abnormal, indicating that GSP had no significant effect on the potential spatial distribution of E. sibiricus. According to the comprehensive analysis of optimum survival environment values for different types, it is concluded that the most suitable habitat of E. sibiricus is a semi-humid area with weak acidic soil rich in nitrogen, phosphorus, and potassium, small fluctuation of air temperature, low temperature, and high altitude.

Based on the natural break classification method, the potential spatial distribution of E. sibiricus was divided into three levels of suitability ( Supplementary Table S1 ). The results showed that the potential spatial distribution mainly spans 33°–42°N and 95°–118°E ( Figure 2B ), which was concentrated in Qilian Mountains (northeastern of QTP), Taihang Mountains (the junction of Loess Plateau and Inner Mongolia Plateau), and Tianshan Mountains, while the rest was scattered in Altai Mountains, Kunlun Mountains (northern of QTP), Hengduan Mountains (central Yunnan-Guizhou Plateau), Qinling Mountains (south of Loess Plateau), and Changbai Mountains (southeast of Northeast China). This study predicted an area of 9.3784 million km² in China, with high and low suitable regions accounting for 14.67% (1.3758 million km²) and 29.44% (2.7613 million km²), respectively. High suitable regions were mainly concentrated in western China, while low suitable regions were mainly distributed around high suitable regions. The high and low suitable regions involve 25 and 33 provincial administrative regions, respectively ( Supplementary Table S2 ), among which the top five provincial administrative regions with the suitable area decreases in high suitable regions are Xinjiang (296,500 km²), Gansu (193,300 km²), Qinghai (125,700 km²), Sichuan (112,600 km²), and Heilongjiang (103,900 km²); the top 5 provincial administrative regions with the largest area of low suitable regions are Inner Mongolia (399,100 km²), Xinjiang (372,600 km²), Heilongjiang (270,300 km²), Qinghai (197,800 km²), and Sichuan (185,300 km²). Overall, the suitable area for E. sibiricus distribution accounts for about 44.11% of the total area in China, and it is concentrated in plateau regions.

The dynamic changes of the potential spatial distribution pattern of E. sibiricus across multiple periods were analyzed. The results showed that the levels of high, low, and unsuitable regions remained relatively stable across different periods ( Figure 3A ), but the area changes within each suitable level were significant during the same period ( Figure 3B ), indicating that the potential suitable pattern of E. sibiricus had migrated with the passage of time. Comparing the current period with the future period, the overall suitable area is estimated to grow between −2.72% and 4.65 million km². Specifically, the area of high and low suitable regions is expected to increase by −10.66% (up to approximately 1.7684 million km²) and 2.89% (up to around 2.8817 million km²), respectively ( Figure 3C ). A deeper analysis of the dynamic changes within the different suitable levels revealed that over time, the percentage range of area growth for high and low suitable regions varied from −100.00% to 59.39% and from −86.56% to 1185.90%, respectively ( Figure 3C , Supplementary Table S3 ). The regions predicted to experience relatively large expansions in their suitable areas are Guizhou, Yunnan, and Henan (high suitability), as well as Taiwan, Guangxi, Beijing, and Fujian (low suitability). Meanwhile, the areas expected to have relatively large contractions in their suitable regions are Shandong, Tianjin, and Chongqing (high suitability) and Zhejiang, Jiangsu, and Anhui (low suitability). Overall, compared to the present, the forecasted suitable area for E. sibiricus demonstrates a general decline; however, the key distribution centers remain relatively stable, such as Qilian Mountains (northeast of QTP), Taihang Mountains (junction of Loess Plateau and Inner Mongolia Plateau), and Tianshan Mountains. Furthermore, the centroid of the suitable habitat is estimated to shift southwestward by approximately 2.93 km ( Figure 3C ).

Based on the natural break classification method, the planting priority of E. sibiricus was divided into three levels of planting ( Supplementary Table S4 ). This study only predicted the natural grassland area of 0.349 million km² in China. The potential planting regions of E. sibiricus were mainly distributed in the Qilian Mountains (northeast of QTP), Taihang Mountains (southeast of Inner Mongolia Plateau), Tianshan Mountains, and Altai Mountains ( Figure 4 ). The high and low selection planting areas were 66,600 km² and 139,300 km², respectively, among which the high and low selection planting regions involve 21 and 20 provincial administrative regions, respectively ( Supplementary Table S5 ). The top 6 provincial administrative regions with the planting area decreases in high selection regions are Xinjiang (17,752.50 km²), Gansu (12,577.50 km²), Qinghai (7,582.50 km²), Tibet (7,335.00 km²), Inner Mongolia (7,155.00 km²), and Sichuan (4,275.00 km²). The top six provincial administrative regions with the largest area of low selection regions are Inner Mongolia (36,090.00 km²), Tibet (28,777.50 km²), Qinghai (20,970.00 km²), Xinjiang (20,205.00 km²), Sichuan (10,755.00 km²), and Gansu (6,007.50 km²). Overall, the suitable area for planting E. sibiricus accounts for 58.99% of the predicted natural grassland and 2.14% of the national area in China.

The PCA and contribution analysis of 23 agronomic traits showed that all agronomic traits explained 58.28% of the overall traits ( Figure 5A ), among which 12 agronomic traits, such as DFR, TGW, FN, SN, NPH, APN, DSLR, FTLR, HY, FGY, FLL, and FLW, contributed significantly to the comprehensive growth evaluation of E. sibiricus. The membership function analysis showed that, in comprehensive evaluation, the varieties S3 (0.47), S5 (0.47), S6 (0.53), S7 (0.60), and S8 (0.49) were superior to E. sibiricus varieties S1 (0.11) and S2 (0.45), while S4 (0.23) was superior to S1 but inferior to S2 ( Figure 5B ). The Mantel Test analysis of the introduction adaptability showed that the Pearson correlation range among all sample plots was −0.58 to 1.00 ( Figure 5C ); for example, the correlation among SP07, SP08, and SP09 was greater than 0.65, indicating that the plots had similar environmental conditions for adaptability distribution. Varieties S3, S5, S6, S7, and S8 had a significant positive correlation with SP03; S6, S7 with SP04; S3, S4, S6, S7, and S8 with SP11; and S5 with SP23 (p < 0.05). It was indicated that S3, S5, S6, S7, and S8 were suitable for planting in the southeast of the Altai Mountains represented by the Altay region (SP03); S6 and S7 were suitable for planting in the western section of the Hengduan Mountains represented by the Changdu City (SP04); S3, S4, S6, S7, and S8 were suitable for planting in the middle part of the Qilian Mountains represented by the Hainan Tibetan Autonomous Prefecture (SP11); and S5 was suitable for planting in the northern foot of the eastern part of the Qilian Mountains represented by the Wuwei City (SP23). Overall, the introduction adaptation regions of the six germplasms are all distributed in the high-altitude mountain region in western China, which indirectly verifies the rationality of the potential distribution and planting distribution pattern of E. sibiricus.

In predicting the potential distribution probability of species, the single ENMs tends to overestimate in places with low species richness and underestimate in places with high species richness (Calabrese et al., 2014). The uncertainty of distribution prediction may distort policy making and planning. In the process of species distribution modeling, previous studies generally rely on multiple evaluation indicators to test the model results constructed by different algorithms based on the same data, selecting the algorithm with the highest statistical accuracy (Li and Wang, 2013; Xu et al., 2015). However, there may be inconsistencies among evaluation indicators. For example, in this study, the model with the highest AUC index was inconsistent with the model corresponding to the highest values of other indexes (Kappa and TSS). In addition, the construction of potential spatial distribution pattern of species is based on three prerequisites: (1) dynamic equilibrium of species distribution and environmental demand (Buckley et al., 2011), (2) idealized species migration ability (Xu et al., 2021), and (3) relatively conservative species ecological niche (Harris, 2015). The simulation results of the potential distribution of E. sibiricus in this study fails to effectively cover the existing distribution records, such as Wuyi Mountain, Luoxiao Mountain and Nanling Region, which show that the optimal algorithm may not achieve statistical optimum under all modeling conditions, nor may it effectively reduce the common sampling deviation (Randin et al., 2006). Nevertheless, to improve the accuracy of the model, using ensemble models to predict species distribution has become the trend of research (Grenouillet et al., 2011; Hao et al., 2020). The ensemble model maps the main trends of species distribution with the overall changes of all models and integrates the advantages of different algorithm models, such as the importance of variables or model response curve (Buisson et al., 2010), so as to effectively achieve and explore the prediction range of different algorithms and find consensus in the prediction of different mechanisms.

In this study, the key environmental factors affecting the spatial distribution of E. sibiricus in the current period are all related to hydrothermal environmental conditions, especially humidity, air temperature, and potential evapotranspiration, which is consistent with the viewpoint that abiotic factors, mainly climate factors, dominated the species distribution pattern on a large scale (Soberon and Peterson, 2005). The response of plants to different environmental conditions gives them significant morphological plasticity (Sultan, 2000). Temperature and precipitation are the main climatic factors affecting forage growth (Li et al., 2018) and also the most important factors affecting genetic differentiation of the Elymus population (Yan et al., 2006). Studies have indicated that Bio01 and Bio12 have significant effects on the phenotype and genetic diversity of Elymus (Qi et al., 2013; Han et al., 2021); in this study, Bio01 and Bio12 were the main bioclimatic factors that significantly affect the potential spatial distribution of E. sibiricus. Compared to precipitation, temperature has a greater influence. Soil conditions can affect the growth and yield of aboveground and underground parts of plants (Liu et al., 2021). The appropriate dose of Ca²⁺ can activate the activity of antioxidant enzymes, alleviate the oxidative damage caused by low temperature to the cell membrane system, and thus improve the cold resistance of E. nutans (Qimei et al., 2021). Application of phosphate fertilizer can not only effectively increase the yield of artificial grassland build by E. nutans, but also prevent the loss of soil organic carbon (Renzneg et al., 2020). In this study, ECa and AP, as important soil factors, significantly affected the spatial distribution pattern. As an alpine plant, E. sibiricus obviously used ECa to improve its cold resistance and AP to increase seed yield to expand population size. Altitude is a comprehensive environmental factor, and the genetic diversity of the Elymus population increases first and then decreases with altitude, with the maximum value appearing at around 3,300 m (Yan et al., 2009). The most suitable altitude for E. sibiricus in this study is 3,355.44 m, which provides evidence for the hypothesis that alpine plants have higher genetic diversity at the most suitable altitude. In addition, environmental factors such as precipitation, temperature, and radiation intensity will change with different altitudes (Lv et al., 2023), while the relative contribution of elevation to the potential distribution of E. sibiricus in this study is only 5.21%, which indicates that elevation may not have a direct impact on the quality of life of species, but may indirectly affect the habitat suitability by affecting other environmental factors. These findings well prove that the key environmental variables screened in this study have practical significance.

The Qinghai-Tibet Plateau, located in southwest China, is the highest and largest plateau in the world, with an average elevation of over 4,000 m. During the uplift between Miocene periods and Quaternary periods, the geographical and climatic characteristics of QTP have undergone significant changes, making it recognized as one of the most important biodiversity hotspots in the world (Mao et al., 2021). QTP once served as a refuge for Northern temperate plants during the Quaternary period and was recognized as one of the important origin and radiation places of modern Northern temperate plants after the Glacial period (Xiong et al., 2022). The QTP is considered a potential center of origin for E. sibiricus population (Xiong et al., 2022, 2024). During the warming process from the last glacial period to Holocene, E. sibiricus gradually spread to Sichuan, Gansu, and Xinjiang and then further expanded to Qinghai and Inner Mongolia (Han et al., 2022; Xiong et al., 2022, 2024), which has formed a wide distribution pattern today. The pattern formed after the expansion of E. sibiricus population is similar to the potential distribution pattern predicted by this study, which is distributed in the vicinity of Qilian Mountains, Taihang Mountains, Kunlun Mountains, Hengduan Mountains, Qinling Mountains, and Changbai Mountains (Xiong et al., 2022, 2024). The potential planting and introduction adaptation areas of E. sibiricus predicted in this study yield similar results. In addition, geographical range is usually considered the basic determinant of species diversity and diversification (Ricklefs, 2006), and isolation (such as geographical, ecological, or behavioral isolation) is an important factor affecting the blocking mechanism of gene flow among populations, which has the possibility of leading to the formation of continent-island model, island model, and isolation-by-distance model (Qu et al., 2004). The potential distribution pattern predicted by this study may be caused by some historical geological events, thus forming the E. sibiricus population to gather and distribute in mountains and its vicinity. However, due to the lack of molecular data support, it is impossible to clarify specific historical geological events. Climate warming is a significant threat to global biodiversity, especially for endemic species in temperature-sensitive alpine areas (Verrall and Pickering, 2020). Temperature has an important influence on the productivity of alpine meadow plants, but the continuous rise of temperature may have a negative impact on alpine grasslands (Fazlioglu and Wan, 2021). Although some studies have suggested that many alpine plants in QTP-Hengduan Mountains may not face a high risk of extinction due to habitat contraction caused by climate warming, some species at extremely high altitudes show significant range contraction (Liang et al., 2018). In this study, with the climate warming and drought intensification in the future, the potential suitable regions of E. sibiricus show a collapse trend and migrate to the high-altitude areas in southwest China. The introduction adaptation regions of six E. sibiricus germplasm are also distributed in the high-altitude regions of mountains in west China. It is speculated that in response to global warming, the population of E. sibiricus will shrink and concentrate in the Qilian Mountains (northeast of QTP), Taihang Mountains (junction of the Loess Plateau and Inner Mongolia Plateau), and Tianshan Mountains. In addition, many studies have recorded the pattern of alpine species migrating to high altitude in response to climate warming (Dullinger et al., 2012; Körner and Hiltbrunner, 2021), which supports the “mountain-top” extinction and “uphill contraction” hypothesis (Sorte and Jetz, 2010), indicating that it is necessary to take protective measures for species inhabiting extremely high altitude or narrowly distributed.