Climate Change Impact on Three Important Species of Wild Fruit Forest Ecosystems: Assessing Habitat Loss and Climatic Niche Shift

: As global biodiversity hotspots, wild fruit forests play key ecological functions, providing essential ecosystem services such as carbon storage, soil retention, and water conservation, and support food chains and biodiversity conservation through key species. Climate change, with rising temperatures and altered precipitation patterns, threatens wild fruit forests by reducing the habitats and numbers of key species, potentially turning these ecosystems from carbon sinks to sources and diminishing overall biodiversity and ecosystem services. However, research on how these changes affect important species’ habitats and carbon dynamics remains insufficient. To address this, we analysed habitat suitability for three important species ( Prunus armeniaca L., Malus sieversii , and Prunus ledebouriana (Schltdl.) Y.Y.Yao with the aim of informing conservation strategies. We used biomod2 to integrate environmental and species data using six methods, encompassing 12 models. We predicted overlapping geographical distributions of three species, analysing their ecological niches and environmental interactions using global datasets to understand their adaptations. This analysis revealed ecological niche shifts and reductions in resource utilisation in both current and future scenarios. Their distribution centres will move poleward under the influence of bioclimatic factors and human activities. In conclusion, protecting P. armeniaca , M. sieversii , and P. ledebouriana is essential for the conservation and overall health of wild fruit forest ecosystems. This study provides new insights into climate change, habitat loss, informing conservation and resilience strategies.


Introduction
The Earth is currently undergoing its sixth mass extinction, primarily driven by habitat fragmentation due to degradation and loss, with climate change being a major factor [1][2][3].The Intergovernmental Panel on Climate Change (IPCC) reports a significant increase of 1.1 • C in average global surface temperature from 2011 to 2020 compared with preindustrial levels [4].This increase, which is linked to escalating greenhouse gas emissions from human activities, has exacerbated global warming.The combined effects of global warming and other stressors significantly impact terrestrial ecosystems, altering their structure and function and causing shifts in species habitats and distributions [5,6].Habitat fragmentation further disrupts interspecies interactions, increases vulnerability, and leads to widespread extinction and extirpation [7][8][9].
Vegetation is crucial as both an indicator and component of climate response, with species survival depending on genetic diversity, adaptability to environmental changes, and migration to suitable climates [10,11].Species in fragmented or rare habitats face higher risks of migration difficulties or extinction [12,13].Climate is a key determinant of species distribution, and changes in distribution patterns are clear indicators of climate shifts [14,15].The complex relationships among climate change, human disturbances, and biodiversity conservation present a multifaceted challenge [16,17].Forests cover a third of the Earth's land area and are vital for global carbon sequestration, holding 56% of terrestrial ecosystem carbon [18].Trees absorb significant carbon dioxide through photosynthesis and store it in biomass, litter, deadwood, and soil organic matter.Natural forests, unlike simpler artificial forests, are more effective in biodiversity conservation and provide key ecosystem services, such as carbon storage, soil retention, and water conservation [19].
The wild fruit forest region, identified as one of the key biodiversity hotspots worldwide, has garnered widespread academic attention owing to its abundant biological resources and unique ecosystem structure.These forests, which have survived geological changes and climatic variation since the Tertiary Period, serve as sanctuaries for thermophilic mesophytic broadleaf trees.Their existence is a result of their unique geological history and local climate conditions of warmth and humidity, in contrast to early desert climates [20,21].However, climate change, overgrazing, tourism, and unsustainable harvesting have severely affected these habitats, causing a drastic reduction in their area and number [22,23].Key species such as Prunus armeniaca L., Malus sieversii, and Prunus ledebouriana (Schltdl.)Y.Y.Yao, all Tertiary relicts, are the dominant species in these wild fruit forests [24][25][26][27][28]. M. sieversii, the ancestor of modern apples, is native to the Tianshan Mountains, including the regions within China, Kazakhstan, and Kyrgyzstan [20].P. armeniaca, from the Rosaceae family, has played a crucial role in the history of cultivated apricots and is prominent in Xinjiang's wild fruit forests [20,26].P. ledebouriana, also in the Rosaceae family, now exists only in relict distributions in Kazakhstan and Xinjiang, China, with notable fossilisation in Europe [29][30][31].The loss of habitats for these three keystone species will directly result in a reduction in biodiversity in the wild fruit forests, which, in turn, will have a negative impact on ecosystem functions and stability.It also affects the continuation of breeding and conservation programs.Understanding the potential overlapping geographical distributions of these species, along with their ecological niche dynamics and conservation area planning, is essential for their utilisation and protection against habitat loss due to climate change.
Our hypotheses focused on the direct impacts of climate change, specifically, the effects of rising temperatures and altered precipitation patterns on the habitats of three important species: P. armeniaca, M. sieversii, and P. ledebouriana.We predicted that increasing global temperatures and more frequent extreme climatic events will significantly reduce suitable Forests 2024, 15,1281 habitats for these species, compress their ecological niches, and decrease the area and biodiversity of wild fruit forests.To predict the future geographic distribution of these important species, we employed SDMs, integrating existing distribution data with future climate projections to estimate likely survival areas.We chose climatic factors-temperature, precipitation, and seasonal variation-and non-climatic factors like soil type and topography to define current ecological niches and project how future climate changes might alter species distributions.Our methodology included the following: (1) selecting high-precision individual models for each species in biomod2 after simulation testing; (2) EM application to forecast possible distribution zones and intersecting areas in both current and future climate conditions, specifically targeting the 2050s and 2090s periods; (3) analysis of ecological niche overlap and dynamics among these species, particularly between native and habitat loss areas; (4) identifying environmental variables that significantly impact their potential distributions; and (5) using the Marxan model to identify priority conservation areas and suitable growth environments.The study primarily aimed to provide a scientific basis for conserving wild populations and germplasm resources of these species.

Data on the Geographical Spread of Target Species
Our research gathered worldwide location records for three important species, P. armeniaca, M. sieversii, and P. ledebouriana.We analysed these data using a comprehensive checklist of SDMs [52].Initially, we searched online databases for the geographical distribution data of these species.Specifically, for P. armeniaca, M. sieversii, and P. ledebouriana, we obtained 110, 484, and 19 records from the Global Biodiversity Information Facility (http://www.gbif.org/,accessed 28 July 2023), and 11, 7, and 10 records from the Chinese Virtual Herbarium (http://www.cvh.ac.cn, accessed 28 July 2023), respectively.In addition, 21, 25, and 17 records were sourced from the literature [24][25][26][27][28].We excluded distribution records that were assigned to capital cities or based on the centroids of provinces.Furthermore, we included records from the Center for Agriculture and Bioscience International (https://www.cabi.org/,accessed 28 July 2023) and conducted country-based extrapolation.In enhancing the precision of environmental variables, only one distribution point was maintained per 10 × 10 km grid, employing ENMTools [53].This process led to the aggregation of 383 location records across the three species (275 for P. armeniaca, 79 for M. sieversii, and 36 for P. ledebouriana), as shown in Figure 1.
Our hypotheses focused on the direct impacts of climate change, specifically, the effects of rising temperatures and altered precipitation patterns on the habitats of three important species: P. armeniaca, M. sieversii, and P. ledebouriana.We predicted that increasing global temperatures and more frequent extreme climatic events will significantly reduce suitable habitats for these species, compress their ecological niches, and decrease the area and biodiversity of wild fruit forests.To predict the future geographic distribution of these important species, we employed SDMs, integrating existing distribution data with future climate projections to estimate likely survival areas.We chose climatic factors-temperature, precipitation, and seasonal variation-and non-climatic factors like soil type and topography to define current ecological niches and project how future climate changes might alter species distributions.Our methodology included the following: (1) selecting high-precision individual models for each species in biomod2 after simulation testing; (2) EM application to forecast possible distribution zones and intersecting areas in both current and future climate conditions, specifically targeting the 2050s and 2090s periods; (3) analysis of ecological niche overlap and dynamics among these species, particularly between native and habitat loss areas; (4) identifying environmental variables that significantly impact their potential distributions; and (5) using the Marxan model to identify priority conservation areas and suitable growth environments.The study primarily aimed to provide a scientific basis for conserving wild populations and germplasm resources of these species.

Data on the Geographical Spread of Target Species
Our research gathered worldwide location records for three important species, P. armeniaca, M. sieversii, and P. ledebouriana.We analysed these data using a comprehensive checklist of SDMs [52].Initially, we searched online databases for the geographical distribution data of these species.Specifically, for P. armeniaca, M. sieversii, and P. ledebouriana, we obtained 110, 484, and 19 records from the Global Biodiversity Information Facility (http://www.gbif.org/,accessed 28 July 2023), and 11, 7, and 10 records from the Chinese Virtual Herbarium (http://www.cvh.ac.cn, accessed 28 July 2023), respectively.In addition, 21, 25, and 17 records were sourced from the literature [24][25][26][27][28].We excluded distribution records that were assigned to capital cities or based on the centroids of provinces.Furthermore, we included records from the Center for Agriculture and Bioscience International (https://www.cabi.org/,accessed 28 July 2023) and conducted country-based extrapolation.In enhancing the precision of environmental variables, only one distribution point was maintained per 10 × 10 km grid, employing ENMTools [53].This process led to the aggregation of 383 location records across the three species (275 for P. armeniaca, 79 for M. sieversii, and 36 for P. ledebouriana), as shown in Figure 1.

Environmental Variables
This study incorporated 39 environmental factors, including 19 bioclimatic, 3 topographic, and 17 soil factors (Table 1).From the Paleoclim database (http://www.paleoclim.org, accessed 29 July 2023), we acquired raster files of 19 bioclimatic variables, each at a 5 arcmin resolution.These variables, originating from detailed monthly records of temperature and precipitation, have formed a globally utilised dataset for ecological studies, spanning from 1979 to 2013.For projections of future climates, specifically the periods 2041-2060 (the 2050s) and 2081-2100 (the 2090s), data were retrieved from the Worldclim2.1 database (http://www.worldclim.org/,accessed 29 July 2023).Each of these periods included 19 climate factors named bio1-bio19 [54].Future climate scenarios were selected from the sixth phase of the International Coupled Model Intercomparison plan (CMIP6).This approach differs from the typical concentration pathway (RCP) scenario used in CMIP5, which combines different shared socioeconomic paths (SSPs) with RCPs, and incorporates aspects of future socioeconomic development.We chose three combined scenarios, SSP1-2.6,SSP2-4.5, and SSP5-8.5, representing low, medium, and high radiation forcing scenarios, respectively.The spatial resolution of these data was 2.5 arcmin (~5 km).Soil-related parameters originated from the World Soil Database (HWSD), courtesy of the Science Data Center of Cold and Arid Regions.Topographic information was sourced from a geospatial data cloud (http://www.gscloud.cn/,accessed 29 July 2023).The prediction of suitable habitats was based on the premise that soil and topographic conditions will remain unchanged for the forthcoming 70 years [55].
The analysis of the 37 environmental factors (excluding soil and topographic factors) employed Pearson correlation and the variance inflation factor (VIF) to assess their correlation and significance.In the R language, we conducted Spearman's correlation and multicollinear VIF evaluations on the point interpolation data.Initially, we filtered out factors showing correlations of <0.8 and VIF values of <10.VIF, also known as the reciprocal of tolerance, indicates the degree of multicollinearity, where VIF < 10 indicates no multicollinearity, VIF > 10 < 100 indicates multicollinearity, and VIF > 100 indicates serious interfactor multicollinearity.The filtered modelling factors are shown in Table 1.

SDM Development and Accuracy Assessment
Our research involved modelling the present and anticipated future geographic distributions of three important species (P.armeniaca, M. sieversii, and P. ledebouriana) utilising global occurrence data alongside environmental variables.This process was facilitated through the biomod2 package within R Studio [56].In biomod2, we applied a suite of 12 distinct algorithms, specifically, ANN, CTA, FDA, GAM, GBM, GLM, MARS, MAX-ENT, MAXNET, RF, XGBOOST, and SRE.For each of these species, a random selection of 75% of their distribution records constituted the training dataset; the remaining 25% formed the test dataset [57].To increase the models' reliability, we repeated this distribution segmentation five times.
Additionally, we generated 1000 pseudo-absence points, repeating this step three times to refine our models.Consequently, 360 models were developed.Through accuracy assessment, we selected four models that consistently showed average values exceeding 0.9 for further analysis.These models included ANN, CTA, FDA, GAM, GBM, GLM, MARS, MAXENT, MAXNET, RF, and XGBOOST.
We constructed an EM using 11 individual models, employing six integration methods: EMmean, EMcv, EMci, EMmedian, EMca, and EMwmean.The EM was used to predict the potential geographical distributions of P. armeniaca, M. sieversii, and P. ledebouriana.For assessing the accuracy of the model, the true skill statistic (TSS) and area under the receiver operating characteristic curve (ROC-AUC) were employed.The ROC curve compares the false-positive rate (1 − specificity) to the true-positive rate (1 − omission rate) [58].The TSS metric, which measures the average omission error, is independent of the size of the verification dataset and combines the benefits of the kappa statistic (occurrence, sensitivity, and specificity) without frequency dependence [59].Higher TSS and AUC values indicate better model accuracy.In the suitability zoning phase, the biomod2 model results were integrated into ArcGIS to produce raster data.The predicted suitability zones for the three species were classified into four grades: high suitability (0.66-1), medium suitability (0.33-0.66), low suitability (0.05-0.33), and unsuitable (0-0.05)[60].

Geographical Distribution Migration and Overlaps
We employed the SDM tool, which was integrated into the ArcGIS software (version 10.4), to examine potential shifts in the geographical distributions of three important species under various climate scenarios.Following the methodology outlined by Brown (2014), we mapped the distribution of these species using the SDM tool (available at http://www.sdmtoolbox.org,accessed on 1 September 2023).Our analysis focused on the changes in their distributional centroids, which were assessed based on a suitability threshold.Specifically, habitats with a probability of >0.3 were classified as 'suitable', while those with a probability of 0.3 or less were considered 'unsuitable'.
We employed the Centroid Changes (Lines) tool from the SDM Toolbox (v2.4) to analyse the migration of potential geographic distribution centres using binarised rasters.Initially, the rasters representing the geographic distributions of the three important species were scaled by factors of 1, 2, and 4 using ArcGIS 10.4's (Environmental Systems Research Institute, lnc.RedLands, CA, USA) Raster Calculator.These scaled rasters were then aggregated to delineate the overlapping distribution areas of the species.This step is crucial for identifying regions in which multiple species converge, indicating a heightened invasive risk.The identification of these common areas is vital for developing targeted conservation strategies to protect ecosystems from the potential impacts of these important species.

Measurement of Ecological Niches
We quantitatively analysed the ecological niches of P. armeniaca, M. sieversii, and P. ledebouriana across different periods and under various climatic conditions.Initially, we compared the ecological niche dynamics between native areas and regions facing habitat loss, utilising occurrence and bioclimatic data with the R Studio (Version 4.2.3,Molly Hill, NJ, USA) ecospat package [61,62].This approach utilised the PCA-env and COUE methods to conduct a comprehensive analysis of the bioclimatic variables associated with the species [48,49].
We then performed climate niche similarity tests between native areas and those experiencing habitat loss, using the ecospat package with 1000 repetitions for each test [63].A significant difference in ecological niche similarity values (p < 0.05) indicated ecological niche dissimilarity.
Furthermore, using the ecospat package, we analysed and visualised the ecological niches of the species, including computing the ecological niche overlap index, 'D'.This index ranged from 0 (no overlap) to 1 (complete overlap), representing the extent of ecological niche overlap.We also used the ENMTool niche breadth module to calculate ecological niche breadths based on current and future potential distribution data [64].Here, 'B1' denotes the minimum and 'B2' denotes the maximum niche breadth.

Marxan Model Construction
The Marxan model, known for its effectiveness in identifying minimum-cost areas for conservation planning [65,66], was used to identify priority conservation areas for P. armeniaca, M. sieversii, and P. ledebouriana under current climatic conditions.Owing to continuous improvements, this model has become popular for land conservation planning.It operates at a spatial resolution of 2 km 2 using square planning units (PUs), each measuring 25,000 m in height and width.In ArcGIS, we used the 'Zonal Statistics as Table ' tool to aggregate species distribution data within each PU, which helped us construct a species distribution matrix.
We set a conservation target that includes 30% of the total habitat area, applying a species protection factor (SPF) of 100.A crucial aspect of the Marxan model is its boundary length modifier (BLM), which was set at 25,000.The BLM acts as a correction parameter for the perimeter of a conservation area [67].By adjusting the BLM, we analysed the balance between the cost, total boundary length, and area.Such an analysis is essential for determining the equilibrium point that leads to a more efficient spatial distribution pattern in conservation areas [68].To ensure the reliability of our findings, the model was run over 100 iterations to determine the optimal configuration of the planning units.
determining the equilibrium point that leads to a more efficient spatial distribution pattern in conservation areas [68].To ensure the reliability of our findings, the model was run over 100 iterations to determine the optimal configuration of the planning units.
The developed EMs demonstrated the highest accuracy in this assessment.For P. armeniaca, M. sieversii, and P. ledebouriana, the TSS values of the EMs were 0.92, and the AUC values were up to 0.98 (Figure 3).Hence, the EMs developed from singular high-precision models significantly enhanced the fit accuracy while diminishing uncertainties in the fitting process.This suggests a high reliability in the predicted potential geographic distributions of the three important species, as inferred through the use of EMs.The developed EMs demonstrated the highest accuracy in this assessment.For P. armeniaca, M. sieversii, and P. ledebouriana, the TSS values of the EMs were 0.92, and the AUC values were up to 0.98 (Figure 3).Hence, the EMs developed from singular high-precision models significantly enhanced the fit accuracy while diminishing uncertainties in the fitting process.This suggests a high reliability in the predicted potential geographic distributions of the three important species, as inferred through the use of EMs.

Present Potential Geographic Spread
The EM projections of the current (1979-2013) potential geographic ranges for P. armeniaca, M. sieversii, and P. ledebouriana are shown in Figure 4. P. armeniaca and M. sieversii share common distribution zones in Armenia, Azerbaijan, China, Iran, Italy, Kazakhstan, Turkey, and Kyrgyzstan, while P. ledebouriana is primarily native and endemic to Kazakhstan.Notably, the shared areas in China, Kazakhstan, and Turkey represent 73.38% of the total joint distribution for P. armeniaca and M. sieversii.China has the largest shared area, at 8.76 × 10 6 ha, accounting for 37.32% of the total, followed by Kazakhstan (27%, 6.34 × 10 6 ha) and Turkey (9.05%, 2.12 × 10 6 ha) (Table 2).The largest global habitat suitability area for P. ledebouriana is significantly concentrated in Kazakhstan (979.17 × 10 6 ha), followed by the global habitat suitability areas for M. sieversii (468.20 × 10 6 ha) and P. armeniaca (445.51 × 10 6 ha).
P. armeniaca has a highly suitable habitat spread over 151.14 × 106 ha, predominantly in Europe (Ukraine, Romania, Russia), Central Asia (Kazakhstan, Turkmenistan, Pakistan, northwestern China), and North America (United States, southern Canada), with sparse distribution in Oceania (Australia).Eastern Europe is the main area of the most suitable habitat for P. armeniaca, accounting for 70.4% of the total suitable area, followed by Central Asia and North America with 21.89% and 0.44%, respectively (Table 3).For M. sieversii, highly suitable habitats, covering 121.99 × 10 6 ha, are mainly in Central Asia (Mongolia, Iran, Turkey, Kazakhstan, Kyrgyzstan, Tajikistan, Turkmenistan, northwestern China), Europe (Georgia, Spain), Africa (Morocco, Algeria, Swaziland), South America (Chile, Argentina, Trinidad, and Tobago), and North America (United States, southern Canada).The highly suitable habitat of M. sieversii is concentrated in Central Asia with 56.05%, followed by North America (16.39%),Africa (10.32%),Europe (6.31%), and South America (6.31%) (Table 4).For P. ledebouriana, highly suitable habitats span 468.86 × 106 ha, predominantly in Europe (Azerbaijan, Ukraine, Russia, Romania, Bulgaria, Germany), Central Asia (Kyrgyzstan, Kazakhstan, Iran, Turkey, northwestern China), and North America (United States, southern Canada).The main habitat of P. ledebouriana is located in Europe, which accounts for 60.51% of the total area, while Central Asia and North America account for 19.66% and 11.89%, respectively (Table 5).

Comparative Analysis of Ecological Niches
The geographic ranges of these three important species partially mirror their fundamental niches, defined as the array of environmental conditions allowing their survival.The ecological niche dynamics of M. sieversii, P. armeniaca, and P. ledebouriana, based on a comparison of the climate niche space between native and lost habitats, are presented in Figure 5. Based on occurrence data and bioclimatic variables for native and habitat loss, the ecological niche overlaps of P. armeniaca, M. sieversii, and P. ledebouriana revealed Schoener's D values of 0.83, 0.74, and 0.76, respectively, indicating that the ecological niche overlap is relatively high between P. armeniaca, M. sieversii, and P. ledebouriana (Figure 5).The ecological niches of P. armeniaca, M. sieversii, and P. ledebouriana in their extirpated habitats were smaller than those in their native habitats.This indicates a decrease in resources that can be jointly utilised in both the current and future periods.In terms of habitat loss, the ecological niches of P. armeniaca, M. sieversii, and P. ledebouriana did not occupy all the ecological niches of their native areas.The null hypothesis of the ecological niche equivalence of P. armeniaca, M. sieversii, and P. ledebouriana based on the bioclimatic variables of the native and invasive areas was not rejected (p = 0.1908, 0.9523, and 0.5238, respectively).
The niche widths of P. armeniaca, M. sieversii, and P. ledebouriana were calculated using the ENMTools (Version 5.26) software package, focusing on various climate conditions.As shown in Table 6, the B1 and B2 values of P. armeniaca, M. sieversii, and P. ledebouriana were all >0.7 and >0.9, respectively.Hence, each period showed no marked distinction between B1 and B2, suggesting that P. armeniaca, M. sieversii, and P. ledebouriana tend towards being generalist species.Furthermore, B1 and B2 exhibited an upward trend in alternative climate scenarios compared with the present period.This trend implies an expansion in the range of resources utilised by these species in forthcoming climate conditions, demonstrating their extensive distribution and robust environmental adaptability.
distinction between B1 and B2, suggesting that P. armeniaca, M. sieversii, and P. ledebouriana tend towards being generalist species.Furthermore, B1 and B2 exhibited an upward trend in alternative climate scenarios compared with the present period.This trend implies an expansion in the range of resources utilised by these species in forthcoming climate conditions, demonstrating their extensive distribution and robust environmental adaptability.

Projected Future Geographic Ranges
The potential global geographical distributions of P. armeniaca, M. sieversii, and P. ledebouriana under SSP126, SSP245, and SSP585 during the 2050s and 2090s are shown in Figures 6-8 and Table 7.For P. armeniaca, the total area of suitable habitat is projected to decrease under most future climate scenarios in the 2050s and 2090s.The most significant reduction occurs in the 2090s under the SSP585 scenario, with the suitable habitat area decreasing to 885 × 10 6 ha, representing a 32.21% reduction from current conditions.Habitat losses are primarily noted in Romania, Ukraine, southern Russia, and sporadically across Central Asia, although some expansions are noted in Central Asia and southern Europe.For M. sieversii, the total area of suitable habitat is projected to decrease under all climate scenarios in the 2050s and 2090s compared with current conditions.The most significant loss occurs in the 2090s under the SSP585 scenario, with the suitable habitat area decreasing to 1389 × 10 6 ha, a 3.71% reduction.Habitat losses are mainly in Romania, Ukraine, southern Russia, and sporadically in Central Asia, although some habitat expansions are observed mainly in Central Asia, with a few in southern Europe.Other notable reductions include the SSP126 scenario in the 2090s with a 3.34% reduction, followed by the SSP245 scenario in the 2050s and the SSP126 scenario in the 2050s with smaller decreases.For P. ledebouriana, the total area of suitable habitat is expected to decrease under all future climate scenarios in the 2050s and 2090s compared with current conditions.The most significant loss is forecasted for the 2050s under the SSP585 scenario, with the suitable habitat area reducing to 517.95 million ha, a 47.1% decrease.The primary habitat losses are noted along the borders of Ukraine, Russia, and Georgia, with smaller affected areas in Kazakhstan, Mongolia, and northwestern China.Some habitat expansions are observed in Moldova and limited areas of southwestern Russia.Other significant reductions are projected under the SSP126 scenario in the 2090s with a 46.68% decrease, the SSP585 scenario in the 2090s with a 45.13% decrease, and the SSP126 scenario in the 2050s with a 43.99% decrease.

Overlapping Geographical Distributions under Climate Change
In the future, the primary regions experiencing a loss of overlapping geographic distribution areas for the three important species will mainly be located in Azerbaijan, Georgia, Kazakhstan, Tajikistan, Kyrgyzstan, Iran, and northwestern China, as illustrated in Figure 9.The area of lost overlapping geographic distributions for the three important species was largest for the 2090s under the SSP245 scenario, followed by the 2050s under SSP126, 2050s under SSP245, 2050s under SSP585, 2090s under SSP585, and 2090s under SSP126.
armeniaca are Friuli-Venezia Giulia, Italy, under current climate scenarios; the centres of potential geographical distributions of M. sieversii are Karaman, Turkey, under current climate scenarios; and the centres of potential geographical distributions of P. ledebouriana are Vasvar, Vas, Hungary, under current climate scenarios.Under the three scenarios from the present to the 2050s and 2090s, P. armeniaca showed an overall trend of moving northward and from northeastern Europe to Central Asia.For M. sieversii, from the present to the 2050s and 2090s, the potential geographic distribution centres generally shifted northward and northwestward in Europe, with the centre oscillating under the SSP245 scenario.For P. ledebouriana, from the current period to the 2050s and 2090s, the potential geographic distribution centres tended to move northward from northeastern Europe to Central Asia, exhibiting a swaying trend in Central Asia.Overall, the potential geographic distribution centres of these three important species predominantly trend northward under the three scenarios projected for the 2050s and 2090s (Table 9).
Forests 2024, 15, x FOR PEER REVIEW 18 of 27 P. armeniaca are Friuli-Venezia Giulia, Italy, under current climate scenarios; the centres of potential geographical distributions of M. sieversii are Karaman, Turkey, under current climate scenarios; and the centres of potential geographical distributions of P. ledebouriana are Vasvar, Vas, Hungary, under current climate scenarios.Under the three scenarios from the present to the 2050s and 2090s, P. armeniaca showed an overall trend of moving northward and from northeastern Europe to Central Asia.For M. sieversii, from the present to the 2050s and 2090s, the potential geographic distribution centres generally shifted northward and northwestward in Europe, with the centre oscillating under the SSP245 scenario.For P. ledebouriana, from the current period to the 2050s and 2090s, the potential geographic distribution centres tended to move northward from northeastern Europe to Central Asia, exhibiting a swaying trend in Central Asia.Overall, the potential geographic distribution centres of these three important species predominantly trend northward under the three scenarios projected for the 2050s and 2090s (Table 9).We utilised the EMca integrated model to evaluate the impact of each environmental variable on the potential geographic distributions of P. armeniaca, M. sieversii, and P. ledebouriana, as detailed in Figure 11.The analysis revealed that for P. armeniaca, annual mean temperature (Bio1), isothermality (Bio3), and precipitation in the wettest month (Bio13) had the most significant cumulative contributions.For M. sieversii, the key bioclimatic variables were precipitation in the wettest month (Bio13), driest month (Bio14), warmest quarter (Bio18), and coldest quarter (Bio19).For P. ledebouriana, the major contributing variables for P. ledebouriana were isothermality (Bio3), precipitation in the wettest month (Bio13), and temperature seasonality (Bio4).These variables predominantly define the multidimensional ecological niche of each species; however, other climatic, soil, and topographic factors play smaller roles.

Priority Protection Areas under Current Climate Conditions
The Marxan model was employed to identify priority conservation areas for the three important species, with the findings imported into ArcGIS to develop a focused conservation strategy.As depicted in Figure 12, these priority areas are predominantly located in

Priority Protection Areas under Current Climate Conditions
The Marxan model was employed to identify priority conservation areas for the three important species, with the findings imported into ArcGIS to develop a focused conservation strategy.As depicted in Figure 12, these priority areas are predominantly located in Armenia, Azerbaijan, China, Iran, Italy, Kazakhstan, Turkey, and Kyrgyzstan.This distribution aligns with the highly suitable habitats for these species as forecasted by the biomod2 model, affirming the precision of these predictions.Furthermore, these crucial conservation zones cover a relatively small portion of the land, exhibiting a compact distribution.This is beneficial for establishing specific conservation and management plans.

Priority Protection Areas under Current Climate Conditions
The Marxan model was employed to identify priority conservation areas for the three important species, with the findings imported into ArcGIS to develop a focused conservation strategy.As depicted in Figure 12, these priority areas are predominantly located in Armenia, Azerbaijan, China, Iran, Italy, Kazakhstan, Turkey, and Kyrgyzstan.This distribution aligns with the highly suitable habitats for these species as forecasted by the bio-mod2 model, affirming the precision of these predictions.Furthermore, these crucial conservation zones cover a relatively small portion of the land, exhibiting a compact distribution.This is beneficial for establishing specific conservation and management plans.

Influences of Environmental Data and Spatial Resolutions on SDM Efficacy
Beyond bioclimatic factors, a range of elements can affect the performance of SDMs [69], including topographic features and human impacts [70,71], which play distinct roles in determining species distributions.In this study, climate, soil, digital elevation model (DEM), and habitat integrity index (HII) data were integrated into the SDMs to project the global distribution patterns of the three important species, with a focus on prediction accuracy [51].We utilised environmental data of varying resolutions.However, existing studies suggest that increased spatial resolution does not necessarily enhance SDM predictive accuracy.Therefore, to reduce uncertainties linked to disparate spatial resolutions in environmental data, we normalised the resolution of all the environmental variables.

Importance of Predictions by EMs
Individual SDMs are often used to predict species colonisation and extinction; however, they can suffer from overfitting or inadequacy [72].EMs, which combine multiple SDM predictions, offer greater accuracy and reduced uncertainty, making them ideal for forecasting the potential geographic distributions of important and endangered species under various conditions, such as climate change and human impacts [34,35,73].In this study, 11 individual SDMs (ANN, CTA, FDA, GAM, GBM, GLM, MARS, MAXENT, MAXNET, RF, and XGBOOST) were used to create the EM.Despite the high TSS (>0.8) and AUC (>0.9) values for the individual SDMs, the EM achieved even better scores (>0.92 and >0.98), enhancing the prediction accuracy.
P. armeniaca, M. sieversii, and P. ledebouriana are vital for the stability and ecosystem services of wild fruit forests, offering significant genetic resources and insights into plant evolution.However, their global geographical and niche overlaps remain unexplored.Using the EM in biomod2, we predicted the global potential distribution and niche breadth of these three important species under climate change conditions, evaluating future habitat loss, overlap of lost and native areas, reserve planning, and key environmental variables influencing their distribution.This study provides a comprehensive approach for conserving and utilising similar species globally.

Response of Spatial Distribution Patterns of Three Important Species to Climate Change
This century, most regions of the northern hemisphere are expected to experience the effects of climate change.Under climate change, species may face three potential outcomes: adapting in place, migrating to track shifting climates spatially temporally, or facing local extinction [74,75].Predictions of the potential geographic distribution of vulnerable species under climate change can serve as an early warning signal [76].This change may cause range dislocations and species turnover, threatening native flora and increasing the risk of extinction [77].The geographic ranges of M. sieversii, Juglans regia, Prunus armeniaca, Crataegus chlorocarpa, Prunus cerasifera, and Sorbus tianschanica are expected to shrink owing to climate change, leading to a decline in the diversity of wild fruit forests in the Xitianshan Mountains [78].Furthermore, studies indicate that the geographic range of M. sieversii has declined owing to climate change, with nature reserves covering only a small fraction of the remaining suitable areas [79].A consequence of climate change is the shift of plant communities to higher elevations [80][81][82][83] and new dimensions [84,85] to follow favourable climate conditions.This inability to adapt threatens rare species, potentially leading to population declines and eventual extinction [86].Contrary to previous studies, our research indicates that the geographic ranges of P. armeniaca, M. sieversii, and P. ledebouriana are diminishing owing to climate change, with suitable habitats moving to higher latitudes, thereby increasing the risk of extinction from habitat loss.Our findings corroborate the hypothesis that climate change intensifies habitat loss for important species, enhance our understanding of how global warming influences the geographic shifts of these species, and offer a scientific foundation for devising future conservation strategies.

Environmental Factors
Restricting the Distributions of P. armeniaca, M. sieversii, and P. ledebouriana Under current climatic conditions, the primary environmental factors affecting the distribution of P. armeniaca, M. sieversii, and P. ledebouriana are temperature and precipitation, followed by topographic factors; soil factors have the least impact on these species.According to the EM model, suitable habitats for P. armeniaca, M. sieversii, and P. ledebouriana are mainly at altitudes of 1000-1400 m, which aligns with their actual living conditions [24,26,27].Although thermohydric conditions play a major role in the potential global geographic distribution patterns of P. armeniaca, M. sieversii, and P. ledebouriana, the constraints of topography and soil factors should not be overlooked.The contribution rates of environmental factors indicate that a combination of factors, including temperature, moisture, altitude, and slope, will jointly affect the potential geographic distributions of P. armeniaca, M. sieversii, and P. Ledebouriana.4.5.Priority Protection Areas for P. armeniaca, M. sieversii, and P. ledebouriana Nature reserves are essential to safeguard natural resources and biodiversity.In exploring systematic reserve zoning for P. armeniaca, M. sieversii, and P. ledebouriana, we identified priority conservation areas in Armenia, Azerbaijan, China, Iran, Italy, Kazakhstan, Turkey, and Kyrgyzstan.These countries offer ideal environments for these species largely because of their unique geographical locations and climatic conditions.The regions, which are mainly in temperate to semi-arid climate zones, consist of mountainous and plateau terrains that foster diverse ecological environments [27,87].This diversity allows species to thrive at various altitudes and microclimates, matching their specific growth habits.
The key growth factors in these areas include ample sunlight, a moderate climate, and balanced moisture.The mountainous regions of Armenia and Azerbaijan offer varied microclimates, whereas plateau areas in Western China and Central Asia provide essential nutrients with unique soil types and textures.The semi-arid climates of Iran and Turkey facilitate efficient water use, whereas the warm climates and good soil drainage of Italy and Kyrgyzstan are beneficial for these species.These conditions not only align with the growth habits of P. armeniaca, M. sieversii, and P. ledebouriana, but also support their biodiversity and genetic diversity.Therefore, these regions are crucial for the growth and conservation of these valuable species and offer significant opportunities for research and preservation.Furthermore, while in situ conservation within these natural reserves is vital, the importance of ex situ conservation efforts cannot be understated.Ex situ strategies, such as the establishment of seed banks and living collections, play a crucial role in complementing in situ measures by providing a 'safety net' against potential loss of genetic diversity due to environmental or anthropogenic pressures [88,89].This dual approach ensures the preservation of a genetic reservoir for P. armeniaca, M. sieversii, and P. ledebouriana, which can be crucial for restoration and research purposes under changing global conditions.The integration of ex situ and in situ conservation strategies will therefore enhance our capacity to maintain and recover these species in their natural habitats, while also allowing for controlled scientific studies and breeding programs that can further support their conservation.

Model Prediction Limitations
This study's modelling and analysis present inherent limitations and challenges.First, the field surveys lacked comprehensiveness, potentially skewing the species distribution data away from typical patterns.This could introduce biases, with notably less thorough surveys in remote and complex terrains compared with more accessible areas.Second, the exclusion of biological factors as predictor variables limited the simulation to theoretical ecological niches, rather than representing the species' actual ecological niches [90,91].Although the impact of human footprint was considered, urbanisation and deeper anthropogenic influences, which are particularly significant in densely populated areas of Europe and beyond, were not explicitly modelled.This oversight may affect the accuracy of our predictions as urban expansion can alter local ecosystems, climate conditions, and species interactions [92,93].Extrapolating models to future environmental conditions beyond current training data impacts prediction reliability [94].Additionally, unmodelled factors such as migration capabilities, barriers, and evolutionary responses to environmental changes could affect species distribution predictions [95].Therefore, it is crucial to consider these factors when applying predictions to field surveys and conservation efforts [96].The environmental variables for this study were sourced from the Paleoclim database, covering the period 1979-2013.This range extends nearly a decade beyond the 1950-2000 period available in the WorldClim database [52,74].However, climate data for the most recent decade remain absent.Consequently, it is essential to include the missing data in future research to enhance the reliability and robustness of the predictions.

Conclusions
The potential geographical distributions of P. armeniaca, M. sieversii, and P. ledebouriana are shaped by bioclimatic variables, the HII, soil attributes, and topography, with global climate change significantly impacting their distribution and overlapping areas.Our EM outperformed individual models such as ANN, CTA, FDA, GAM, GBM, GLM, MARS, MAXENT, MAXNET, RF, and XGBOOST, showing that it is more reliable for predicting the geographical distributions of these important species.
The EM predictions showed that these species are primarily distributed in Central Asia and Europe and undergo ecological niche changes during invasion.Under climate scenarios SSP126, SSP245, and SSP585, suitable habitats are projected to decrease by the 2050s and 2090s.In the 2050s, the overlapping geographic distribution areas will mainly be in Azerbaijan, China, Iran, Kazakhstan, and Georgia.By the 2090s, these areas are expected to shift predominantly to China, Kazakhstan, Kyrgyzstan, and Georgia, with a general trend of moving northward.
Bioclimatic variables and elevation are significant influencers of their distribution, whereas the cumulative impact of topography and soil properties is low.The potential loss of distribution areas for these species is a serious concern, threatening the genetic diversity and adaptability of wild fruit forest ecosystems to environmental change.
In conclusion, the protection of P. armeniaca, M. sieversii, and P. ledebouriana extends beyond biodiversity conservation, and is essential for ecosystem vitality, scientific advancement, and economic resilience.A comprehensive conservation strategy that includes habitat protection, natural reproduction support, and management plans is crucial.This approach, guided by EMs and ecological studies, is the key to preserving biodiversity and ensuring sustainable coexistence in the face of climate change, habitat loss, and carbon sequestration.

Figure 2 .
Figure 2. Evaluation indices of individual predictive models (AUC: area under the receiver operating characteristic, TSS: true skill statistic, ANN: artificial neural network, CTA: classification tree analysis, FDA: flexible discriminant analysis, GAM: generalised additive model, GBM: generalised

Figure 6 .
Figure 6.Potential global geographical distribution of Prunus armeniaca in the 2050s and 2090s predicted using the ensemble model (EM).The scenarios include SSP126, SSP245, and SSP585.Figure 6. Potential global geographical distribution of Prunus armeniaca in the 2050s and 2090s predicted using the ensemble model (EM).The scenarios include SSP126, SSP245, and SSP585.

Figure 6 .
Figure 6.Potential global geographical distribution of Prunus armeniaca in the 2050s and 2090s predicted using the ensemble model (EM).The scenarios include SSP126, SSP245, and SSP585.Figure 6. Potential global geographical distribution of Prunus armeniaca in the 2050s and 2090s predicted using the ensemble model (EM).The scenarios include SSP126, SSP245, and SSP585.

Figure 7 .
Figure 7. Potential global geographical distribution of Malus sieversii in the 2050s and 2090s predicted using the ensemble model (EM).The scenarios include SSP126, SSP245, and SSP585.Figure 7. Potential global geographical distribution of Malus sieversii in the 2050s and 2090s predicted using the ensemble model (EM).The scenarios include SSP126, SSP245, and SSP585.

Figure 7 .
Figure 7. Potential global geographical distribution of Malus sieversii in the 2050s and 2090s predicted using the ensemble model (EM).The scenarios include SSP126, SSP245, and SSP585.Figure 7. Potential global geographical distribution of Malus sieversii in the 2050s and 2090s predicted using the ensemble model (EM).The scenarios include SSP126, SSP245, and SSP585.

Figure 8 .
Figure 8. Potential global geographical distribution of Prunus ledebouriana in the 2050s and 2090s predicted using the ensemble model (EM).The scenarios include SSP126, SSP245, and SSP585.

Figure 8 .
Figure 8. Potential global geographical distribution of Prunus ledebouriana in the 2050s and 2090s predicted using the ensemble model (EM).The scenarios include SSP126, SSP245, and SSP585.

Figure 10 .
Figure 10.Centroid changes in the distribution of Prunus armeniaca, Malus sieversii, and Prunus ledebouriana in future climate scenarios.

Figure 10 .
Figure 10.Centroid changes in the distribution of Prunus armeniaca, Malus sieversii, and Prunus ledebouriana in future climate scenarios.

Figure 12 .
Figure 12.Global priority conservation areas for Prunus armeniaca, Malus sieversii, and Prunus ledebouriana as predicted by the Marxan model.

Figure 12 .
Figure 12.Global priority conservation areas for Prunus armeniaca, Malus sieversii, and Prunus ledebouriana as predicted by the Marxan model.

Table 1 .
Participation modelling environment variables.

Table 2 .
Area of overlap under the current climate conditions.

Table 6 .
Niche widths for three important species.

Table 6 .
Niche widths for three important species.

Table 7 .
Global area of distribution of three important species under future climate scenarios (×10 6 ha).

Table 7 .
Global area of distribution of three important species under future climate scenarios (×10 6 ha).

Table 9 .
The centres of potential geographical distributions of three important species.