Located in the heart of the Eurasian continent, Central Asia comprises the sovereign nations of Kazakhstan, Kyrgyzstan, Tajikistan, Turkmenistan, and Uzbekistan, as well as China’s Xinjiang Uygur Autonomous Region, a provincial-level administrative division. The total area of this expansive region is approximately 560 × 104 km². The terrain in this region is complex, with the high-elevation Altai Mountains, Tianshan Mountains, Pamir Plateau, and Hindu Kush Mountains in the southeast; the Turan Plain and the Caspian Sea littoral in the west; and the Kazakh Hills in the north. The topography is elevated in the southeast and lower in the northwest. The climate is mainly temperate with dry and semi-dry conditions, and certain regions experience influences from Mediterranean and plateau climates. The temperature changes drastically, with a range of 20°C–30°C between day and night. The mean temperature during summer falls between 20°C and 30°C, while in winter, the average temperature drops to below 0°C (Hua et al., 2022). Rainfall across Central Asia is mainly concentrated during the cooler months of winter and spring, varying from more than 1,000 mm in the western Hisar Mountains and Fergana Mountains to less than 100 mm across the eastern regions (Huang et al., 2024). Central Asia is rich and unique in plant resources. Currently, 9,520 species of higher plants have been identified; among these, 20% are unique to the region, classified under 138 families and 1,176 genera. The botanical geography of Central Asia is categorized into five large regions and 33 small regions, with over 65% of the species exhibiting a Central Asian-specific distribution (Zhang et al., 2020).

In the current investigation, location data for 197 species of B. tianschanica were obtained via the Global Biodiversity Information Facility (GBIF, 2024) and Chinese Virtual Herbarium (2024), with 78 obtained through field investigation. The data were saved in CSV format and imported into the ArcGIS 10.2 software to remove duplicates and invalid entries. To minimize spatial autocorrelation between geographic location coordinates and to maintain consistency with the spatial extent of environmental variables, only one of the two samples existing within the same grid cell was retained. Finally, 116 species’ distribution records ( Figure 1 ) were sorted out and saved in ASCII format, which will be used for subsequent distribution model analysis. The national border data were from the Ministry of Natural Resources-Standard Map Service (2024) [Drawing No. GS (2021) No. 5443].

In this study, 19 (Bio1–Bio19) climate data of SSP-RCP2.6 (SSP1-2.6), SSP-RCP4.5 (SSP2-4.5), and SSP-RCP8.5 (SSP5-8.5) were downloaded from the World Climate Data Network in the present period (1970–2000) and the future period (2041–2060 and 2061–2080). The SSP1-2.6 trajectory represents an environmentally sustainable route, with greenhouse gas emissions minimized, and the global temperature is projected to increase by 1.5°C by 2100; SSP2-4.5 is a moderate development trajectory, where emissions of greenhouse gases are at a medium level, and the global temperature is expected to climb by 2.1°C by 2100; SSP5-8.5 is a fossil fuel-intensive path, with significantly increased greenhouse gas output, and the global temperature will see an ascent to 4.7°C by 2100 (Zhou et al., 2021). At the same time, digital elevation model (DEM) data were sourced from the General Bathymetric Chart of the Oceans (GEBCO Compilation Group, 2024) to consider the influence of topography on species distribution. Although many research efforts typically focus on bioclimatic variables and topographic parameters for constructing the model, given that soil characteristics play a crucial role in influencing species distribution, this study also downloaded soil data from the Food and Agriculture Organization’s databases (Fischer et al., 2008) to ensure the comprehensiveness and accuracy of the model. Multicollinearity may lead to inaccuracy and over-fitting of model prediction results. Multicollinearity can result in erroneous model predictions and overfitting. To mitigate this, a Jackknife analysis was conducted for the 36 ecological variables within the MaxEnt model to assess their contribution to model predictions. Variables with no contribution were eliminated, and those with significant contributions were selected. Additionally, ENMtools were utilized for a correlation analysis of soil and climate variables. As a result, 27 environmental variables were chosen for model analysis ( Table 1 ), reducing multicollinearity while enhancing the model’s predictive accuracy.

The gathered occurrence data and ecological variable information on B. tianschanica were converted to ASCII format using ArcGIS 10.2, enabling their integration with MaxEnt 3.4.1 for model processing. In the process, the training subset was allocated 75%, while the testing and validation subset was assigned a 25% share. The number of iterations was fixed at 1,000, and data simulation involved 10 repetitions of the training process (Wu et al., 2022). To assess the limiting influence of individual environmental factors on the spatial distribution patterns of B. tianschanica, the Jackknife module in the MaxEnt 3.4.1 software was employed for calculating the relative contribution and the significance through permutation for each ecological parameter. The receiver operating characteristic (ROC) curve was applied to assess the precision of the model’s predictions (Zhu et al., 2017). The area under the curve (AUC) score ranged from 0 to 1.0, and the higher the AUC figure, the better the predictive performance of the model (Wang et al., 2007). The simulation results were output in the Cloglog format and saved in ASCII format, and the remaining parameters used the default settings of the MaxEnt 3.4.1 software.

The random forest model was constructed through the Python software, with B. tianschanica as the target tree species. First, the distribution point data of B. tianschanica were read from the Shapefile, and the background points without the distribution of the target tree species were generated. Then, the TIFF files of all environmental indicators were read from the specified directory, the environmental indicator data of the species points and background points were extracted, and the outliers were removed. After that, the random forest model was trained using the training set data, the importance of the features was calculated, and the top 10 features were selected based on the importance of retraining the model. Finally, the model was applied to forecast the validation set, the performance metrics were computed, and the ROC graph was created. In addition, the full prediction scope was estimated, and the forecasted data were exported as a new TIFF image.

The AUC values were obtained after simulation using the MaxEnt 3.4.1 software and repeated 10 times ( Table 2 ). The findings suggest that the average training set AUC for the potential distribution model of B. tianschanica under current climate conditions was 0.970, and the average training set AUC for the future climate scenario predictions was above 0.97, demonstrating that the outcomes from the MaxEnt model exhibit high accuracy and reliability and that it is capable of forecasting the potential geographic distribution for B. tianschanica. The mean AUC score for the RF model with respect to the prediction results of B. tianschanica was 0.873 ( Table 2 ), reflecting a considerable predictive precision. The forecasting performance of the RF model was inferior to that of the MaxEnt model, and the results concerning the potential suitable habitat area of B. tianschanica were slightly less accurate compared to those of the MaxEnt model.

Based on the outcomes of the MaxEnt analysis ( Table 3 ) ( Figure 2 ), by selecting the six environmental factors with larger contribution rates, it is found that the percentage of gravel volume in the lower layer, the percentage of gravel volume in the upper layer, elevation, precipitation of the driest month, mean temperature of the coldest quarter, and organic carbon content significantly influence the potential distribution of B. tianschanica, and their cumulative contribution rate reached 62.1%. The collective contribution of soil attributes (percentage of gravel volume in the lower layer of soil, percentage of gravel volume in the upper layer of soil, and the level of organic carbon within the lower layer of soil) amounted to 33.6%, whereas the aggregate contribution of climatic factors (precipitation of the driest month and mean temperature of the coldest quarter) was 17.2%, and the combined effect of topographical variables (elevation) was 11.3%.

The MaxEnt model’s environmental variable response curves ( Figure 3 ; considering the parallel trends across the SSP1-2.6, SSP2-4.5, and SSP5-8.5 climate projections, only the SSP2-4.5 scenario is illustrated here for exemplification) elucidate the correlation of B. tianschanica’s occurrence likelihood with pivotal ecological variables. Generally, a plant’s presence probability exceeding 0.5 indicates suitable conditions for growth (Guo et al., 2019). Currently, optimal growth conditions for B. tianschanica include a lower soil gravel volume percentage between 2.8% and 29.8%, an upper soil gravel volume percentage between 2.1% and 11.8%, an elevation range of 695 to 2,211 m, an average temperature in the coldest quarter between −11.1°C and −2.8°C, and precipitation in the driest month ranging from 0.2 to 8.2 mm. Moreover, the presence probability of B. tianschanica is maximized when the lower soil organic carbon content exceeds 1.7%. Predictions for 2041–2060 suggest changes under suitable growth conditions: the lower soil gravel volume percentage will adjust to 3.1%–20.6%, the upper soil gravel volume percentage will range from 2.1% to 15.3%, the suitable elevation will decrease to 638–1,836 m, the average temperature in the coldest quarter will rise to −7.7°C to −1.1°C, and the precipitation in the driest month will increase to 0.5–13.9 mm, with the lower soil organic carbon content above 0.3%. Further projections for 2061–2080 indicate continued adjustments in growth conditions: the lower soil gravel volume percentage will be 2.8%–14.7%, the upper soil gravel volume percentage will be 1.9%–11.2%, the suitable elevation will be 650–1,877 m, the average temperature in the coldest quarter will range from −8.1°C to −0.7°C, the precipitation in the driest month will be 0.4–12.7 mm, and the lower soil organic carbon content will remain above 0.3%.

The outcomes of the RF model ( Figure 4 ) suggest that terrain is the predominant influence on the occurrence of B. tianschanica, with climate playing a secondary role and soil factors exerting the weakest effect. In detail, altitude emerges as the most crucial terrain-related element, while the most pivotal climatic element is rainfall during the peak rainy period. Presently ( Figure 5 ), B. tianschanica flourishes optimally at altitudes spanning 1,419 to 3,909 meters, accompanied by a wet season rainfall of 69 to 204 mm. Nevertheless, future projections for the period 2041–2060 anticipate a shift: the most favorable growth zones will migrate to higher altitudes, ranging from 1,878 to 3,639 m, with a reduction in wet season rainfall to 61–192 mm. Come 2061–2080, the region suitable for growth is anticipated to ascend further to altitudes between 2,386 and 3,266 m, alongside an uptick in wet season rainfall, varying from 70 to 225 mm.

Utilizing the MaxEnt algorithm, the projected occurrence map for B. tianschanica is depicted in Figure 6 . Regions of prime habitat are predominantly found in the Tianshan Mountains, the Ili River Valley, around Lake Issyk-Kul, and among other locales, and are also scattered across the Turpan Basin, encompassing a zone of 5.81 × 10⁴ km², which constitutes just 6.37% of the overall habitable zone. Zones of moderate habitat preference are largely concentrated in the Tianshan Mountains, along the Ili River, and within the Turpan Basin, covering an expanse of 22.03 × 10⁴ km², representing 24.17% of the aggregate habitable zone. Areas with minimal habitat suitability exhibit an extensive spread, totaling 63.29 × 10⁴ km², or 69.45% of the entire habitable zone. In addition to the Tianshan Mountains, these regions extend to the Irtysh River, Ulungur River, Bogda Mountains, Kazakh Uplands, Lake Balkhash, Amu River, and the midsection of the Syr River.

In contrast to the MaxEnt model, the cumulative suitable habitat for B. tianschanica as predicted by the RF model encompasses an area of 336.21 × 10⁴ km², exceeding threefold the size estimated by the MaxEnt model and showing a more concentrated pattern. Within this, the prime habitat zone spans 89.55 × 10⁴ km², predominantly located in the Tianshan Mountains, the Ili River Valley, Lake Balkhash, Lake Issyk-Kul, the Kazakh Uplands, along the Irtysh River, and the upper course of the Syr River; the zones of moderate habitat suitability are not only adjacent to the prime habitat zones but also extend into the lush regions at the southern periphery of the Karakum Desert, covering an area of 92.54 × 10⁴ km²; the least suitable habitats stretch eastward toward the environs of the Junggar Basin, northward toward the Turgay Depression and the Ural River Basin, and westward toward the southeastern shores of the Caspian Sea, encompassing an area of 154.12 × 10⁴ km².

Examining the information presented in Figure 7 and Table 4 , observations indicate variations in the viable cultivation zones for B. tianschanica across two timeframes (2041–2060 and 2061–2080) as predicted by the MaxEnt model under various climate projections. In the SSP1-2.6 and SSP5-8.5 scenarios, from the present era to 2041–2060, the aggregate suitable cultivation zone for B. tianschanica expanded by 21.58 × 10⁴ km² and 38.87 × 10⁴ km², respectively. The most substantial growth occurred under the SSP5-8.5 projection, expanding by 42.70%. However, under the SSP2-4.5 projection, the aggregate suitable zone in 2041–2060 measured 77.43 × 10⁴ km², marking a 15.03% reduction from the current era. Progressing from the present to 2061–2080, the overall suitable zone for B. tianschanica experienced growth across varying climate projections, yet the rate of expansion waned as CO₂ levels rose, with increases of 19.66%, 10.06%, and 4.98%. On the whole, the high- and moderate-suitability zones for B. tianschanica saw minimal change, predominantly situated in the Tianshan Mountains, along the Ili River, and in the Turpan Basin, whereas the low-suitability zones expanded toward the Kazakh Uplands.

The outcomes of the RF model indicate that from the current era to the 2041–2060 timeframe, the overall viable region for B. tianschanica saw an expansion of 1.63 × 10⁴ km² under the SSP2-4.5 projection, whereas it exhibited a downward trend across other timeframes or projections, particularly pronounced during the 2061–2080 period. This is primarily observed in the shrinkage of the least suitable zones in the Kazakh Uplands and the Turgay Depression region.

Upon analyzing the projected future habitat shifts as forecasted by both models, it is evident that aside from the contraction in the SSP2-4.5 projection (2041–2060) as per the MaxEnt model, there was a general trend of expansion in other projections and periods, with substantial rates of change. Conversely, the RF model displayed an inverse pattern. Beyond the growth in the SSP2-4.5 projection (2041–2060), there was a decline in habitat suitability across other projections and periods, with notable fluctuations in the rates of expansion and contraction.

This research utilized the MaxEnt and RF algorithms, integrating climatic, edaphic, and morphological data from Central Asia, alongside the sampled occurrences of B. tianschanica, to investigate the impact of ecological variables on the prospective range of the species. The findings are considered credible. Field data for some species were collected through extensive surveys, with GPS technology used for precise location marking, enhancing the accuracy and dependability of the model beyond literature or database reliance. Both models demonstrated high predictive precision, with MaxEnt’s AUC exceeding 0.97 and RF’s AUC over 0.86, suggesting MaxEnt’s slightly superior performance. However, RF predicts a much broader potential distribution for B. tianschanica. This discrepancy may stem from differences in model assumptions and sample bias handling (MaxEnt alleviates sample bias through strategic selection and weight adjustment of background points, while RF employs bootstrap sampling methods to effectively reduce bias). MaxEnt, based on maximum entropy, may predict a more even distribution when data points are scarce or unevenly distributed, potentially underestimating actual species presence in certain areas, whereas RF behaves oppositely. The predicted primary distribution areas for B. tianschanica include the Tianshan Mountains, Ili River Basin, Issyk-Kul Lake, Turpan Basin, Irtysh River, Ulungur River, Bogda Mountain, Kazakh Hills, Balkhash Lake, Amu Darya, and the middle reaches of the Syr Darya River. The identified areas are consistent with previous studies (Ding et al., 2021), thereby enhancing the model’s reference value. Furthermore, areas outside the documented regions are identified as potential suitable habitats, where future climatic conditions are expected to support the growth of B. tianschanica.

The distribution of species is governed by various ecological elements, such as climatic conditions, in conjunction with soil and terrain characteristics (Wang and Xing, 2021; Wu and Wang, 2021). This study, focusing on B. tianschanica, employed the MaxEnt and RF models to analyze the relationship between its distribution pattern and these environmental factors. The MaxEnt model indicates that the key ecological elements influencing the occurrence of B. tianschanica are soil properties (such as the proportion of gravel in the lower and upper soil strata and the organic carbon content in the lower soil layer), climatic variables (including rainfall during precipitation of the driest month and mean temperature of the coldest quarter), and terrain attributes (specifically elevation). Presently, the favorable conditions for the growth of B. tianschanica are characterized by a lower soil gravel percentage of 2.8% to 29.8%, altitudes between 695 and 2,211 m, mean temperatures in the coldest quarter from −11.1°C to −2.8°C, and a range of 0.2 to 8.2 mm for precipitation in the driest month. The temperature and moisture needs of plants dictate their vertical distribution, which subsequently impacts individual development and the spatial pattern of tree species (Yu et al., 2016). The simulation forecasts alterations in the thermal and hydration requirements of B. tianschanica moving forward. The average temperature of the coldest quarter is projected to rise by roughly 2°C, prompting a shift to lower altitudes ranging from 650 to 1,877 m. Simultaneously, an enhanced necessity for hydration is anticipated, with the precipitation in the driest month anticipated to escalate to a range of 0.5 to 12.7 mm. Soil texture plays a pivotal role in determining water retention and aeration, both of which are essential for optimal root development and efficient nutrient absorption in plants (Benning and Moeller, 2021). Well-structured soil facilitates robust root growth, which in turn accelerates plant growth rates, improves overall health, and enhances adaptability to environmental conditions. To preserve soil moisture, it is projected that the volume percentage of gravel in the subsoil will decrease by approximately 15%. This adjustment aims to enhance the soil’s water-retention capabilities, thereby promoting healthier root development and superior overall plant performance. The random forest model’s projections reveal that the foremost ecological elements governing the spread of B. tianschanica encompass relief (elevation), meteorological conditions [precipitation of the wettest quarter, precipitation of the warmest quarter, temperature annual range (Bio5–Bio6), and annual precipitation], as well as soil attributes (proportion of silt). B. tianschanica exhibits optimal growth at altitudes between 1,419 and 3,909 m, where the precipitation of the driest quarter is from 69 to 204 mm during the current epoch. As CO₂ concentrations escalate in the future, the vertical zone amenable to the proliferation of B. tianschanica will contract to a range of 2,386 to 3,266 m. Furthermore, the peak rainfall during the height of the rainy season is projected to increase, ranging from 70 to 225 mm. The two analytical models agree that B. tianschanica will exhibit heightened hydration needs in the coming years, whereas the forecasts for thermal requirements vary, resulting in divergent shifts in the optimal altitude spectrum for vegetation. These variances may stem from multiple causes. First, the temporal misalignment between bioclimatic and soil datasets can introduce significant biases in evaluating future plant growth conditions. Second, the study may not have fully accounted for all environmental determinants of plant growth, such as solar radiation and anthropogenic influences. Additionally, the limitations of data sources—namely, the coarse resolution (2.5-minute grid) and restricted temporal scope of the climate data—impede the accurate representation of microscale climatic variability and long-term climatic trends. Collectively, these issues may compromise the robustness and reliability of the predictive models.