Introduction

Global climate change presents significant challenges to the geographical distribution and ecological adaptability of plant species1,2. These environmental shifts manifest as increased frequency of extreme weather events, disrupted seasonal patterns, and profound impacts on the distribution and intensity of key climatic factors such as temperature, precipitation, and light3. Consequently, plant growth cycles, reproductive strategies, and available habitat are subjected to mounting pressures4. As primary producers within ecosystems, the growth conditions of plants are intrinsically linked to the stability and efficiency of fundamental natural processes, including carbon and water cycles. Even minor alterations in ecological niches can initiate cascading effects throughout the food web.

Furthermore, the scarcity and uneven distribution of water resources, compounded by soil degradation phenomena (e.g., salinization, acidification, and fertility decline), exacerbate the survival pressures on plant populations. The synergistic effects of these environmental stressors compel plant populations to continuously adapt or migrate to new, more favorable habitats. However, for many species, particularly those characterized by narrow ecological niches and limited dispersal capabilities, such environmental changes can lead to dramatic reductions in habitat availability or even local extinction5. Scientific evidence indicates that climate change not only diminishes overall biodiversity but also intensifies ecosystem fragmentation by facilitating the invasion of alien species and disrupting established interspecies interactions6. These disruptions can alter species balance, render food web structures fragile, and severely threaten the provision of essential ecosystem services, such as air purification, water conservation, and climate regulation. Shifts in spatial distribution of species can force species adapted to specific environments out of their optimal ranges, thereby increasing their extinction risk, especially for those already endangered7. Therefore, the accurate prediction and comprehensive assessment of the specific impacts of environmental changes on the geographical distribution of particular plant species represent an urgent priority in biodiversity conservation.

Concurrent with an increasing understanding of environmental change, rapid advancements in Geographic Information Systems (GIS) technology have underscored the prominence of Species Distribution Models (SDMs). SDMs, which are based on the concept of ecological niches, are increasingly utilized for predicting species habitats and distribution patterns under various environmental conditions8,9,10. By correlating known species occurrence records with relevant ecological data, SDMs can forecast potential suitable habitats across broad geographical regions and facilitate spatiotemporal inferences11. This capability is essential for informing conservation planning, habitat restoration initiatives, and evidence-based policy-making.

In recent years, progress in statistical techniques has led to the development of diverse algorithms within the SDM framework, including Generalized Linear Models (GLMs), Generalized Additive Models (GAMs), and Generalized Boosting Models (GBMs). Each algorithm offers distinct features for addressing complex species distribution questions. Among these, the Biomod2 platform is notable for its ensemble modeling strategy12. Functioning as a comprehensive framework that integrates multiple modeling algorithms (e.g., GLMs, GAMs, Random Forests), Biomod2 can leverage the strengths of different approaches, thereby circumventing the limitations inherent in single-model applications and enhancing the accuracy and reliability of predictions13,14. Biomod2 also permits user-defined parameter adjustments tailored to specific species and regional characteristics, facilitating predictions that better align with real-world ecological scenarios.

Lilium davidii var. willmottiae (E. H. Wilson) Raffill (L. davidii), commonly known as the Lanzhou Lily, is a perennial herbaceous plant belonging to the Liliaceae family15. This lily, endemic to China, possesses significant economic and medicinal value, being recognized for its “medicinal and edible” properties. Similar to other Lilium species, L. davidii also holds ornamental value due to its vibrant flower coloration16. The core cultivation area for this species is situated in the mountainous regions of Gansu Province, Northwest China17. L. davidii primarily thrives in Lanzhou, Gansu Province, where it has a cultivation history spanning over 400 years and has been designated as a National Geographical Indication product of China. The bulbs of L. davidii are rich in various bioactive compounds, including flavonoids, alkaloids, phospholipids, polysaccharides, saponins, amino acids, proteins, dietary fibers, vitamins, and trace elements. These constituents are reported to exhibit beneficial effects such as immune system enhancement, blood sugar and lipid reduction, and anti-allergy, anti-oxidation, and anti-aging properties18. However, intensified human activities and ongoing environmental changes pose considerable risks of habitat degradation and genetic resource erosion for L. davidii. Consequently, enhancing the protection of natural L. davidii resources is of particular importance.

Current research concerning L. davidii predominantly focuses on aspects such as seedling cultivation techniques16chemical composition analysis19challenges associated with consecutive cropping6and genetic diversity assessment20. Investigations into predicting the suitable habitats for L. davidii under prevailing environmental conditions remain relatively limited. Modeling and predicting the suitable habitat of L. davidii can therefore provide essential guidance for optimizing the management of its core populations, conserving existing resources, and informing introduction and cultivation strategies. To address this knowledge gap, an ensemble modeling strategy, utilizing the Biomod2 framework, was employed to forecast the current potential distribution patterns of L. davidii based on its occurrence data and relevant ecological variables. The primary objectives were to delineate the current potential suitable habitats for L. davidii and to identify the key ecological factors influencing its distribution. The findings are intended to provide scientific theoretical support for the conservation, sustainable utilization, and potential introduction of L. davidii resources.

Materials and methods

Occurrence records

The occurrence records for L. davidii within its distribution range in China were meticulously compiled from multiple sources to ensure comprehensive spatial coverage. An extensive series of field surveys conducted by the research team between 2010 and 2019 yielded an initial 98 distribution records. In addition to these primary field data, 56 occurrence records for L. davidii within China were retrieved from the Global Biodiversity Information Facility (GBIF; accessible at https://www.gbif.org/), and a further 10 records were obtained from pertinent published literature.

To ensure the quality and reliability of the dataset for modeling purposes, several data cleaning and filtering steps were implemented. Initially, records with missing or imprecise coordinates, exact duplicates, or those clearly falling outside the known ecological range of L. davidii were excluded. Subsequently, to address potential spatial autocorrelation and ensure a consistent spatial resolution across the dataset, a spatial thinning process was applied. This involved creating a 5 km × 5 km buffer around each occurrence point and removing redundant records within these buffered areas, retaining only one point per grid cell. This procedure resulted in a final dataset of 120 unique and spatially refined occurrence records used for subsequent analyses. The geographical distribution of the final 120 occurrence records is presented in Fig. 1.

Fig. 1
figure 1

Geographical distribution records of L. davidii.

Full size image

Environmental variables and selection

For this study, a set of 19 bioclimatic variables (bio1–bio19) with a spatial resolution of 2.5 arc-minutes, representing the baseline period of 1970–2000, were sourced from the WorldClim database (version 2.1). Additionally, three topographical variables – elevation (altitude), slope, and aspect – were derived from a digital elevation model (DEM) using ArcMap software (v10.8, ESRI, Redlands, CA, USA). These 22 initial environmental variables (detailed in Table 1) are commonly employed in species distribution modeling studies. All environmental variables, in raster format, were standardized to the WGS1984 datum and resampled to match the spatial resolution of the bioclimatic data.

To mitigate potential multicollinearity among environmental variables, which can lead to model overfitting and affect prediction accuracy, a systematic variable selection process was implemented. This process involved two main steps21: Initially, each of the 22 environmental variables was individually used, along with the L. davidii occurrence records, to run preliminary MaxEnt models. The contribution rate of each variable to these preliminary models was assessed. Any environmental variable with a contribution rate of zero was excluded from further consideration. For the remaining variables, a Pearson correlation analysis was conducted using R software to generate a correlation matrix. Based on the results of both the correlation analysis and the preliminary model contribution rates, a final set of variables was selected. Specifically, if two or more variables exhibited a high correlation (Pearson correlation coefficient |r| > 0.8), the variable with the lower contribution rate in the preliminary MaxEnt models was removed. This iterative process ensured that the final set of environmental variables used for constructing the definitive species distribution models was both ecologically relevant and minimally correlated.

Table 1 Environmental variables.
Full size table

Species distribution modeling and evaluation

Species distribution modeling was conducted using the biomod2 package (v4.2.5) within RStudio (v4.4.1). The modeling process integrated the 120 refined occurrence records of L. davidii with the selected environmental variables. Twelve individual species distribution modeling algorithms available within biomod2 were employed: Artificial Neural Network (ANN), Classification Tree Analysis (CTA), Flexible Discriminant Analysis (FDA), Generalized Additive Model (GAM), Generalized Boosted Model (GBM), Generalized Linear Model (GLM), Multivariate Adaptive Regression Spline (MARS), Maximum Entropy Model (MaxEnt), Random Forest (RF), and Surface Range Envelope (SRE).

For model calibration, the occurrence data were randomly partitioned into a training set (75% of records) and an independent testing set (25% of records) to evaluate model performance. Given that true absence data for L. davidii were unavailable, pseudo-absence points were generated. Two sets of pseudo-absence data were created, each containing 500 randomly selected points from the entire study area, excluding locations with known presence records22. This number of pseudo-absence points was chosen to provide a robust representation of the background environmental conditions across the study area, enhancing the ability of the model to discriminate between suitable and unsuitable habitats. While some approaches advocate for a larger number of pseudo-absence points (e.g., 5000–10000)23the use of 500 points per set in this study, coupled with multiple model runs and ensemble forecasting, was considered sufficient to characterize the environmental background for L. davidii within the biomod2 framework, particularly when balancing model complexity and computational resources.

Species distribution model evaluation

To account for variability arising from random initialization, each of the twelve individual modeling algorithms was run independently 5 times, resulting in a total of 50 individual model projections. The performance of each individual model run was evaluated using the True Skill Statistic (TSS) and the Area Under the Receiver Operating Characteristic Curve (AUC). Following the individual model runs, an ensemble model was constructed to synthesize the predictions and reduce uncertainty. Only individual model runs demonstrating good predictive performance, defined by a TSS score ≥ 0.75, were selected for inclusion in the ensemble. The algorithms that consistently met this criterion and were included in the final ensemble model were ANN, CTA, RF, GBM, FDA, GLM, MaxEnt, GAM, and MARS. The final ensemble model was then constructed by calculating a weighted average of the predictions from these selected high-performing individual models. The weights assigned to each selected model run were proportional to their respective AUC values, giving greater influence to models with higher discriminatory power. This weighted averaging approach aims to produce a more robust and reliable prediction of the potential suitable habitat for L. davidii.

The predictive performance of both individual models and the final ensemble model was assessed using several standard metrics22: AUC values range from 0 to 1, where a value of 0.5 indicates a model performing no better than random, and a value of 1 indicates perfect discrimination. AUC values were categorized as: < 0.6 (fail), 0.6–0.7 (poor), 0.7–0.8 (fair/moderate), 0.8–0.9 (good), and > 0.9 (excellent). True Skill Statistic (TSS): TSS values range from − 1 to + 1, where + 1 indicates perfect agreement and values of zero or less indicate performance no better than random. TSS scores > 0.75 were considered to indicate excellent model performance24. Kappa measures the agreement between predicted and observed presences/absences, corrected for agreement expected by chance. Values range from − 1 to + 1, with 1 indicating perfect agreement25.

Following the construction and evaluation of the ensemble model, the resulting habitat suitability prediction map for L. davidii was imported into ArcGIS software (v10.8) for further analysis and visualization. The continuous habitat suitability values, ranging from 0 (unsuitable) to 1 (highly suitable), were reclassified into four distinct categories to facilitate interpretation and management applications: Highly Suitable: Probability of presence ≥ 0.75; Moderately Suitable: 0.5 ≤ Probability of presence < 0.75; Poorly Suitable: 0.25 ≤ Probability of presence < 0.5; Unsuitable: Probability of presence < 0.25. This classification scheme allows for a clear delineation of areas with varying degrees of predicted suitability for L. davidii.

Results and discussion

Environmental variable selection

To mitigate the risk of model overfitting arising from potential correlations among environmental variables, a systematic screening and optimization process was implemented. From an initial set of 22 environmental variables, eight critical variables were identified as essential for modeling the suitable habitat of L. davidii following a rigorous selection procedure that involved eliminating variables with zero contribution and applying Pearson correlation analysis (Fig. 2). The selected key variables include elevation (Elev), slope, aspect, isothermality (bio3), temperature seasonality (bio4), minimum temperature of the coldest month (bio6), precipitation of the driest month (bio14), and precipitation seasonality (bio15).

Fig. 2
figure 2

Results of correlation analysis of environmental variables.

Full size image

Model performance and reliability

The performance of the individual modeling algorithms and the ensemble model (EM) was evaluated (Fig. 3). Among the individual models, the Random Forest (RF) model demonstrated the highest average accuracy, followed by Generalized Additive Models (GAMs) and Generalized Boosted Models (GBMs). Conversely, the Surface Range Envelope (SRE) model exhibited the lowest accuracy, with a TSS score of 0.561. Excluding the SRE model, the remaining nine individual models displayed excellent performance metrics, with an average TSS score exceeding 0.84 and an average AUC score greater than 0.93. The ensemble model (EM), constructed by combining these nine high-performing models, yielded robust and satisfactory results, achieving an AUC of 0.855 and a TSS of 0.957. Consequently, all subsequent analyses of habitat suitability were based on the outputs from this ensemble model.

Fig. 3
figure 3

Evaluation indices of individual predictive models for L. davidii distribution.

Full size image

Contribution of environmental variables and response curves

The ensemble model identified Temperature Seasonality (bio4), elevation (Elev), and the minimum temperature of the coldest month (bio6) as the most significant environmental factors influencing the geographic distribution of L. davidii. The response curves (Fig. 4) indicated specific optimal ranges and values for these variables. The optimal range for Temperature Seasonality (bio4) was determined to be between 519.65 and 937.41 (standard deviation *100), with the most suitable value around 725.89. For elevation (Elev), the favorable range was identified as 1586–2967 m, with a peak suitability around 2340 m. The suitable range for the minimum temperature of the coldest month (bio6) was found to be between − 17.78 °C and 1.90 °C, with an optimal value of approximately − 4.49 °C.

Fig. 4
figure 4

Environmental variable response curve.

Full size image

Current potential distribution of suitable habitats for L. davidii

The potential suitable habitats for L. davidii in China, as predicted by the ensemble model, were reclassified into four categories: highly suitable, moderately suitable, poorly suitable, and unsuitable areas (Fig. 5). The highly suitable area, covering approximately 35.79 × 105 km2, is primarily concentrated in Lanzhou City and its surrounding regions in Gansu Province, with extensions into parts of Shaanxi and Henan Provinces. These regions represent the core suitable zones for L. davidii. The moderately suitable area encompasses approximately 23.18 × 105 km2, and the poorly suitable area covers about 39.47 × 105 km2. These areas typically extend outward from the highly suitable zones. Notably, the model predictions also indicated potential suitable growing conditions for L. davidii in the Xinjiang region. The area of the primary distribution area provinces is shown in Table 2.

Fig. 5
figure 5

Potential geographic distribution of L. davidii based on the ensemble model at present.

Full size image
Table 2 The area of the primary distribution area provinces (×104 km2.
Full size table
Table 2 The area of the primary distribution area provinces (×105 km2.
Full size table

Discussion

This study successfully employed an ensemble species distribution modeling approach to predict the current potential suitable habitats for L. davidii in China and to identify the key environmental factors shaping its distribution. The high predictive accuracy of the ensemble model (AUC = 0.855, TSS = 0.957) underscores the robustness of the biomod2 framework for this type of ecological assessment, consistent with findings from other SDM studies on various plant species26.

Key environmental drivers of L. davidii distribution

The analysis revealed that temperature seasonality (bio4), elevation (Elev), and the minimum temperature of the coldest month (bio6) are the primary determinants of L. davidii distribution. The preference for moderate to high temperature seasonality (bio4, optimal range 519.65–937.41) suggests that L. davidii is adapted to environments with distinct seasonal temperature variations. Such seasonality is often crucial for the phenological development of temperate bulbous plants, influencing processes like dormancy breaking, sprouting, flowering, and bulb development27. Extreme temperature stability or, conversely, overly erratic and unpredictable fluctuations outside this optimal range could disrupt these critical life cycle stages, thereby limiting the persistence of species.

Elevation (optimal range 1586–2967 m, peak at 2340 m) emerged as a significant topographical factor. This preference for mid-to-high altitudes in regions like Gansu, Shaanxi, and Henan is consistent with the known distribution of many Lilium species, which often inhabit mountainous or subalpine environments28. Altitude influences local climate by creating gradients in temperature, precipitation, solar radiation, and atmospheric pressure. The identified optimal elevation range likely provides a specific combination of cooler temperatures, adequate moisture (potentially from orographic precipitation or reduced evapotranspiration), and possibly reduced competition compared to lower elevations. This aligns with the traditional cultivation areas of L. davidii, which are predominantly mountainous13.

The minimum temperature of the coldest month (bio6, optimal range − 17.78 °C to 1.90 °C, optimum at −4.49 °C) highlights the species’ cold tolerance and winter requirements. The ability to withstand temperatures as low as −17.78 °C indicates a significant adaptation to cold winter conditions, which is essential for perennial bulbs that overwinter in the ground29. The optimal temperature around − 4.49 °C suggests that while L. davidii is cold-hardy, it may also require a certain period of chilling for proper dormancy breaking and subsequent vigorous growth in spring, a common trait in many temperate geophytes30,31. Temperatures consistently above 1.90 °C during the coldest month might not provide sufficient chilling, while temperatures far below − 17.78 °C could exceed its physiological tolerance limits, leading to tissue damage or mortality.

Influence of other environmental variables

While temperature seasonality (bio4), elevation (Elev), and the minimum temperature of the coldest month (bio6) were identified as the primary drivers, other selected environmental variables, including slope, aspect, isothermality (bio3), precipitation of the driest month (bio14), and precipitation seasonality (bio15), also contribute to defining the suitable habitat for L. davidii.

Slope and aspect are crucial topographical features that modulate microclimatic conditions at a local scale32. For instance, aspect influences the amount and intensity of solar radiation received, which directly affects soil temperature, moisture evaporation, and light availability for photosynthesis. Slope gradient, on the other hand, impacts soil drainage, water runoff, and soil stability. For a bulbous plant like L. davidii, well-drained soils are often preferred to prevent bulb rot, and specific light conditions dictated by aspect might be optimal for its growth and flowering33. Although their individual contribution rates might be lower than the top three variables, their combined effect can significantly refine habitat suitability within broader climatically favorable zones.

Isothermality (bio3), which quantifies the extent of day-to-night temperature oscillations relative to the annual temperature range, likely plays a role in the physiological processes of L. davidii. A certain degree of diurnal temperature fluctuation can be beneficial for plant growth by influencing respiration rates and carbohydrate metabolism34. The specific optimal range for isothermality found in this study suggests that L. davidii prefers environments where these daily fluctuations are balanced within the context of overall annual temperature variations.

Precipitation patterns, particularly the precipitation of the driest month (bio14) and precipitation seasonality (bio15), are critical for water availability, especially for a species that relies on stored reserves in its bulb but still requires adequate moisture during its active growth and flowering periods35. While Lilium species are generally not adapted to waterlogged conditions, prolonged drought during critical developmental stages can severely impact growth, bulb replenishment, and reproductive success. Therefore, regions with a certain minimum level of precipitation during the driest part of the year and a degree of predictability in seasonal moisture availability are likely to be more suitable.

Model performance and predicted habitat distribution

The ensemble model demonstrated high predictive accuracy, and the predicted current potential distribution of L. davidii (Fig. 5) largely aligns with its known core cultivation and natural distribution areas, particularly in Gansu, Shaanxi, and Henan provinces17. This congruence between model predictions and empirical observations lends credibility to the model’s output and its utility for conservation and management planning. The concentration of highly suitable areas around Lanzhou City in Gansu Province is consistent with its designation as a National Geographical Indication product, reflecting the region’s historically favorable environmental conditions for this specific lily variety36.

The model also identified moderately and marginally suitable areas extending from these core zones. These areas may represent transitional zones where one or more environmental factors are sub-optimal but still within the tolerance limits of L. davidii. Such regions could be important for future range expansion or for identifying populations with unique adaptations, warranting further investigation.

Implications for conservation and cultivation management of L. davidii

A particularly noteworthy finding is the prediction of potentially suitable habitats for L. davidii in parts of the Xinjiang region. This is significant as Xinjiang is generally outside the traditionally recognized distribution range of this specific variety. This prediction opens new avenues for research and potential agricultural applications. The unique climatic conditions in Xinjiang, such as high diurnal temperature ranges, abundant sunshine during the growing season, and potentially different soil compositions, could offer distinct advantages or challenges for L. davidii cultivation37. For example, ample sunlight and significant temperature differences between day and night might promote bulb carbohydrate accumulation and potentially influence the concentration of medicinal compounds. However, expanding cultivation to Xinjiang also presents several potential risks and challenges that require careful consideration. These include: Xinjiang is predominantly an arid and semi-arid region, and water resources are a critical limiting factor for agriculture. Large-scale introduction of L. davidii would need to consider sustainable water management practices and the potential impact on local water balances38. While the model considered broad climatic and topographical factors, specific soil properties (pH, texture, nutrient content) in Xinjiang’s potential suitable areas need to be assessed for their compatibility with the requirements of L. davidii. Therefore, while the model provides a valuable first-pass assessment, extensive field trials, ecological adaptability studies, and thorough risk assessments are crucial before any large-scale introduction or cultivation efforts are undertaken in Xinjiang.

Beyond Xinjiang, the identification of extensive highly and moderately suitable areas in provinces such as Gansu, Shaanxi, and Henan carries significant implications for the species’ conservation and sustainable utilization. For existing cultivation regions within these provinces, our models can guide optimized land-use planning by pinpointing areas that are environmentally most congruent with L. davidii’s ecological requirements. This allows for more targeted and efficient resource allocation in agricultural management. For conservation efforts, areas predicted to remain highly suitable, especially those within or adjacent to existing protected areas, can be prioritized for in situ conservation strategies, safeguarding genetic diversity and natural populations.

Furthermore, for areas identified as suitable but currently underutilized for L. davidii cultivation, our findings suggest potential for agricultural expansion. However, any such expansion should be approached with caution and scientific rigor. This includes conducting targeted field surveys to verify predicted suitability at finer spatial scales, implementing pilot cultivation projects to assess local adaptability and yield, and ensuring that new cultivation initiatives adhere to principles of sustainable agriculture, particularly regarding soil health and water resource management. The integration of our habitat suitability maps with administrative boundaries and land-use classifications can serve as a powerful tool for policymakers and local communities to make informed decisions regarding L. davidii conservation prioritization and cultivation management.

Limitations of the study

While this study provides valuable insights, several limitations should be acknowledged. Firstly, the species distribution models were based primarily on climatic and topographical variables. Although these factors are fundamental drivers of species distribution, the explicit inclusion of edaphic (soil) variables, such as soil type, pH, organic matter content, and nutrient availability, was constrained by the lack of readily available, high-resolution, and spatially continuous soil data covering the entire study extent. These edaphic factors are known to significantly influence plant growth and distribution, particularly for species with specific soil requirements like Lilium. Incorporating such detailed soil information in future studies could potentially refine the predictions of suitable habitats and provide a more comprehensive understanding of the ecological niche of L. davidii. Secondly, the occurrence data, while compiled from multiple sources and subjected to cleaning, may still be subject to some degree of sampling bias. Certain accessible areas might be overrepresented, while remote or less explored regions could be underrepresented. Although spatial thinning was applied to reduce clustering, inherent biases in the original data collection efforts can influence model outcomes. Thirdly, this study primarily focused on modeling the current potential distribution of L. davidii. While these findings establish a crucial baseline for conservation and management, they do not account for future environmental dynamics. Future research should, therefore, incorporate projections under various climate change scenarios to assess how the suitable habitats of L. davidii might dynamically shift, expand, or contract in the future. This is particularly important for a species with economic and cultural significance, as long-term climate change could pose significant threats to its viability in traditional cultivation areas. Finally, and critically, this study, like many others employing SDMs, models a single species in isolation, which presents specific limitations. SDMs predict potential suitable habitats based on observed environmental correlations. However, they inherently do not account for complex ecological processes beyond abiotic environmental drivers. This includes biotic interactions and dispersal limitations. The exclusion of these factors can lead to an overestimation of the species’ potential fundamental niche, potentially depicting a broader suitable habitat than its actual realized niche in nature. Furthermore, as correlative models, SDMs provide a static representation of habitat suitability and are inherently limited in their ability to capture dynamic niche processes, such as species adaptation, evolutionary responses, or shifting biotic interactions over time, especially under changing environmental conditions. Future research should strive to integrate these complex biotic and dispersal factors, perhaps through more mechanistic modeling approaches or by linking SDMs with dynamic population models, to provide a more comprehensive understanding of species distribution patterns.

Conclusion

This study successfully predicted the current potential ecological distribution of L. davidii in China with high predictive accuracy using an ensemble species distribution modeling approach. Key environmental factors, notably temperature seasonality, elevation, and minimum temperature of the coldest month, were identified as significantly shaping the distribution of L. davidii, with delineated optimal ranges offering critical insights into its ecological niche. The model confirmed highly suitable habitats in traditional regions like Gansu, Shaanxi, and Henan, and importantly, identified potential suitability in Xinjiang, suggesting new research avenues. These findings provide a robust scientific basis for targeted conservation and sustainable utilization strategies. Given the environmental sensitivities of species, proactive measures are essential, including enhanced in situ and ex situ conservation, promotion of climate-adaptive cultivation practices, dynamic habitat monitoring, and explicit modeling of future climate change impacts to anticipate range shifts. Careful evaluation of new potential cultivation areas, alongside policy integration and stakeholder collaboration, will be crucial for adaptive management. Future research should prioritize field validation, comprehensive ecological adaptability assessments, incorporation of edaphic and biotic factors into models, and projections of future distributions under climate change to ensure the long-term persistence and sustainable use of this culturally significant lily.