C. heterophylla occurrence records across China were compiled from authoritative sources: the Global Biodiversity Information Facility (GBIF; https://www.gbif.org/), Chinese Virtual Herbarium (CVH; https://www.cvh.ac.cn/), Plant Photo Bank of China (PPBC; https://ppbc.iplant.cn/), and published literature (He et al., 2023; Liu and Chen, 2024). Records lacking precise coordinates were georeferenced using Baidu Map Coordinate Picker (https://api.map.baidu.com/lbsapi/getpoint/index.html) based on textual descriptions, and corresponding uncertainties were estimated according to the specificity of each location description. To mitigate the effects of spatial autocorrelation and sampling bias inherent in herbarium and literature-based data (Taylor et al., 2020), the dataset was rigorously cleaned by removing duplicate and ambiguous records. Spatial thinning was then performed using the R package spThin (Allouche et al., 2006), with a filtering distance of 5 km to ensure that only one occurrence record was retained per 5 ×5 km grid cell (Wang et al., 2024). The final dataset comprised 243 spatially independent occurrence records (Figure 1), which were formatted into a CSV file for input into the MaxEnt model.

Thirty-one environmental variables potentially influencing the distribution of C. heterophylla were initially selected, comprising 19 bioclimatic variables (under CMIP6 SSPs scenarios) (Urban et al., 2016), 9 soil parameters, and 3 topographic factors. In this study, soil parameters and topographic factors are generally considered to remain constant under future climate projections. To reveal its core climatic niche, the model excludes disturbance factors such as human activities, thereby simulating potential suitable distribution areas under future climate change (Ren et al., 2025; Shan et al., 2025). 1970–2000 climate data and future climate projections for the 2050s (2041-2060) and 2090s (2081-2100) were obtained from WorldClim (version 2.1; http://www.worldclim.org) at 2.5 arc-minute resolution. Future climate projections were derived from the BCC-CSM2-MR model developed by the China Meteorological Administration (CMA), considering three Shared Socioeconomic Pathways (SSPs): SSP1-2.6 (low radiative forcing), SSP2-4.5 (intermediate), and SSP5-8.5 (high) (Wang et al., 2024). Soil data came from the Harmonized World Soil Database (HWSD v1.2; Cold and Arid Regions Scientific Data Center), while topographic data were obtained from Geospatial Data Cloud (http://www.gscloud.cn). All topographic and soil datasets were resampled to a 2.5 arc-minute resolution using ArcGIS 10.8 to ensure consistency with the climatic data layers.

To mitigate model overfitting caused by multicollinearity among environmental variables (Wang et al., 2020), we performed a structured variable selection procedure. Initial MaxEnt runs were conducted to evaluate variable importance based on percent contribution and permutation importance. A variable was excluded only if both of these values were exactly 0% and its response curve in the jackknife test fully overlapped with the curve generated using all variables. Environmental factor correlations were analyzed using IBM SPSS Statistics (version 28.0; IBM Corp., Armonk, NY, USA). By combining the contribution rates of variables with correlation matrix analysis, factors exhibiting pairwise correlation with the coefficient |r| > 0.8 underwent selection refinement: the lower-contribution variable in each correlated pair (determined from preliminary analysis) was eliminated (Yi et al., 2016). Further consideration was given to thexbiological plausibility of each variable, evaluating its association with the ecological requirements of C. heterophylla and substantiating findings through ecological theory. For instance, the selected variables are closely linked to the species’ water requirements and temperature adaptability, exhibiting not only statistical significance but also clear biological plausibility. This process yielded nine key environmental factors for final model construction (Table 1).

MaxEnt v3.4.4 was used to predict the suitable habitats for C. heterophylla across three time periods. The model incorporated 243 occurrence points with 25% randomly allocated for testing and 10 replicate runs. Outputs were formatted as logistic probabilities (0 to 1). Environmental variable importance was assessed through jackknife testing, with key factors identified by joint consideration of percent contribution and permutation importance. Model accuracy was evaluated using the Area under the Receiver Operating Characteristic (ROC) Curve (AUC) (Phillips et al., 2006). AUC values (0–1 scale) indicate predictive accuracy: <0.6 (failed), 0.6-0.7 (poor), 0.7-0.8 (fair), 0.8-0.9 (good), and 0.9-1.0 (excellent) (Zhang et al., 2024).

Default MaxEnt parameters may produce overly complex models prone to overfitting (Morales et al., 2017). In this study, the regularization multiplier (RM) and feature combination (FC) in the MaxEnt model were optimized using R’s Kuenm package (Cobos et al., 2019). From baseline settings (RM = 1, FC=LQHPT), RM was tested across 0.5-4.0 (0.5 increments) with randomized combinations of five feature classes: linear, quadratic, product, threshold, and hinge. C. heterophylla occurrence data were randomly partitioned (75% for training, 25% for testing). Model complexity was tested based on the evaluated using δAICc values and 5% missing rates, where lower δAICc indicates superior predictive performance (Giraldo and Kucuker, 2024).

MaxEnt output files (ASC format) were batch-converted to raster format in ArcGIS. Predicted suitability for C. heterophylla was reclassified into four categories based on mean logistic values: unsuitable (0-0.2), low suitability (0.2-0.4), medium suitability (0.4-0.6), and high suitability (0.6-1) (Gu et al., 2021; Xu et al., 2019). Distribution map for 1970-2000, 2050s and 2090s periods were generated using a China basemap. Suitable habitat areas were quantified per period using ArcGIS Spatial Analyst tools.

Based on the potential habitat distribution maps for 1970–2000 and future periods generated by the MaxEnt model, we quantitatively analyzed spatial pattern changes using the Distribution Change Analysis module in SDM toolbox (v2.0). Habitat transitions were classified into four categories (-1: range expansion; 0: species absence; 1: stable habitat; 2: range contraction). Spatial change maps were generated in ArcGIS 10.8. Centroid positions for 1970–2000 and future suitable habitats were calculated using the Centroid Changes (Liners) tool in SDM toolbox (v2.0), with migration distances indicating directional shifts in habitat distribution.

The MaxEnt model simulated C. heterophylla potential habitats using 243 occurrence records and nine environmental factors. Default parameters (RM = 1, FC=LQPTH) yielded δAICc=88.48, whereas optimized parameters (RM = 3, FC=LQPH) achieved δAICc=0, indicating superior model performance (Table 2). Ten cross-validations with optimized settings produced a mean AUC of 0.933, confirming high predictive accuracy for habitat suitability (Figure 2).

Analysis of regularization training gain, permutation importance, and jackknife tests from the MaxEnt model simulation identified key environmental factors influencing C. heterophylla distribution. Bio16 (Precipitation of wettest quarter) showed highest contribution (39.5%), followed by Bio9 (Mean temperature of driest quarter, 22.2%), Alt (Altitude, 16.2%), and Bio3 (Isothermality, 7.1%), collectively accounting for 85% of total contribution. Permutation importance highlighted Bio16 (41.1%) and Alt (27.6%) as most significant, totaling 68.7% (Table 3; Figure 3). In contrast, the remaining variables—REF Depth (4.0%), Bio19 (3.9%), T_bs (3.5%), Slope (2.3%), and Bio4 (1.2%)made markedly lower contributions, collectively accounting for only 14.9%. Results indicate that precipitation during Bio16 (Precipitation of wettest quarter) constitutes the dominant factor in the model. Alt (Altitude) is identified as a secondary yet critical determinant, while temperature-related variables (Bio9, Bio3) further refine habitat suitability. Together, these form the core ecological drivers for C. heterophylla.

C. heterophylla occurrence probability followed unimodal responses to environmental gradients (Figure 4). Using probability > 0.5 as threshold (Araújo et al., 2019) the optimal range for Bio16 (Precipitation of the Wettest Quarter) was found to be between 300 mm and 450 mm, with the highest probability (approximately probability = 0.6) occurring at 350 mm. The suitable range for Bio9 (The mean temperature of the driest quarter) was -12°C to 2°C, peaking at -2°C (approximately probability = 0.56). The optimal Alt (Altitude) range was 100 to 700 m, with a peak at 300 m (approximately probability = 0.63). For Bio3 (Isothermality), the suitable range was between 24 and 30, peaking at 28 (approximately probability = 0.57).

Future habitat suitability for C. heterophylla shows substantial overlap (69.20%~77.56% retention) with 1970–2000 suitable areas across all scenarios, mainly concentrated in Heilongjiang, Jilin, Liaoning, Beijing, Tianjin, Hebei, Shanxi, Shandong. Over all, the rate of habitat expansion ranges from 18.83% to 37.78%, with newly suitable areas mainly emerging in northern and high-altitude marginal regions. Conversely, range contraction occurs predominantly along the southern margins of the historical distribution. The highest increase is observed under the 2090s SSP5-8.5 scenario, while the smallest increase is observed under the 2050s SSP1-2.6 scenario. Suitable areas are expected to expand mainly in the northern regions of Heilongjiang, Inner Mongolia, northern Qinghai, Gansu, and northern Ningxia. The rate of loss is 22.44% ~ 30.70%, with the highest loss under the 2090s SSP2-4.5 scenario, and the lowest loss under the 2050s SSP1-2.6 scenario. The suitable area for C. heterophylla is anticipated to contract mainly in southern Hubei, southern Hunan, southern Henan, eastern Sichuan, and southern Guizhou (Table 5; Figure 7).

Northern Heilongjiang, Inner Mongolia, northern Qinghai, Gansu, northern Ningxia, southern Hubei, southern Hunan, southern Henan, eastern Sichuan, and southern Guizhou represent core change zones for future habitat suitability shifts in China.

C. heterophylla distribution centroids exhibited consistent northward shifts across all scenarios. During the 1970–2000 period, the centroid was located near Dazhangjia Village, Longyao County, Xingtai City, Hebei Province (114.849°E, 39.409°N). Under the SSP1-2.6 scenario, the centroid is projected to shift to the vicinity of Zhoukoudian, Fangshan District, Beijing (115.94°E, 39.66°N) by the 2050s, and further toward Fengtai District, Beijing (116.402°E, 39.857°N) by the 2090s. Under the SSP2-4.5 scenario, it is expected to migrate to the area around Miyun District, Beijing (116.911°E, 40.598°N) by the 2050s, and eventually reach Huangtukang Township, Weichang Manchu and Mongolian Autonomous County, Chengde City, Hebei Province (117.697°E, 41.764°N) by the 2090s. Under the SSP5-8.5 scenario, the centroid will move to Datou Mountain Township, Weichang Manchu and Mongolian Autonomous County (117.269°E, 41.858°N) by the 2050s, and finally arrive at Tianqiao Town, Fengning Manchu Autonomous County, Chengde City (117.185°E, 41.014°N) by the 2090s.

The results from the environmental variables replacement contribution analysis and jackknife test indicate that Bio16 (Precipitation of wettest quarter, 39.5%), Bio9 (Mean temperature of driest quarter, 22.2%), Alt (Altitude, 16.2%), and Bio3 (Isothermality, 7.1%) represent primary environmental factors governing the distribution of C. heterophylla (Table 3; Figure 3). However, discrepancies exist regarding the identification of primary environmental factors across different studies. Liu and Chen (2024) reported that the Bio6 (Minimum temperature of coldest month) and human footprint were the main drivers for hazelnut distribution. This divergence may arise from their analysis of multiple C. heterophylla species collectively rather than species-specific responses. The ecological adaptability of different Corylus species varies significantly, and pooled analyses may fail to capture their species-specific environmental requirements. Similarly, He et al. (2023) identified Bio4 (Temperature seasonality), Bio16 (Precipitation of wettest quarter), and Bio15 (Precipitation seasonality) as dominant factors, potentially because their study did not account for topographic influences. Topographic parameters, including altitude, slope, and aspect, significantly influence plant distribution patterns, particularly in regions with complex geomorphology, by mediating habitat suitability through their effects on localized climatic conditions and edaphic properties (Huang et al., 2024). Consequently, the integration of multidimensional environmental variables is critical for robust identification of the mechanistic drivers governing phytogeographic distributions.

Bio16 showed an optimal range of 300–450 mm, with the highest probability of occurrence peaking at 350 mm (Figure 4), reflects the species’ sensitivity to water supply during the critical summer fruit development stage. In many temperate regions, peak precipitation coincides with summer (Fick and Hijmans, 2017), providing essential soil moisture for nut filling and oil synthesis. Moderate rainfall ensures high yields and quality, while excessive rainfall leading to waterlogging can cause root asphyxiation and increase disease incidence (Marsal et al., 1997). Bio9 exhibited a suitable range of -12°C to 2°C, indicating strong adaptation to cold and dry winters. This physiological trait enables C. heterophylla to survive extreme winter temperatures as low as -30°C to -40°C (Erdogan and Mehlenbacher, 2000). Regarding Alt (Altitude), this study identified a constrained distribution range of 100–700 m, peaking at 300 m. Wang et al. (2020) reported a broader suitable altitude range of 0–3000 m for C. mandshurica. This divergence suggests a clear altitudinal niche partitioning between the two species, which may reduce direct competition and facilitate their coexistence in sympatric regions. Huo et al. (2016) demonstrated that the northern populations of C. heterophylla and C. mandshurica exhibit high climatic similarity, though C. mandshurica generally occurs at higher elevations than C. heterophylla in sympatric regions. This finding is in line with the results of this study. Bio3 has an optimal range of 24 to 30. As the ratio of Bio2 (Mean diurnal temperature range) to Bio7 (Temperature annual range), this variable ecologically reflects the relative magnitude of diurnal versus seasonal temperature variation. The identified optimum indicates that the species thrives in regions with pronounced annual temperature seasonality. This climatic preference aligns with the physiological traits of the C. heterophylla as a temperate deciduous shrub, particularly its dependence on cold winters (Fadón et al., 2020). Sufficient low-temperature accumulation is crucial for meeting the “chilling requirement” necessary to break dormancy, thereby ensuring synchronized spring bud break and normal inflorescence development (González De Andrés et al., 2023). Conversely, in regions with Bio3 values significantly exceeding 30 (e.g., maritime climates), elevated winter temperatures may fail to meet cold requirements, triggering dormancy disorders that impair normal growth and development (Luedeling et al., 2011). Thus, the distribution and adaptation of C. heterophylla depend not only on growing-season temperature conditions but critically on temperature patterns exhibiting pronounced seasonal rhythms to synchronize its phenological processes with the external environment.

The analysis of environmental variables and their impact on the potential distribution of C. heterophylla under 1970–2000 and future climate scenarios provides significant insights into the species’ ecological adaptability and future prospects. Using the optimized MaxEnt model, we predict the species’ geographic distribution under different time periods. Under 1970–2000 climatic conditions, C. heterophylla is primarily distributed in Northeast China (Changbai Mountains and Xiao Xing’anling ranges), North China (Yanshan Mountains), and Northwest China (northern slopes of the Qinling Mountains) (Figure 5). The total suitable area covers 210.85×10⁴ km² (21.96% of China’s land area), with the highly suitable area accounting for 39.09×10⁴ km² and being concentrated in Jilin, Liaoning, Hebei, Shanxi provinces (Table 4). Future climate scenario simulations consistently project a contraction in the total suitable habitat range of C. heterophylla. This trend aligns with earlier studies reporting accelerated range loss in its southern low-elevation populations (Yang et al., 2022). Model results further suggest that although low-suitability areas will gradually decrease, medium- and high-suitability zones are expected to expand in a spatially heterogeneous manner across climate scenarios, particularly in regions such as northern Heilongjiang, northern Inner Mongolia, and northwestern provinces including Gansu and Ningxia (Figure 7). This latitudinal and altitudinal expansion is driven by combined effects of climate warming and projected shifts in precipitation patterns. Altered thermal and moisture conditions collectively enable previously cold-limited northern and high-elevation areas to gradually meet the eco-physiological requirements of C. heterophylla (Álvarez-Álvarez et al., 2025). These newly expanded areas represent potential future habitats for the C. heterophylla, holding critical conservation value for maintaining its population persistence and mitigating biodiversity loss caused by the species’ southern distribution decline.

Under future climate change scenarios, the suitable distribution range of C. heterophylla exhibits a bipolar migration pattern characterized by “northward expansion and southward contraction” (Figure 7). Newly suitable habitats are predominantly distributed in northern Heilongjiang, northern Inner Mongolia, Gansu, Qinghai, and other northwestern regions, whereas substantial habitat loss is observed in southern China, including Hubei, southern Henan, and Sichuan. Compared to the SSP1-2.6 and SSP2-4.5, SSP5-8.5 scenario shows the most significant changes in suitable areas. This can be attributed to the substantial increases in both precipitation and temperature across northern China under this high-emission pathway (Zhou et al., 2020), which collectively create thermal hygric conditions that more closely match the species’ requirements during critical growth phases (Fick and Hijmans, 2017). In contrast, under the SSP2-4.5 and SSP1-2.6 scenarios, the more moderate warming and limited precipitation increases are insufficient to generate equivalent habitat gains in the north. Consequently, the net change in these scenarios is dominated by habitat losses in the southern parts of the range, resulting in a greater overall reduction in total suitable area. CMIP6-based projections of future global temperatures reveal persistent warming trends across all emission scenarios, albeit with significant variations in magnitude. Zhao et al. (2021) further estimate a global temperature increase ranging from 0.6 to 7.8 °C, which may exacerbate the frequency and intensity of extreme climate events. Additionally, CMIP6 models project increased precipitation across most regions of China, with the most substantial enhancement occurring under the high-emission scenario (SSP5-8.5) (Álvarez-Álvarez et al., 2025). The northward migration of suitable habitats can be attributed to synergistic improvements in hydrothermal conditions in northern regions, which optimize the growth environment for C. heterophylla. Conversely, in the southern parts of the current distribution range, future projections indicate that precipitation may exceed the species identified optimal upper threshold of 450 mm (Figure 4; Xu et al., 2022). Such excessive rainfall could result in periodic soil waterlogging, elevating the risk of root hypoxia—a condition known to be detrimental to C. heterophylla (Bailey-Serres and Voesenek, 2008).

The distribution centroid of C. heterophylla demonstrates a consistent northward migration trend under all future climate scenarios relative to its historical position (1970–2000) in Hebei Province (Figure 8). This migratory trend is most pronounced under the higher-emission scenarios (SSP2-4.5 and SSP5-8.5) and is strongly linked to projected changes in precipitation patterns. Climate models indicate a general increase in growing-season precipitation across northern China (Zhou et al., 2020). This enhanced water supply may transform previously water-limited areas in the north into new suitable habitats for C. heterophylla. Under high-emission scenarios, while traditional distribution areas like North China may experience intensified seasonal droughts and precipitation uncertainty, some more northern regions could benefit from modest precipitation increase or greater climatic stability (Zhang et al., 2021). This shift in precipitation patterns favors a contraction of suitable habitat toward these more favorable high-latitude zones. Conversely, in the southern parts of the current range, increased precipitation is projected to frequently exceed the species’ optimal upper threshold (450 mm; Figure 4), leading to waterlogging stress and habitat degradation. Supporting this, Wang et al. (2019) documented upward shifts in suitable ranges for most plant species in Changbai Mountain in response to rising temperatures and altered precipitation patterns, with significantly greater migration rates and distances under high-emission scenarios compared to low-emission ones. These findings collectively suggest that enhanced warming under high-emission scenarios likely drives more substantial displacements of C. heterophylla’s distribution center.