Keys to the global treeline formation: Thermal limit for its position and moisture for the taxon-specific variation

Significance

Alpine treelines delineate spatial boundaries of tree life forms. Different physiological limits among tree taxa could influence the formation of treelines and their response to environmental changes. However, taxon-specific variability and geographical patterns are overlooked by global-scale studies. We integrate over 2,000 records of treeline species from 39 mountain regions globally and uncover a dual control mechanism underlying treeline formation: Heat deficits, averaging around 35% below genus- and species-level thermal optima, consistently restrict the presence of treeline; moisture gradients drive the taxonomic variation among treeline distribution patterns. This dual-control mechanism explains why treelines, despite sharing thermal constraints, exhibit notable biogeographic divergence. It also provides a theoretical basis for more accurate predictions of future treeline dynamics under climate change.

Abstract

Alpine treeline is a prominent biogeographic feature worldwide, determined by the physiological limit of tree life form. There are considerable variations in the various dimensions of physiological limit among tree taxa; thus, varied environmental drivers and spatial patterns are expected for different tree taxa at treelines. However, such taxonomic variability of treeline is often overlooked in large-scale studies. Here, we assembled, to our knowledge, the most comprehensive dataset of tree species at alpine treelines, drawing from research conducted over the past half-century, encompassing over 2,000 records across 38 mountain regions and 43 countries. Using this extensive global dataset, we examined the spatial patterns and environmental drivers shaping different tree taxa at treelines worldwide. The highest tree richness at treelines was found in mid-latitude mountains of the Northern Hemisphere, reflecting floristic differentiation caused by continental isolation. Moisture and climatic variability, particularly seasonal fluctuations, determine the turnover of tree taxa at treelines. Heat limitations appear to restrict the establishment of all genera, effectively defining treeline positions. Heat conditions at treeline positions tend to be about 35% below the genus- and species- level thermal optima. This thermal threshold can effectively explain the global pattern of uppermost tree elevation. Our findings highlight the synergic effects between heat and moisture in determining the taxonomic variation in treeline formation, offering insights for alpine treeline studies under climate change.

Alpine treelines mark the upper elevation limits of tree existence (1) and serve as key indicators of ecosystem responses to climate change (2–4). The treeline presence, representing the abrupt absence of tree life forms [upright woody plants that reach a minimum height of 2 m (5, 6)], is shaped by the adaptation of trees to harsh alpine conditions (6, 7). Theoretically, the treeline reflects the potential physiological limit of trees, defining the edge of their fundamental niche (1). Based on the growth limitation hypothesis, Körner (8) proposed that low temperatures are the primary factor constraining the physiological activities of trees in high mountain regions (5). Root zone temperature at treeline sites (9), combined with the optimal fitting of air temperatures across approximately 300 treeline locations worldwide (10), has been used as a theoretical threshold for predicting potential treeline elevation. This threshold assumes treelines occur in regions with a growing season air temperature of approximately 6.4 ± 0.7 °C, and has been widely adopted (11–13). However, this concept focuses on the niche of tree life forms rather than specific taxa, the application of the niche concept inevitably involves the consideration of different dimensions of the niche (e.g., physiological limits) and their variation among different taxa (14). In fact, beyond insufficient heat, factors such as drought, extreme climate events, and poor soil conditions can also limit trees in mountain regions (11, 15–17). As different taxonomic groups exhibit distinct adaptive traits and tolerance, there are considerable variations in the various dimensions of physiological limit among tree taxa (18). For example, at the Mt. Everest treeline in the Central Himalaya, differences in water-use traits cause Abies spectabilis to depend more on summer temperature, while Betula utilis mainly relies on spring precipitation (19, 20). This complicates the idea of a single climatic factor or a universal threshold for predicting treeline positions (10). Since determining an environmental threshold for treeline presence is important for predicting alpine ecosystem dynamics under climate change (11), accounting for taxonomic variation in treeline environmental responses is necessary, yet this remains largely unaddressed in global-scale studies.

The environmental threshold determining treeline presence represents a “ceiling” for the physiological activities of different tree taxa (8), meanwhile, there may also be “filtering” factors that control which taxa occupy specific treeline locations, thereby shaping the current treeline taxa pattern. However, few studies have focused on large-scale pattern of different tree taxa at the treeline, primarily due to the limited spatial extent of available treeline species data (21). Most existing research on treeline taxa was at local scales, and they often treated different tree taxa as part of the community composition rather than the primary subjects (18, 22–25). Overall, ignoring taxonomic differences at the treeline leads to an incomplete understanding of the ecological processes driving treeline formation (26, 27), as well as how species compete and distribute under the resource constraints of high-elevation environments (28, 29), limiting our ability to forecast the dynamics of treeline under climate change. Therefore, it is critical to provide a panoramic description of the actual distribution and variability of the treeline taxa in geographical and environmental space.

We compiled, to our knowledge, the most comprehensive dataset of global Tree Species on Alpine Treeline (TSAT) [see Xie et al. (30) or Dataset S1]. The TSAT dataset included 2,235 records of tree species on treeline positions from 739 studies since 1980 (Fig. 1 A–D and SI Appendix, Figs. S1–S4 and Method S1), providing a much-expanded spatial extent of treeline data. We focus on two key questions based on the TSAT: 1) What environmental factors drive variation of the distribution patterns among treeline taxa? 2) What are the environmental drivers and corresponding thresholds that determine the treeline presence of different taxa and the subsequent absence of tree forms at global scale? In addition to temperature, factors such as moisture and soil conditions are also known to limit tree growth at treeline (11, 15–17). Their effects on treeline formation have not been comprehensively tested in large-scale studies. Therefore, we incorporated a wide range of environmental factors in our analysis, including the overall regime and variability of temperature and precipitation, as well as various soil properties. We first performed dimensionality reduction on 43 key climate and soil environmental factors (SI Appendix, Table S1) using principal component analysis (PCA), and used phylogenetic methods and decision trees to identify the main environmental factors shaping regional differences in treeline taxa. Then, we applied a bivariate model to identify the main environmental drivers of transition between tree form presence and absence at treelines with different taxa. We developed a Relative Distance to Optimum (RDO) index, to measure the environmental pressure on plant taxa beyond their suitable ecological niche. Based on RDO we proposed an environmental threshold for treeline presence, depicting the uppermost tree limit considering taxon-specific characteristics. It illustrates how much a given environmental factor deviates from a taxon’s optimal requirement. Our analysis was conducted mainly at the genus level, which can be less prone to inaccurate information at the species level (e.g., species misidentification and inconsistent naming) (31, 32). In addition, tree species from the same genus often share similar traits and ecological strategies (33, 34). These reasons make genus a suitable level of study for analyzing tree–environment relationships at a global scale. We also performed analysis at the species level based on common treeline species as an additional validation of genus-level results. We specifically focused on the top 10 common treeline genera (SI Appendix, Fig. S5) when conducting quantitative analysis, covering 2,133 of the 2,235 total sampled points, to improve robustness.

Fig. 1.

Overview of the Tree Species on Alpine Treeline (TSAT) database and the global spatial pattern of tree taxa at treelines. (A) Map of the global distribution of treeline plants at the genus level. For species-level maps, see SI Appendix, Figs. S1–S4. (B–D) Three hotspots of high tree genus diversity at the treeline (Rocky Mountains, Alps, and Tibetan Plateau). (E) taxonomic richness and mean elevation (m a.s.l.) across 5° latitude bands. The colored points represent the number of treeline tree families (orange), genera (blue), or species (red). The gray points represent the average treeline elevation involved in the TSAT dataset for each 5° latitude band. The error bars represent the SD for treeline elevation.

Results and Discussion

Spatial Pattern and Diversity of Global Alpine Treeline Trees.

The TSAT database included a total of 114 tree species, belonging to 22 genera and 10 families distributed across the world and forming alpine treelines (Fig. 1 A–D and SI Appendix, Figs. S1–S4). Trees belonging to the family Pinaceae are the most successful at treeline in the Northern Hemisphere, with four genera (Pinus, Picea, Abies, and Larix) accounting for 76% of all the records. Among them, Pinus was observed in the majority of samples, constituting 35% of the Pinaceae treelines, and includes the greatest number of species, mainly concentrated in North America and Europe (Fig. 1A and SI Appendix, Fig. S1). Picea treelines extend to the northernmost latitudes and dominate in Alaska (Fig. 1A and SI Appendix, Fig. S1). Abies treelines are concentrated in the southeastern Tibetan Plateau and the Rocky Mountains (Fig. 1A and SI Appendix, Fig. S1), and Larix treelines are mainly distributed in the mid- and high-latitude regions of Eurasia (Fig. 1A and SI Appendix, Fig. S1). Except for the Pinaceae treeline, Juniperus treelines concentrate around Qinghai–Tibetan Plateau (Fig. 1A and SI Appendix, Fig. S2). Betula treelines are the most widespread broadleaf treelines in the Northern Hemisphere (Fig. 1A and SI Appendix, Fig. S1). In the Southern Hemisphere, all treeline trees are broad-leaved (Fig. 1A and SI Appendix, Fig. S5). Polylepis and Nothofagus treelines, which appear in the Andes and New Zealand, as well as tropical and temperate regions, respectively, have the widest distributions (Fig. 1A and SI Appendix, Figs. S1 and S2). Eucalyptus and Erica treelines dominate Australia and eastern Africa, respectively (Fig. 1A and SI Appendix, Fig. S2). Additionally, treelines of 12 other genera, such as Sorbus, Fagus, and Tsuga, have been sparsely recorded worldwide (SI Appendix, Figs. S2–S4).

The summary of richness patterns (SI Appendix, Table S2 and Fig. S5) and latitudinal variation (Fig. 1E) reveal the global pattern of treeline taxon diversity. A more complex composition of treeline taxa is observed in the mid-latitudes of the Northern Hemisphere compared to other regions (Fig. 1E). More than 10 treeline species and five genera have been recorded in hotspots such as the Hengduan Mountains (Qinghai–Tibetan Plateau), the Rockies, and the Alps (Fig. 1 B–D and SI Appendix, Table S2). Within these regions, no single genus accounts for more than 50% of the records, and at least three genera comprise over 10% of the observed treelines (SI Appendix, Table S2). Conversely, in other regions, although 10 treeline species are found in the northern Andes, all belong to the same genus (Polylepis) (SI Appendix, Table S2). In Alaska, despite the presence of treelines from six different genera, approximately 94% of the records originate from a single genus (Picea) (SI Appendix, Table S2).

The diversity and data density of treeline taxa in the mid-to-high latitudes of the Northern Hemisphere are both much higher than in the tropics and the Southern Hemisphere. Sampling bias may affect the diversity patterns, especially at species-level; however, the unequal sample size may reflect the true biogeographic distribution of treelines. The ratio of treeline samples between the Northern and Southern Hemispheres in our dataset is approximately 9:1 (2014: 221), and is similar as that in other global treeline datasets (9, 10, 21, 35). In particular, Zou et al. (36) studied the global distribution of treelines with 10 m-resolution remote sensing data, which is expected to be unbiased survey of treelines. They found that 88.2% of grid cells with treelines are in the Northern Hemisphere, and 11.8% are in the Southern Hemisphere. Therefore, the lower sample size and diversity in the Southern Hemisphere could be attributed to the Southern Hemisphere’s narrower land and mountains.

This treeline tree diversity spatial pattern may relate to flora differentiation resulting from the breakup of Pangea (37). From a pancontinental flora perspective, most of Asia, Europe, and North America are parts of the Laurasia, while South America, Africa, Australia, New Zealand, and the Indian subcontinent belong to Gondwana (38). The Laurasian region is dominated by Pinaceae species (39), which thrive in alpine habitats due to rapid radiation, cold tolerance, and efficient seed dispersal (40, 41). In contrast, the Gondwanan flora is characterized by distinct broadleaf groups isolated across separated continents (42), with evolution limited by geographic isolation, inefficient seed dispersal, and stable climates (42). This has led to more specialized, narrowly distributed treeline species in the Southern Hemisphere.

Within the continent, latitudinal patterns of taxon richness and treeline elevation are similar in temperate regions north of 40°N with the highest density of records (Fig. 1E). 40°N passes through the Alps-Himalayas Mountain belt and the southern Rocky Mountains; in these regions, treelines belong to a large number of genera, though they are restricted to relatively small areas (Fig. 1 B–D). This may be explained by the effects of greater mass elevation effects (MEE) in these regions (43), which would tend to create more heterogeneous environmental conditions, influence evolutionary divergence, and thereby lead to increased phylogenetic diversity in mountainous areas (44, 45). This phenomenon is reflected in the positive effects of mountain elevation and distance from coastlines on treeline taxon diversity (SI Appendix, Fig. S6), with both acting as proxies for MEE (11, 43). Therefore, local environmental differences may have a strong influence on local tree taxonomic turnover across a continent. We collected occurrence records from GBIF for each treeline tree taxon across the globe (SI Appendix, Fig. S7). Such records could broadly delineate the global extent of occurrence (46, 47). As shown in SI Appendix, Fig. S7, the extent of occurrences of these treeline taxa, especially several conifer genera and Betula across Eurasia and North America, tend to overlap considerably, reflecting multiple taxa theoretically co-occurring (46). Thus, for a given treeline site within the overlapped extent of occurrence, the taxon actually occupying the location would demonstrate better environmental adaptability and competitive capacity compared to other theoretically co-occurring taxa (48, 49).

Moisture and Climatic Variability Shape Distribution Patterns of Treeline Taxa.

We used PCA (SI Appendix, Fig. S8) to reduce dimension for 22 climate variables (i.e., 19 bioclimatic factors plus growing season length, mean temperature, and precipitation) and 21 soil property variables, respectively (see SI Appendix, Table S1 for the 43 variables). Based on the correlation among these variables (SI Appendix, Fig. S8 C and F) and their contribution (SI Appendix, Fig. S9) to each principal component, we selected four representative dimensions from climate and soil factors, respectively, in the following analysis (SI Appendix, Method S2). The eight environmental dimensions are overall moisture (Moisture), heat (Heat) regime, precipitation seasonality (P. seasonality), seasonality of diurnal temperature range (T. range), soil cation exchange capacity (Soil. cation), nutrient content (Soil. nutrient), physical properties (Soil. physical), and mineral and salinity content (Soil. mine). The pairwise correlations between each two of these dimensions are weak, with the absolute value of Pearson correlation coefficients generally below 0.3 and a maximum value of r = 0.29 between soil cation exchange and precipitation seasonality (SI Appendix, Fig. S10). SI Appendix, Fig. S11 shows the distribution of treeline genera across various environmental dimensions at treelines and occurrence on a global scale.

The phylogenetic signal analysis of environmental characteristics at the treeline revealed significant Pagel’s λ values for moisture regime, P. seasonality, and T. range (0.91, 0.61, and 0.49, respectively, P < 0.05 in all the three cases); meanwhile, λ for heat regime and the four soil dimensions were not significant (P > 0.05). This suggested that overall moisture regime and climatic periodic variability at the treeline show strong phylogenetic conservatism, with the adaptation of treeline species to these environmental constraints being influenced by evolutionary history.

However, the significant phylogenetic signal may actually represent biogeographic conservatism at such a broad spatial scale (50). We constructed a classification and regression tree (CART) to identify the factors driving the differentiation of tree genera at the treeline (Fig. 2A). As anticipated, the long-term isolation of the Laurasian and Gondwanan plates (contributing 21.0% to the differentiation of tree genera), first distinguished conifer treelines in the Northern Hemisphere from broadleaf treelines in the tropics and Southern Hemisphere. However, within the continental interior, we still found evidence of the dominating environmental filtering effect of moisture and climatic variability among the genera at treelines, according to the ranking of environmental factors contributing in the CART: moisture regime (22.7%), P. seasonality (16.8%), T. range (14.8%) soil nutrients (9.4%), heat regime (4.9%), soil cation exchange (2.9%), and mineral/salinity content (1.8%). Notably, overall moisture’s contribution (22.7%) exceeded that of the continent located (21.0%). Fig. 2B shows the distribution of treelines for each genus along the top two important environmental dimensions.

Fig. 2.

The variations of the environmental factors associated with the distribution pattern of treeline genera. (A) Classification and regression tree (CART) analysis of the spatial differentiation of treeline genera. The categorical variable (genus) with the largest number in the sample of this node and its proportion in all samples of this node is shown in the lowermost boxes. The CART shown here is the optimal decision-tree model that has been pruned and simplified based on the relationship between the cross-validation error (X-val error) and the complexity parameter (see SI Appendix, Fig. S12 for the unpruned model and details of pruning). “Pangaea region” factor indicates whether a treeline sample belongs to Laurasia or Gondwanan region. The Laurasia group contains samples in Eurasia and North America, mainly in the northern hemisphere, while the Gondwanan group contains samples in South America, Africa, and Oceania, mainly in the southern hemisphere. Although the Himalayan region is at the interface between Gondwana and Laurasia, considering spatial, geological, and biogeographical perspectives, we have classified the Himalayan and the broader Tibetan Plateau samples within the Laurasia group in this study. (B) Distribution pattern of treelines for each genus along the top two important environmental dimensions (moisture and P. seasonality) of CART. Points and error bars represent mean and SD, respectively.

According to the CART, Abies treelines thrived under the highest moisture levels (Fig. 2A), likely because Abies spp. relies on shallow roots drawing water from surface soils (51), enabling it to thrive in high-moisture environments by focusing on rapid growth and competition rather than drought tolerance. In contrast, Pinus spp. has deep roots that access groundwater, which adapts to surface drought (52); however, it is sensitive to moisture fluctuations and prone to xylem embolism under extreme drought (53). This explains why Pinus treeline species tend to occur in areas with minimal precipitation variation (Fig. 2A). Diurnal temperature variation affects photosynthesis and respiration rates (11, 54–56). Picea spp. thrives in stable, warm nights, and lower diurnal temperature variation (Fig. 2A) due to efficient water use resulting from smaller leaves and fewer stomata (57, 58), while Pinus spp. are characterized by longer needles than other Pinaceae and higher transpiration excels in large temperature swings (Fig. 2A). In the Southern Hemisphere, Polylepis and Nothofagus treelines are found in tropical and temperate zones respectively, and their primary distinction lies in precipitation seasonality rather than thermal differences (Fig. 2A). Polylepis is well adapted to dramatic seasonal shifts in rainfall, likely due to their dormancy during the dry season (59). The CART was pruned to ensure robustness and reduce overfitting; however, the original unpruned CART may still reveal some weak but noteworthy patterns (SI Appendix, Fig. S12). Juniperus spp. and Larix spp. treelines prefer drier regions (SI Appendix, Fig. S12) due to juniper’s thick cuticle and larch’s small, deciduous leaves (57, 60). Betula spp. tend to thrive in stable, well-watered regions (SI Appendix, Fig. S12), highlighting its competitive disadvantage and disturbance dependence in extreme conditions compared to conifers.

A Unified Threshold of Cold Tolerance for the Presence of Treeline Taxa.

While moisture and climatic variability were associated with the variation among the distribution patterns of different treeline genera (Fig. 2), we found heat regime to be a consistent dominating predictor for the presence of different treeline genera (Fig. 3A). Heat regime showed a lack of phylogenetic conservatism across treeline genera revealed by insignificant λ values (λ = 0.53, P > 0.05). Heat regime at most treelines were similar (SI Appendix, Fig. S13A); however, they were significantly lower than the optimal thermal condition for genera. The optimal thermal condition for each genus is defined as the mean of heat dimension associated with all occurrences of each genus (SI Appendix, Fig. S13A). Such levels of deviation were not observed in dimensions of the other environmental conditions (SI Appendix, Fig. S13). We developed a Relative Distance from Optimum (RDO) index to quantify such deviations, which is calculated as the difference between the actual treeline environmental conditions and the optimal conditions of the treeline taxon divided by the global range of the same environmental dimension for the taxon (see Eq. 1 in Materials and Methods for detailed calculation for RDO). Using RDOs as independent factors, we constructed a bivariate model with tree presence (coded as 1) around the treeline or absence (coded as 0) as the dependent variable (Materials and Methods). Deviation from optimal heat (RDO_heat) also emerged as the primary key factor at genus levels (Fig. 3A and SI Appendix, Fig. S14), underscoring that treelines form due to low heat conditions approaching the tolerance limits of plants. There is a broad consensus that temperature defines the ecological niche boundary of trees based on the growth limitation hypothesis (1); therefore, while heat cannot serve as a significant variable for spatial differentiation or for determining competitive dominants across treeline genera (Fig. 2A), it does set the growth limit for treeline-forming trees with various genera.

Fig. 3.

Environmental drivers and thresholds of treeline presence based on the Relative Distance from Optimum (RDO) index. (A) Effects of fixed factors were calculated based on a Bivariate Generalized Linear Mixed Model (BGLMM). The fixed effect factor is the RDO of each environmental dimension, the effect size (y-axis) is defined by the coefficient of fixed variables, and it is also the derivative at the 50% probability threshold. The height of the bars represents the mean of the regression coefficients obtained from 100 random sampling model iterations for each fixed effect variable. The larger the absolute coefficient, the stronger the effect. The error bars represent the boundaries of the 95% CI of the regression coefficients from the 100 iterations. Specifically, the lower bound corresponds to the minimum of the CI lower limits (2.5% percentile) across all 100 iterations, while the upper bound corresponds to the maximum of the CI upper limits (97.5% percentile) across the 100 iterations. (B) Fitted logistic regression of RDO of heat regime (the best factor in A) and treeline presence. The y-axis represents the presence (0) or absence (1) of tree life forms around the actual treeline positions. The label “above treeline (no tree)” at the Top corresponds to a value of 0, indicating that no tree life forms grow in these areas. Conversely, “below treeline (tree)” at the Bottom corresponds to a value of 1, indicating that tree life forms can grow. The steepest slope of a curve, indicated by the “treeline” y-label and the colored points, theoretically occurs at the 0.5 probability mark (y-axis), corresponding to the RDO threshold (x-axis) for treeline presence. This inflection point captures the abrupt transition from areas below the treeline, where trees are present, to areas above it, where trees are absent. Thresholds at species-level are shown in SI Appendix, Fig. S15. The probability density curves displayed at the Top and Bottom of the figure represent the distributions of absence (0, no tree) and presence (1, tree) samples, respectively. The x-axis aligns with the main logistic curve plot, showing the actual RDO_heat values for each sample point. The y-axis indicates the unitless distribution density, which is omitted here for simplicity.

Locations closer to optimal heat levels are more likely to be colonized by trees, with tree survival rates being the most sensitive to increasing heat constraints. When actual heat regimes dropped to an average of 35% (mean of RDO_heat ≈ −0.35 ± 0.12) below optimal levels, treelines formed (Fig. 3B). These thresholds were related to abrupt changes at the inflection point (steepest slope near a probability of 0.5) on logistic curves, resulting from the bivariate model, and highlight the transition from “tree” to “no tree” states (11). Given that some species may occupy unique ecological niches compared to their congeners, this may affect the representativeness of the genus-level RDO_heat threshold. Additionally, we calculated the species-level RDO_heat-derived threshold for 58 species that had five or more treeline samples (SI Appendix, Fig. S15), with an average threshold of approximately –0.4. Most species show values close to the genus average (SI Appendix, Fig. S15); however, one or two species per genus showed different thresholds (SI Appendix, Fig. S15). Larger RDO_heat threshold values indicate globally widespread species with broad ecological ranges, such as Scots pine (P. sylvestris) and Norway spruce (Picea abies). On the other hand, smaller RDO_heat absolute values may indicate species endemic to high altitudes or sensitive to environmental changes, such as the white bark (Pinus albicaulis) and foxtail pines (Pinus balfouriana), which are endangered or near-threatened species (61).

Although the RDO_heat-derived thresholds were similar across most genera, primarily concentrated around 30 to 40% (Fig. 3B), this threshold varied slightly for the Erica and Polylepis genera, both of which are mainly found in tropical regions. This variation underscores differences in how nontropical and tropical trees adapt to the treeline environment. Nontropical species typically first adapt to cold conditions in temperate or frigid zones before reaching the treeline, whereas tropical alpine species often bypass this step (62). Consequently, tropical treeline species may have lower cold resistance and may not be accurately represented in distribution models based solely on heat thresholds. Additionally, treelines dominated by Erica spp. demonstrated the greatest deviation from their thermal niche (RDO_heat = −0.57, Fig. 3B). The tropical monsoon climate in East Africa causes soil moisture to increase with elevation because of decreasing evaporation rate (63, 64), making Erica treelines the only group where moisture levels exceed the optimal niche of the genus (Fig. 2B and SI Appendix, Fig. S11), possibly compensating for heat lacking (8). For Polylepis spp. treelines, the threshold was nearly 0 (Fig. 3B). Previous studies suggested that these broadleaf trees lack ecotype adaptation (25), and their distribution edges do not form true climatic treelines (5). Moreover, the lack of differences between treeline and global heat optimum for Polylepis spp. (SI Appendix, Fig. S13A) highlights its alpine endemic nature (SI Appendix, Fig. S7). In addition, tropical alpine environments experience “winter every night” in contrast to temperate ones. Plants in these regions must cope with freeze-thaw cycles, avoiding ice damage at night and managing dehydration during the day. As illustrated in SI Appendix, Fig. S14, the Polylepis treeline demonstrates superior adaptation to diurnal seasonality (T. range) compared to other genera.

Overall, despite accounting for adaptive differences among taxa, the effect of heat on treeline presence exhibits greater consistency than other environmental factors, as reflected across taxa in the strength, direction, and threshold of this effect. SI Appendix, Table S3 summarizes the logistic curves fitted for each genus or species across various environmental dimensions. The coefficient of variation (CV) indicates that the effect size of RDO_heat and the thresholds it influences show the lowest variability at both genus and species levels, with effects maintaining the same direction (P < 0.001, t test) (SI Appendix, Table S3). Although the RDO_{Soil.physical}-derived threshold has the lowest CV at the genus level, its overall effect is minimal (Fig. 3A). The second factor, RDO_moisture (Fig. 3A), displays a consistent direction of effect across taxa (p < 0.001, t test); however, the effect strength and the thresholds it determines show considerable variability (SI Appendix, Fig. S16). These findings further support the robustness of treelines as indicators of a similar global thermal condition (5), while suggesting that the role of taxonomic differences is minor.

RDO_heat as an Ecological Threshold of Alpine Treeline.

Apart from treeline positions, about 3% of the other recorded locations of treeline genera worldwide fall in areas where the heat is below the RDO_heat -derived thresholds (Fig. 4A). These colder locations may not actually have been located above the treeline, but plants distributed here should be similar to those observed above the treeline. Combined with 30 m resolution raster data of global tree (Materials and Methods), we found tree cover and height were significantly lower in colder locations than in warmer locations (P < 0.05, Wilcoxon test, Fig. 4 B and C). A 13% tree cover and a tree height of 2 m best distinguished colder from warmer locations, although classification performance was moderate (AUC < 0.7) (Fig. 4 B and C). The 13% tree cover closely aligns with the forest threshold (10%) (65), and a tree height of 2 m is widely recognized as the critical treeline height (6). Thus, in terms of its definition, this RDO_heat-derived threshold tends to represent the prediction of actual treeline rather than the species line, marking the upper elevation limit of the realized ecological niche of tree life forms, with immature (seedlings, saplings) or shrub-like (krummholz) individuals possibly occurring above it (8, 66, 67).

Fig. 4.

Ecological significance of the RDO_heat-derived threshold for treeline presence. (A) Heat distribution at global GBIF sample points for different genera. The violin plot shows the frequency without outliers. The black hollow dots mark the mean (optimal thermal niche), the solid black dot marks the treeline presence threshold from Fig. 3B, and the blue and red dots indicate points with heat regimes below and above this threshold (sample size n = 119,264). The y-axis represents the score of the Heat dimension, which is unitless. (B and C) Boxplots for comparison of tree cover and height between the positions with heat regimes below and above the RDO-derived threshold. “***” indicates highly significant group differences based on the Wilcoxon test (P < 0.001). The dashed horizontal lines indicate the optimal classification according to ROC; “×” indicates median value. (D–F) show linear models (blue lines) fitted between the elevation of the actual treeline and RDO_heat-defined treeline (D), actual treeline and growing season defined potential treeline (E), actual treeline and ENM-defined treeline (F). Sample sizes are all 2133, and the linear fits are highly significant (P < 0.001). The red dashed lines represent the y = x fit.

There are another two commonly used niche boundary definition methods, including potential treeline definition based on the established growing season temperature of 6.4 ± 0.7 °C (10, 12), and Ecological Niche Model (ENM) using all kinds of environmental factors (see Materials and Methods for details). Based on their respective definitions, the former reflects the theoretical uppermost tree distribution determined solely by the physiological limit of tree life forms without any disturbances (1). The latter, on the other hand, defines taxonomic line, which is the distribution boundary of tree taxonomic groups without considering life forms (8). The above two methods and the RDO_heat -derived treeline estimation showed strong fit with each other (SI Appendix, Fig. S17) while the latter was significantly lower (t test, P < 0.05), and the latter outperformed them in fitting the TSAT-recorded actual treeline elevation, with the highest R² (0.97), smallest mean bias (97 m), and a regression coefficient closer to 1 (Fig. 4 D–F). The RDO_heat-derived thresholds exhibit an elevation overestimation rate of 51% and an underestimation rate of 5%. In comparison, the other two methods show overestimation rates of 90 and 69%, and underestimation rates of 8 and 13%, respectively. Those results further highlighted that RDO_heat-derived thresholds better match the actual tree-form distribution boundaries, which are lower than both theoretical potential treelines and taxonomic lines. They also capture taxon-specific tree tolerance, while offering a more straightforward and accurate simulation based on logistic models than complex ENMs. A more precise understanding of the environmental drivers of treelines can improve the predictions of treeline dynamics and their impacts on alpine ecosystems under future climate change, offering crucial insights into the future dynamics of alpine endemic habitat (12), providing forward-looking guidance for biodiversity conservation efforts.

Conclusion and Outlook

Overall, this study provides essential data and methodologies for global biogeographical research, with a focus on taxonomic variation at the treeline. It improves upon previous studies that only examined environmental controls. We indicate that in treeline formation, heat limits all tree growth as a “ceiling” by restricting basic physiological functions (1). Trees are absent and treelines are established when the available heat falls below approximately 35% of the thermal optimum for each genus within its niche range, especially for treelines in temperate zone. This 35% threshold presents an effective approach for describing observed treeline elevation patterns while accounting for taxonomic differences. Beneath this “ceiling” of heat, supply and seasonality of water act as “filters” in determining the taxonomic variation of treelines, by immediate impacts on photosynthesis of plants (68) These “filters” display stronger pressures than soil factors. This is likely because the stress of water availability depends entirely on environmental conditions and taxa’s adaptive traits, thus the stress of water is more fixed, while soil nutrient deficiencies could be partly offset by processes like litter decomposition (69). In summary, while prior research acknowledged heat as the primary factor influencing treelines, our work verifies this principle from a taxonomic perspective. We also expand on this understanding by highlighting the importance of moisture in shaping treeline patterns. Our findings demonstrate that moisture, along with its variability, determines which tree species can survive at the treeline, while temperature largely governs where the treeline occurs.

There are still some limitations and caveats: As this study integrates global observational data, it is inherently subject to sampling bias. The greater number of treeline records available in the Northern Hemisphere compared to the Southern Hemisphere may influence perceived patterns of treeline species diversity at a local or regional scale. However, this disparity also reflects the actual biogeographic and floristic patterns observed worldwide. Relying on global biodiversity databases, the measured niche width may diverge from the true ecological values. To minimize this discrepancy, we have implemented measures to improve the reliability of the RDO calculation. Future research could involve transplant experiments or controlled field studies to directly assess the ecological niche of each treeline taxon. However, such studies may be challenging due to logistical difficulties and resource limitations. Generally, this study would enhance our understanding of taxa sensitivity to extreme environmental conditions, provides robust tools for predicting future dynamics of alpine vegetation, and identifies vulnerable endemic tree species crucial for mountain biodiversity conservation in the face of climate change.

Materials and Methods

Construction of the TSAT Dataset.

The Tree Species on Alpine Treeline (TSAT) dataset (30) is based on a systematic review of studies at alpine treeline at different scales published from 1980 to the present. Each study has meticulously documented the latitude, longitude, and species information for one or more alpine treeline points. Regarding species information, the selected studies must report the dominant tree species, defined as a tree species recorded as the highest elevation of a treeline site, or, in the absence of such a record, the tree species that accounts for the highest relative abundance within the treeline site. Approximately ~1,000 records were taken from the screening of the ToTE (21), which is a dataset of treeline species compiled from literature integration following the Preferred Reporting Items in Systematic Reviews and Meta-Analysis (PRISMA) protocol. We performed additional curation and validation of these records. See SI Appendix, Methods S1 for more dataset collection details and criteria. We calibrated the scientific names of the tree species using the “Taxonstand” package in R to follow The Plant List, which is a broadly accepted and widely accessible working list of known plant species developed and disseminated in direct response to the Global Strategy for Plant Conservation (70). Finally, the TSAT dataset contains 2,235 records based on 739 nonrepetitive field observations, across 39 mountain regions, 43 countries, and 6 continents (Fig. 1A and SI Appendix, Figs. S1–S4).

To enhance the reliability and robustness of our subsequent quantitative analyses, we focused exclusively on the top 10 most common treeline genera, each with at least 10 records (i.e., present at 10 treeline sites). These top 10 genera encompass 2,133 of the total 2,235 records. When the number of sampling points for a given genus is too small, the estimated thresholds may be disproportionately influenced by environmental conditions at one or a few sites or outliers, resulting in biased or unrepresentative outcomes. The use of threshold of 10 is expected to help mitigate such biases, thus strengthening the robustness of genus-level inferences. While the “top 10” genera selected for detailed analysis meet this minimum requirement, many of them have substantially more than 10 records.

Collection of Worldwide Distribution Records of Treeline Tree Species.

To compare with the environmental conditions at the treeline and to calculate the RDO index, we downloaded global comprehensive occurrence records for each treeline tree species from the Global Biodiversity Information Facility (GBIF) database using the R package “rgbif” (71). GBIF is an international network and data infrastructure that provides data-holding institutions worldwide with universal standards, best practices, and open-source tools for sharing species observation information. To ensure the veracity of the occurrence data, the accuracy of recorded locations, and that the records adequately represent the species in their natural states, we employed the “CooperativeCleaner” R package (72) for data cleaning. The criteria for cleaning involved removing records that a) lacked coordinates, b) had coordinate uncertainty greater than 10 km, c) were located in urban areas or biodiversity institutions such as zoos and museums, and d) were based on fossil material samples or living collections. By the end of February 2024, we obtained a total of 119,262 records for all the treeline species.

Regardless, records based on human observations are susceptible to biases and limitations, including errors in species identification, inconsistencies in species naming over time, and the presence of complex subspecies, factors that can lead to unreliable species records that are difficult to correct. These issues may result in inaccurate matching of global occurrence points for treeline species during GBIF data aggregation. Additionally, some species, particularly alpine endemics, may have very limited records in GBIF, increasing uncertainty in their distribution estimates. In contrast, genus-level records are generally more complete and less prone to errors, especially in understudied regions such as alpine areas. Therefore, in this study, RDO calculation and modeling analyses were primarily conducted at the genus level. Nonetheless, we also performed supplementary analyses for certain species with a higher number of records to validate our findings.

Dimensional Reduction of Global Environmental Indicators.

Environmental factors influencing alpine treelines are diverse. Regardless of considering taxon-specific environmental adaptations, factors other than temperature may play a significant role. For example, Camarero et al. (73) found the positive temperature-tree growth relationship on alpine worldwide has been fading along with global warming in the past decades. Tree-ring data showed that water has been the main limiting factor for radial growth in recent decades (17, 74). Moreover, variability in climate, such as diurnal temperature range and the seasonal hydrothermal fluctuations, has also been shown to potentially limit tree survival at the treeline by intensifying freeze–thaw cycles and increasing the risk of frost events (8). In addition to climate-related environmental factors, tree regeneration experiments have found that soil moisture availability controls seedling recruitment at treelines (15). Soil properties, such as nutrient content and mechanical composition, also influence seedling regeneration and tree growth at treeline (75). Due to poor soil development in alpine regions (76), soil conditions may also serve as important limiting factors. Those studies indicate the influence of factors other than temperature on treeline dynamics. Since the last century, 48% of the 166 treeline sites worldwide have not significantly risen with climate warming, and some have even declined (35). This indicates that warming is not the only environmental factor causing treeline elevational change. Therefore, a more comprehensive set of environmental variables should be incorporated into global-scale studies of treeline distributions. However, previous studies investigating the contributions of multiple environmental factors to treeline elevation or distribution (11, 12, 77) have not considered the influence of differences among treeline taxonomic groups.

In this study, environmental factors are considered from both climatic and soil perspectives, including overall regime and variability of temperature and precipitation, physical and chemical properties (i.e., cationic exchange capacity, content of nutrients, mineral and salinity) of soil. The basic indicators are derived from the Climatologist at High Resolution for the Earth’s Land Surface Areas (CHELSA) v2.1 and the Harmonized World Soil Database (HWSD) v2.0. CHELSA provides very high resolution (30 arc sec, ~1 km) climatic data specifically designed for hydrological and ecological assessments, including climate layers for various time periods and variables, ranging from the Last Glacial Maximum, to the present, and to several future scenarios (78). This study utilized present multiyear (1981−2010) average data of 19 bioclimatic variables including length, mean temperature, and precipitation of the growing season period (GSL, GST, and GSP). The 19 bioclimatic variables (see details in SI Appendix, Table S1) were derived from temperature and rainfall data, capturing essential ecological trends, seasonality, and extreme climatic conditions, which aid in the study of species distribution and climate impact on biodiversity (79). Hydrothermal conditions during the growing season are crucial in the estimation of potential treeline elevation (1, 5, 9, 11), and climatic periodic fluctuations, particularly seasonal temperature variations, would influence plant water use efficiency and growth by affecting processes such as freeze-thaw cycles and photosynthesis (8, 80, 81) The HWSD offers very high resolution (30 arc sec, ~ 1 km) comprehensive soil information, facilitating assessments of soil resources at global, national, and regional scales. Surface soil directly influences the nutrient availability, water retention, and root development, which are essential for tree growth at high elevations (8, 82). Consequently, this study selected various physicochemical indicators of topsoil (0 to 30 cm) for further analysis (SI Appendix, Table S1). We then extracted the aforementioned environmental indicators into the TSAT, the global location records of each species from GBIF, as well as sample points created for our subsequent linear mixed model analysis in ArcGIS v10.8 software.

To identify the principal factors that capture the majority of the environment variables, we applied PCA using the R package “FactoMineR” (83) to reduce the dimensionality of 22 climatic variables and 21 soil physicochemical properties, respectively. For the representative dimensions selected to indicate environmental conditions for further analysis, we considered the explanatory power of each dimension for all the factors, as well as the ecological significance reflected by the contribution and correlation of the variables to each dimension. See SI Appendix, Method S2 for details of selection and the ecological meaning of principal component dimension.

Phylogenetic Signal Detection and CART for Environmental Treeline Distribution.

To explore the environmental mechanisms that lead each treeline taxon to occupy their current spatial space, first, we used phylogenetic signal detection to assess whether the variation of extreme environmental toleration of treeline taxa, represented by their geographical distribution, is constrained by phylogenetic relationships. Pagel’s λ (84) is widely used to measure phylogenetic signals because it better prevents pseudobranching from species showing bias in large datasets (85, 86) compared with Blomberg’s K (87). λ values closer to 0 indicate weak signals, while higher values suggest stronger conservatism. We used the “phylosig” function in the “phytools” R package (88) to calculate λ. The phylogenetic tree (SI Appendix, Fig. S5) was constructed using the “V.PhyloMaker” package, based on the mega-tree “GBOTB.extended.tre”(89).

Then, we employed Classification and Regression Tree (CART) analysis to distinguish the environmental conditions needed for different treeline taxa to occupy treeline positions. The response variable was the genus, and explanatory variables were environmental factors and the Pangaea region (Laurasia or Gondwanan) in which the sample points were located. The introduction of the Pangaea region factor broadly delineates the floristic differences between the Northern and Southern Hemispheres. Although the Himalayan region lies at the interface between Gondwana and Laurasia (90), its flora predominantly exhibits characteristic Laurasian biogeographic features, including temperate and subalpine taxa, rather than typical Gondwanan taxa (91), especially in alpine areas. Considering both geological and biogeographical perspectives, we have classified the Himalayan and the broader Tibetan Plateau samples within the Laurasian group. Therefore, the Laurasia group comprises samples from Eurasia and North America, mainly in the Northern Hemisphere, while the Gondwanan group includes samples from South America, Africa, and Oceania, mainly in the Southern Hemisphere. CART determined the optimal environmental cut points between treelines of different taxonomic groups. To improve the robustness of the model and reduce overfitting, we used a postpruning method to systematically prune a fully grown CART, removing branches based on a complexity parameter (cp) to balance tree complexity and error. Cross-validation was employed to select the optimal cp, helping prevent overfitting and enhancing model performance (92). Calculations above were performed using the “rpart” package (93), and visualizations were done with “rpart.plot” (94).

Calculation of the RDO Index.

We defined the Relative Distance to Optimum (RDO) as the deviation of the environmental value at a given location from the optimal environmental value within the range of ecological niche of a given genus or species (see the below Eq. 1). Compared to using raw environmental values as thresholds, RDO is a relative index, allowing an easier comparison of different taxonomic groups along different environmental dimensions. Trees from distinct groups respond differently to environmental pressures; thus, the same environmental condition can have different impacts. For example, under identical temperature conditions, a tree of a cold-tolerant taxon may still be within its optimal ecological niche, whereas a less cold-tolerant taxon might already experience stress that limits growth. Therefore, it could be harder to find generalized patterns across different species based on the variety of raw (absolute) environmental values, and a relative value (RDO) could overcome this limitation by scaling the values to the magnitude of 1.

$${\mathit{RDO}}_{i,j}=\frac{E_j-{\mathit{OE}}_{i,j}}{{E_{\mathit{upper}}}_{i,j}-{E_{\mathit{lower}}}_{i,j}}.$$ [1]

In Eq. 1, i and j represent the given treeline genus or species and environmental dimensions (including eight principal components of climate and soil conditions derived from the PCA), respectively. E_j is the actual environmental value of the sample point (such as treeline position), and OE_i,j is the optimal environmental value (ecological niche optima) of the given genus or species. The denominator is the ecological niche breadth, which refers to the range of each environmental factor a plant can survive, calculated as the difference between the upper (E _ upper_i,j) and lower (E _ lower_i,j) limits of environmental factor tolerance. OE_i,j, E _ upper_i,j and E _ lower_i,j were obtained from the GBIF global occurrence records of treeline tree taxa, based on the mean, maximum, and minimum values for a given genus or species across specified environmental factors; when conducting genus-level analysis, the GBIF global occurrence records cover all species under each treeline genus; while for species-level analysis, the occurrence records correspond to each specific treeline species. In summary, E_j depends only on the environmental dimension values at the sample location, while OE_i,j, E _ upper_i,j and E _ lower_i,j depend on the taxon and environmental dimension under consideration and do not vary with the sample location.

We acknowledge that sampling bias inherent in global occurrence data based on field observation is difficult to fully eliminate. Field observation-based sampling often misses human-inaccessible areas, leading to errors in coordinate records and misidentification of species within the same genus or with similar appearances (95). As a result, it is unlikely that we will ever perfectly replicate the true niche width based solely on global observations. To mitigate these limitations and ensure that our observations align as closely as possible with the theoretical distribution patterns, we implemented several measures. First, we employed the largest and most widely used global species occurrence database (GBIF), aiming to obtain a wide coverage of species occurrence data. Second, to reduce the influence of sampling and spatial biases, we selected only one occurrence point per environmental pixel, thereby minimizing redundancies and overrepresentation of certain conditions. Furthermore, to mitigate the impact of outliers, potentially caused by incorrect GBIF records, we applied outlier removal based on the Interquartile Range (IQR) method (96). The IQR is calculated as the difference between the third quartile (Q3) and the first quartile (Q1), and outliers are values below Q1 − 1.5 × IQR or above Q3 + 1.5 × IQR (97). Using this approach, the adjusted (after excluding extreme values) niche width bounds are denoted as E _ upper_i,j and E _ lower_i,j. Some studies use the peak of the performance curve to represent the OE_i,j (98, 99); however, since this curve is based on density distribution or abundance histograms, its peak can be influenced by bin size and sampling (100). Thus, using the mean value provides a more comprehensive and robust approach to further avoid the issue of anomalous peaks that may result from sampling biases.

Bivariate Generalized Linear Mixed Model (BGLMM) for Treeline Presence.

To explore key environmental factors driving treeline presence and the variations of these drivers among different species, we utilized a BGLMM and ranked environmental factors based on fixed effect size. A bivariate model simulates a binomial distribution and is widely used to explore the determinants of environmental event occurrences due to its concise form and the valuable insights it provides (101, 102). The “pglmm” function in the R package “phyr” was used for BGLMMs, which can fit models with “nested” covariance structures under Bayesian mode (103).

The dependent variable in this study was the presence or absence of tree around treeline points. Presence was coded as 1, while absence was coded as 0. This variable was derived from sample points within a 150 s × 150 s (~5 km × 5 km) area, which included 25 raster pixels around each treeline point. Sample points with elevations higher than the nearest treeline point were assigned a value of 0, indicating the absence of trees. Conversely, points with elevations lower than the treeline point were assigned a value of 1, indicating the presence of trees. The elevation of each point was assigned using the 30 m resolution Global Digital Surface Model from Advanced Land Observing Satellite (ALOS, v 3.2) (104).

Fixed-effect variables included the RDO_i,j of each of the 8 environmental dimensions at the sample points around treeline. It is important to note that nearly half of these sample points are located above the treeline, where environmental conditions theoretically do not support tree growth or only allow survival in krummholz form. Consequently, the environmental conditions at these points may fall outside the ecological niche width of the trees, potentially even well below E _ lower. As a result, the RDO values for these points may be substantially lower than 1, or may have no lower limit. The random factor includes genera, mountain ranges, and 5° latitude bands those the treelines belong to. Based on the mathematical foundation of the bivariate model, the fixed effect size for each variable reflects the abruptness of the 0-1 transition between tree presence and absence, i.e. indicating the treeline position. This corresponds to the derivative at the 50% probability threshold, where a larger absolute value signifies stronger control by the environmental factor on treeline presence (11, 105).

We also employed a random sampling approach to exclude the potential impact of uneven sampling density across mountains worldwide. Based on the distribution of “0, 1” sample points around the actual treeline, we found that the mountain range with the fewest sample points had only 40 points. Therefore, we randomly selected 40 sample points from all the 38 mountains or regions each time, built a subset of the data, and performed modeling. This sampling and modeling were repeated 100 times, and we recorded the mean of the regression coefficients (effect sizes) for each dependent variable across these 100 iterations.

Threshold for Treeline Presence Defined by RDO.

The primary driving factor was defined as the RDO_i,j with the largest absolute fixed effect size. To identify the RDO_i,j threshold driving treeline presence, we located the inflection point on the logistic curve of the bivariate model. First, we extracted the predicted probabilities from the logistic regression model across the range of the primary factor, while keeping other variables constant. Next, we computed the first derivative of the logistic curve with respect to this factor and identified the point where the derivative (slope) reached its maximum (typically around the 50% probability threshold). The horizontal coordinate value corresponding to this point is the threshold for treeline presence of the given genus of plant. This point represents the critical threshold of change between 0 and 1, and is commonly used to identify ecological boundaries (106, 107), making it suitable for describing treeline thresholds.

We assume that the RDO_i,j threshold represents the actual treeline, meaning that trees distributed near this environmental threshold worldwide should be similar to those observed at the treeline, reflecting the growth limit of the species. Since a height of approximately 2 m is often used as a critical threshold for defining tree life forms, and treeline trees are generally found outside of forested areas [defined as areas with >10% tree cover (65)], our assumption can be viewed as a validation that RDO thresholds can categorize the global occurrence of treeline plants (from GBIF) into two groups: those with plant height exceeding 2 m and those located outside forests. These two groups are significantly differentiated, supporting the ecological relevance of the thresholds. To validate this, we used 30-m resolution tree cover and height data to assign GBIF sample points and applied the Wilcoxon test to compare values on either side of the RDO_i,j threshold. Tree cover data were obtained from the mean value of Hansen Global Forest Change v1.11 (108), Global Forest Cover Change (GFCC) v3 (109) and Copernicus Global Land Cover Layers-Collection 2 (110). Tree height data were obtained from Global Forest Canopy Height 2019, which was calculated from Global Ecosystem Dynamics Investigation (GEDI) lidar data (111). Additionally, we identified the optimal boundary for tree cover and height between the two types of sample points (on either side of the RDO_i,j threshold) based on the maximum sum of the sensitivity and specificity of the receiver operating characteristic (ROC) curve according to the area under the curve (AUC) (112, 113). However, the uncertainty in this process may be related to data resolution matching. Since occurrence data points from GBIF lack specific information such as tree height and cover, we utilized global raster data for these variables. The available data have a resolution of 30 m, which can only represent average canopy conditions within a certain area surrounding each occurrence point, rather than characteristics of individual trees. The validation conducted in this study aims to identify general trends, and the large number of occurrence points (N > 100,000) may help mitigate the impact of resolution-related accuracy errors.

Comparing RDO-Defined Thresholds With Two Other Boundary Categories.

Based on niche theory (1), in addition to the treeline threshold, we defined two other models to predict global treeline distribution using the RDO_i,j at the logistic curve inflection point: the potential treeline model based on mean growing season temperature (9, 10) and the plant distribution boundaries simulated by the ENM. For each actual treeline recorded in TSAT dataset, we identified the nearest position for all three of the above theoretical treeline modeling results.

For the RDO_i,j - defined treeline, we first input all climate variable raster data into the constructed PCA space and assigned the second axis scores, representing heat regimes, back to the original raster. Next, we calculated RDO_heat values for each genus and extracted the transitional raster between those above and below the threshold for each genus. These transitional rasters define the theoretical treeline distribution based on this method.

For potential treeline determined by mean growing season temperature, we extracted the raster area with a mean growing season temperature of 6.4 ± 0.7 °C (10). This fixed temperature threshold reflects commonalities in tree life forms under extreme environmental conditions, and it has been widely applied by many previous studies for predicting treeline elevations at different scales from mountain ranges to continents to the globe (11–13). However, we recognize that this approach does not account for variations across different taxa and regions. One of the aims of our study is to introduce a taxonomic perspective, which could provide a more nuanced understanding of treeline distribution patterns. By comparing the results derived from the fixed temperature model with those from our taxonomically informed model, we seek to evaluate whether incorporating such biological variations can enhance the accuracy of treeline elevation predictions across multiple spatial scales.

For ENM simulations, we employed Maxent methods using the “dismo” R package “maxent” function (114). Occurrence data points included records in TSAT as well as in GBIF, and background points were randomly generated in equal numbers within a 2-degree buffer around the occurrence points. Following Feng et al. (14), we reported a checklist for the reproducibility of the Maxent ENM (SI Appendix, Table S4). The maximum sensitivity plus specificity threshold for ENM (SI Appendix, Table S5) was employed as theoretical treelines under the ENM method.

The theoretical treeline area was calculated using 30 s (~1 km) raster data, while treeline elevation was derived from 30 m-resolution ALOS DSM data. To calculate elevation differences, we first found the maximum and minimum 30 m-resolution ALOS DSM values within each 30 s pixel. If the actual treeline was below the minimum or above the maximum, we calculated the difference accordingly; otherwise, the difference was 0 (11). The elevation of each actual treeline plus the difference equals the theoretical treeline elevation. Finally, we fitted linear models between different treelines and compared the effectiveness of various models in simulating actual treelines.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Data, Materials, and Software Availability

The TSAT dataset we conducted in this study are available at https://doi.org/10.6084/m9.figshare.28966370 (30). Climatologist at High Resolution for the Earth’s Land Surface Areas (CHELSA) v2.1 database can be obtained from https://chelsa-climate.org/downloads/. The Harmonized World Soil Database (HWSD) v2.0 can be obtained from https://www.fao.org/soils-portal/data-hub/soil-maps-and-databases/harmonized-world-soil-database-v20/en/. Hansen Global Forest Change v1.11 can be obtained from https://storage.googleapis.com/earthenginepartners-hansen/GFC-2023-v1.11/download.html. Global Forest Cover Change (GFCC) v3 can be obtained from https://lpdaac.usgs.gov/products/gfcc30tcv003/. Copernicus Global Land Cover Layers-Collection 2 can be obtained from https://land.copernicus.eu/en/products/global-dynamic-land-cover. Global Forest Canopy Height 2019 can be obtained from https://glad.umd.edu/dataset/gedi. All process data form and R codes used for the analysis in this study are available at https://doi.org/10.6084/m9.figshare.28968662 (115). All study data are included in the article and/or supporting information.