1. Introduction

One of the most ubiquitous patterns in ecology is that most species are rare, whereas only a few species are abundant in local communities (Gaston, 1994). Uncovering the mechanisms leading to this universal ecological pattern has puzzled ecologists for decades (Rabinowitz, 1981; McGill et al., 2007). Despite important advances in understanding how many rare species coexist (McGill et al., 2007), community-level studies primarily focus on the effects of biotic and abiotic factors at local scales (Vellend, 2016). Accordingly, how large-scale effects (e.g., regional historical and current environment) shape species rarity (quantified by both relative abundance in a local assemblage and geographic range defined by the number of occupied sites) across large spatial scales remains largely unclear.

Regional and local drivers are likely to co-determine patterns of species rarity in local communities (Arellano et al., 2014). Within the framework of community assembly theories, the species-pool hypothesis offers a useful perspective for integrating regional and local influences (Jiménez-Alfaro et al., 2018; Jarzyna et al., 2025). This hypothesis states that local species diversity (e.g., species richness and rarity based on species abundance) depends on regional species diversity through regional processes (e.g., speciation, extinction, and large-scale dispersal) and local abiotic and biotic filtering (Vellend, 2016). For example, local species richness and rarity should depend on species-abundance distribution in the regional species pool (Jiménez-Alfaro et al., 2018). Specifically, a species' prevalence in the regional species pool may determine its probability of entering a particular local community when a dispersal event occurs, shaping local-scale rarity (Vellend, 2016). Therefore, compared to widely distributed species, rare species are less likely to disperse to a focal site within a given region (Cornell and Harrison, 2014).

Regional factors such as centennial and millennial changes in macroclimatic conditions (e.g., long-term climatic stability), geographical characteristics, and anthropogenic activities have been reported to relate both regional species pool size and species rarity (Kreft and Jetz, 2007; Svenning et al., 2015; Enquist et al., 2019). For example, macroclimatically stable regions often support a greater number of rare species (Enquist et al., 2019; Song et al., 2021). At these broad scales, anthropogenic stressors can alter species richness, rarity and evenness within ecological communities by modifying species' relative abundance (Hillebrand et al., 2008). In addition, local environmental conditions (e.g., microclimate and topographical heterogeneity, and soil conditions) play a critical role in shaping population viability and thus contribute to variation in species abundance within local communities (Ricklefs and He, 2016; Vellend, 2016). For example, local environmental filtering may lead to uneven distribution of species abundance and even local extinction of some rare species (Storch et al., 2022). It is thus necessary to consider the complex interplay of regional and local factors when interpreting patterns of species rarity.

In this study, we disentangled the influence of regional and local factors on local richness and rarity of tree species in subtropical and tropical forests of China. This region is renowned for its exceptionally high plant diversity and rich evolutionary history, hosting more than 80% (~26,000 species) of Chinese vascular plants and nearly 80% of the country's endemic plant genera (Lu et al., 2018). Under the influence of the Tibetan Plateau uplift and a typical monsoon climate, species-rich evergreen broad-leaved forests have developed and diversified in this region. As the world's largest continuous subtropical forest area, it harbors nearly 70% of the world's subtropical forests dominated by families such as Fagaceae, Lauraceae, and Theaceae (Li et al., 2025), while other regions at the same latitudes are deserts or semi-deserts (Song et al., 2013). Given the global significance of this region, it is crucial to understand how regional species pools and large-scale drivers shape the richness and rarity of local species assemblages in this region. To assess how regional and local effects contribute to local tree species diversity, we used the conceptual framework presented in Fig. 1. We hypothesized that species pool size, along with regional and local environmental factors, affect tree species richness and rarity, leading to the following predictions.

(1) According to the species pool hypothesis, regional species pool size is a positive driver of local species richness because it determines the species number potentially available for colonization (Zobel, 1997). While environmental filtering and dispersal limitation shape community assembly, their effects on richness are constrained by the regional pool size (Srivastava, 1999). Therefore, we expect that regional species pool size positively influences local tree richness, as shown in recent studies (Ricklefs and He, 2016; Jiménez-Alfaro et al., 2018; Jarzyna et al., 2025). In contrast, local species rarity depends not only on species arrival but also on post-colonization processes, such as local abundance regulation (Dobrowski, 2011; Vellend, 2016). Rare species constitute a non-random subset of the regional species pool, and are often characterized by narrow climatic tolerances and limited dispersal ability, which may reduce their dependence on the regional species pool (McGill et al., 2007). Thus, we predict that regional species pool size would have a weaker effect on local species rarity than on species richness.

(2) Regional species pool sizes depend on the abundance of suitable habitat available historically, which is influenced by environmental factors like climate, climatic stability, and human activities (Zobel et al., 2011). Consequently, regional environmental conditions can affect local diversity both directly and indirectly through changes in the regional species pool, with a greater impact on species richness than on species rarity. Moreover, compared with common species, rare species might be highly sensitive and often respond rapidly to anthropogenic disturbances (Hillebrand et al., 2008; Xu et al., 2019a; Song et al., 2021). Hereby, we expect that local rarity would decline with increasing human activity.

(3) As proposed by Janzen's mountain pass hypothesis (Janzen, 1967), climatic variations in (sub)tropical mountains is relatively low, creating physiological barrier between species inhabiting valleys and mountain passes. This uniformity promotes allopatric speciation and ultimately leads to high species diversity (Ghalambor et al., 2006). Accordingly, we expect a negative relation between species diversity and local climatic variations in the studied forests. We further predict that the low climatic seasonality exerts a stronger effect on species rarity than on richness, because rare species tend to have narrow climatic tolerances and high sensitivity to environmental change (Gherardi and Sala, 2015; Shrestha et al., 2018).

2. Material and methods 2.1. Study area

The study was conducted in Chinese subtropical and tropical forest regions. These regions cover nearly one-third of the total land area of China, ranging from 18 to 34°N in latitude and 90 to 123°E in longitude. The high geodiversity of the study area creates a wide range of habitats for highly diverse species, with nearly 80% of Chinese endemic vascular plant genera occurring in this region (Lu et al., 2018). The orogeny of the Himalayas, the uplift of the Qinghai-Xizang Plateau, and the subsequent development of the East Asian and Indian monsoons have huge impacts on the topography and climate of these forests (Hong and Blackmore, 2015). Accordingly, the great and complex climatic and topographic heterogeneity result in high plant diversity. The study area is subject to a typical monsoon climate, with cool, dry winters and warm, wet summers. Mean annual temperature varies from 15 ℃ to 22 ℃, and annual precipitation from ~800 mm to 3000 mm (Song et al., 2013). Since the study area occupies several densely populated and economically developed regions such as Yangtze River Delta and Pearl River Delta Megalopolis, the biodiversity of subtropical and tropical forests is also facing an unprecedented threat (Song et al., 2013).

2.2. Plant community data

We used plant community data from the Vegetation Archive of Subtropical and Tropical Regions in China (VAST-China; http://www.givd.info/ID/AS-CN-006). The dataset (VAST-China v.2.0) compiles 808 data sources, includes 11,640 vegetation plots and 11,773 vascular plant species. All species names were standardized following the World Flora Online (WFO, 2024) and the Flora of China (http://www.efloras.org/) using the R package U.Taxonstand (Zhang and Qian, 2023) and its online version (Zhang et al., 2025). Due to the different abundance scales used in vegetation surveys (mostly Braun-Blanquet cover-abundance, Drude abundance, and relative importance value), following the methods proposed by van der Maarel (1979, 2007) and Song et al. (2013), we transformed these abundance scales to the median of Ordinal Transfer Value (OTV) for the purpose of cross-comparison and data synthesis. The median of OTV was a plausible scale grades to define the relative dominance of species in a community from 0 to 100%. To reasonably represent species completeness, we first filtered the data to keep plots with at least three species or high species completeness estimated by the sum of median of OTV ≥ 70% proposed by van der Maarel (1979, 2007).

To ensure a representative regional species pool in this study, we selected the vegetation plots located in subtropical and tropical forest regions according to Chinese vegetation classification system (The Editorial Board of Vegetation of China, 1980). These regions share the similar origins of the vegetation and evolutionary history (Chang, 1993; Lu et al., 2018). Evergreen broad-leaved forest and tropical monsoon rain forest are the zonal vegetation types. We first excluded the plots with strong human disturbance (managed forests, plantations, and mangrove forests), non-forest plots, and the ones with missing geographical coordinates. To avoid the potential overfitting effects of vegetation plots size on plant diversity, only plots with known sizes ranging from 100 m² to 10000 m² were selected. After this filtering, over 90% of the studied plots had the areas < 1000 m². The plot size was used as a covariate for all the data analyses. Furthermore, we retained the plots originally recorded as tree layer. To maximize data utilization, we also included the plots where the tree and shrub layers were not separated in the original references, treating them as woody plant layer in our analyses. This was justified by their low proportion and the dominant species of tree species within them. According to the above criteria, 3923 vegetation plots were selected for regional species pool and further filtered to quantify local plant diversity, including 3802 plots (97%) with detailed plant records at tree layer and 121 plots (3%) at woody plant layer.

The whole study area was divided into 365 grid cells at 100 km spatial resolution (the regional scale used in this study). The 3923 sampling plots were assigned to grid cells and occupied a total of 212 grid cells. The number of plots within each grid cell ranged from 1 to 171. To reduce the influence of spatial sampling bias (i.e., oversampling of certain areas), we retained no more than five plots with exact geographic locations (e.g., site-specific coordinate imprecision or rounding) in each grid cell. For the 23 grids (10.8% of total grid cells number) with more than 40 plots, we randomly sampled 40 plots, because total plant richness of 40 plots was not significantly different from the one of all the plots within each grid (Wilcoxon test, P = 0.09). Finally, among 3923 plots, 2960 plots were selected to quantify the local species diversity across all grid cells. Overall, 50142 occurrence records from 3307 tree species were included. Among these species, 1185 species (36%) are endemic to China, 328 species (9.83%) are threatened, and 28 species (0.85%), such as Nyssa shweliensis and Magnolia angustioblonga, are critically endangered (Qin et al., 2017).

2.3. Measures of species diversity

Species richness and rarity of local-scale tree communities were used. Measures of species rarity are based on a multidimensional concept, encompassing local abundance, geographical range/range size, habitat specificity, and population size (Gaston, 1994). Assuming that one species may be rare because it always has few individuals at many local sites or many individuals at few sites (Enquist et al., 2019), local abundance and geographical range were widely used to calculate species rarity (Laffan et al., 2013). Here, we considered both relative abundance in a local assemblage and the number of occupied sites to measure rarity. Species rarity was calculated using the same algorithm as the endemism indices, but with weights based on sample counts instead of groups occupied (Laffan et al., 2013). The formula was as follows: Species rarity = ∑t∈TatAt, where a_t was the relative abundance of the species t in a site, and A_t was the total number of samples occupied by species t across the whole study area. The higher the value, the higher the species rarity in a community. The calculation of species rarity was done by using the software 'Biodiverse' (Laffan et al., 2010). The correlation coefficients showed that species rarity had relatively low correlations with richness (Pearson's r = 0.43^***), indicating that these two measures can well represent different diversity properties of the communities.

2.4. Regional species pools and environmental factors

Regional species pools refer to the set of species that can potentially immigrate to a local community, thereby influencing its composition and dynamics (Jarzyna et al., 2025). Using 100 km × 100 km grid cells as the operational units at regional scale, we defined the regional species pool as the number of all recorded tree species from all forest vegetation plots within each 100 km grid, i.e. the filtered regional species pool (Cornell and Harrison, 2014). For each one of 143 focal grid cells that were occupied by sampling plots, we used Chao2 richness estimation to redefine regional species pool within each cell (Fig. S2). This is a rarefaction method with nonparametric estimation and can consider the dark diversity (Pärtel et al., 2025) of the species pool (Chao and Chiu, 2016). To reduce differential sampling intensity across grid-cells and better represent the regional species pool, the grid cells with fewer than 5 plots were excluded from the analyses (thus, the number of plots in each grid-cell varied from 5 to 40). To evaluate whether the differences in plot number could affect the estimation of the regional pool, we selected the grids with at least 20 plots, randomly sampled 20 plots for each grid to estimate regional species richness, repeated the same data analyses, and then compared the results with the results using the full dataset. We found that this comparison yielded overall consistent results (Tables S1 and S2). In addition, to test the scale-dependent effects of regional species pool size and judge whether the number of plots per region would lead to a biased estimate of regional species richness, we also used the 50 km and 200 km grid cells to define the regional species pool size and redone analyses at these two scales. Generally, the relative contributions of regional species pool size on local richness and rarity had little change (Table S3). Considering the geographic patterns of mountains and tree distributions in the study region and following the spatial scale used in previous macroecological studies in China (e.g., Huang et al., 2016; Xu et al., 2018) and globally (Enquist et al., 2019; Pärtel et al., 2025), the grid cell of 100 km was considered appropriate for this study.

We used mean annual temperature (TempRegion) and annual precipitation (PrecRegion) as regional climatic variables derived from CHELSA (Karger et al., 2017) from 1979 to 2013 averages at a 30 arc-second resolution. The anomalies of mean annual temperature (TempAnomaly) since the Last Glacial Maximum were used to test the paleoclimate-change hypothesis, using the PaleoClim (http://www.paleoclim.org) dataset at a 2.5 arc-minute resolution (Brown et al., 2018). Elevation range (ElevRange) was used as a proxy of regional habitat heterogeneity (Rahbek and Graves, 2001), based on a digital elevation model at 90 m resolution (http://www.earthenv.org/DEM). We used the elevation data at 90 m resolution to calculate the averages of each 100 km grid, and the differences between the top 5% maximum and the bottom 5% minimum elevations to represent the elevational ranges. Human population density (HUMAN) was used as one regional factor to test the effects of human disturbance. HUMAN (persons per km²) was derived from the History Database of the Global Environment (HYDE3.1) (Klein Goldewijk et al., 2011). All these environmental variables were transformed to the Albers conical equal-area projection and aggregated at a 100 km resolution using the "lets.addvar" function of the R package "letsR".

2.5. Local environmental factors

At the local scale (the site level), the importance of current climate, topography, and soil conditions on local species diversity has been evaluated (e.g., Harrison et al., 2006; Ricklefs and He, 2016). The current climate was derived from the CHELSA at a 30 arc-second (~1 km) resolution (Karger et al., 2017). Due to the study region being affected strongly by East Asian and Indian monsoons, temperature seasonality (TempSeason) and precipitation seasonality (PrecSeason) were selected as local climatic variables. The elevation of each site was the survey value in the field. Slope and topographical position index (TPI) were obtained from the EarthEnv at a 1 km resolution (Amatulli et al., 2020). Soil pH at a resolution of 250 m was extracted from SoilGrids (Hengl et al., 2017). Since site-specific environmental data was unavailable, we performed supplemental analyses of the ranges of temperature and precipitation seasonality at 1 km resolution within each 100 km grid cell, and found that there were huge seasonal variances within all 100 km grid cells (Fig. S1), suggesting that 1 km resolution could well represent local climatic conditions of the studied plots.

2.6. Statistical analysis

We used boosted regression trees (BRT) to partition the independent relative effects of regional and local factors on species richness and rarity on the 2960 forest plots. Since the objective was to use BRT to assess the relative importance of predictors, we did not split the data into training and testing subsets. The learning rate was 0.001, which was ideal shrinkage value to avoid high variance in prediction deviance (Elith et al., 2008). The optimal number of trees (minimizing the prediction bias) was determined through 10-fold cross-validation, a method that effectively reduces overfitting. To increase the model accuracy, randomness was included using a bagging fraction of 0.5. Model performance was evaluated using the cross-validated correlation coefficient, an indicator of predictive accuracy and robustness based on internal cross-validation folds (Elith et al., 2008).

To evaluate the effect of regional species pool size on two diversity measures, we performed BRT models with and without considering regional species pool size, respectively, and compared the relative contribution of all variables between the two models (Jiménez-Alfaro et al., 2018). To evaluate the contributions of predictors, we used the relative influence of the variables and the partial dependence plots (Elith et al., 2008). The relative influence referred to the contribution of each variable and the values sum reaching to 100 (Friedman, 2001). Partial dependence plots (PDPs) were generated to visualize the marginal effect of each predictor variable on the response variable (species richness and rarity) after accounting for the average effects of all other variables in the model (Friedman, 2001). For a given variable x, the partial dependence was computed by averaging model predictions across the entire dataset after systematically varying the focal predictor's values while holding all others constant at their observed values. This procedure allowed interpretation of the functional relationship between the predictor and the response, independent of interactions with other covariates. The fitted values of predictors on the x-axis were on the scale and standardized to the 0–1 range. The partial dependence plots were used to show the relationships of BRTs models fitted to the local species richness and rarity at three spatial scales in Figs. S5–S7.

Structural equation models (SEMs) were used to investigate the direct effects of regional species pool and the direct and indirect effects of environmental factors on species richness and rarity. SEMs enabled the evaluation of hypothesized causal relationships in data sets with more than one dependent variable and the effects of dependent variables on one another (Grace, 2006). We used a piecewise SEM (Lefcheck, 2016), assuming that the response variables in the models followed a Gaussian distribution. Due to the existence of spatial autocorrelation (Moran's I < 0.001 on the residuals of non-spatial multiple regression models, Fig. S3), we fitted spatial simultaneous autoregressive error models in the SEMs (Moran's I > 0.05). Standardized coefficients for SEM paths were used to compare the relative importance of predictors. Chi-square test was used to evaluate the performance of SEM models, and the P value of all models was greater than 0.05. Based on previous studies of regional and local effects on richness (Harrison and Cornell, 2008; Jiménez-Alfaro et al., 2018), we developed a priori theoretical SEMs (Fig. 1) with all predictors to assess the relative importance of regional and local effects.

To reduce multicollinearity, the variables with the absolute value of Pearson correlation coefficient < 0.75 and easier to interpret ecologically were retained (Fig. S4). For all the analyses, we log₁₀-transformed plant richness, soil pH, PrecSeason, PrecRegion, HUMAN, regional species pool, and sqrt-transformed ElevRange and TempAnomaly (Table 1). To eliminate the unit dimensions of predicted variables for comparing their relative importance, all predictor variables were standardized to the 0–1 range. All statistical analyses were carried out using R 3.6.3 software (R Core Team, 2019). The BRTs models were fitted using R package "dismo", SEMs were calculated with R package "piecewiseSEM", and Moran's I value and SARs were calculated using R packages "spdep" and "spatialreg".

3. Results 3.1. Regional drivers of species richness and rarity

The geographical patterns were different between tree richness and rarity (Fig. 2). Local richness decreased with increasing latitude (Fig. 2A), while species rarity showed a clear decline from southwest to northeast except for high rarity in the main island of Taiwan (Fig. 2B). Regional species pool showed higher correlations with richness (Pearson's r = 0.35^***) than rarity (Pearson's r = 0.21^***). The results of BRT (Table 2) and SEMs (Figs. 3 and 4) showed that species pool size was the best predictor for richness, but had a relatively low positive effect on rarity. Among regional drivers, TempRegion was the most important predictor for accurately predicting local species richness, followed by PrecRegion (Table 2; Fig. S6A and B), while HUMAN and TempRegion were the main predictors for species rarity (Table 2; Fig. S6C and D). Including species pool in the models reduced the importance of both regional factors, while the explained variance increased slightly (Table 2).

Regional environment significantly affected local diversity directly (Figs. 3 and 4) and indirectly via the effects on regional species pool (Fig. 4A and B). In detail, regions with warmer temperatures, higher elevational heterogeneity, and greater paleoclimate stability harbored larger species pools, which subsequently supported higher local richness and rarity. Moreover, the effects of regional factors (e.g., TempRegion and TempAnomaly) on local richness were larger than those on rarity. However, human density showed strongly negative correlations with rarity whether species pool was considered or not and had no correlation with richness.

3.2. Local drivers of species richness and rarity

Local environmental factors directly influenced local plant diversity (Table 2; Fig. 3A and B, 4A and B). Specifically, slope positively affected species richness. In contrast to richness, species rarity was strongly influenced by local climatic seasonality. TempSeason and PrecSeason emerged as the most significant predictors, accounting for over 50% relative influence on local species rarity (Table 2; Fig. S6C and D). Furthermore, we determined that plot size can explain approximately 20% and 10% of the total explained variance in species richness and species rarity patterns, respectively. It is important to note that while plot size is a key predictor of local species diversity, particularly in terms of species richness, it does not alter the relative significance of other local and regional environmental factors, such as the size of regional species pools (Fig. 4 and Table 2).

4. Discussion

The regional species pool is a foundational concept for understanding ecological processes that occur between local and extensive spatial scales, shaping local species diversity. However, there is still a lack of precise definitions of regional species pools (Jarzyna et al., 2025) and empirical studies often underestimate its importance due to dark diversity (Pärtel et al., 2025). Generally, regional species pools are the set of species that can potentially immigrate to a local community, influencing local community composition and dynamics (Jarzyna et al., 2025). In this study, we defined the regional species pool as the filtered set of all recorded species from vegetation plots, and explicitly tested its role together with regional and local environmental factors in shaping local tree species richness and rarity across subtropical and tropical forests in China. Our results supported the role of regional species pools and showed that tree richness was mainly shaped by regional factors via positive effects of species pool, whereas rarity was influenced jointly by local climatic seasonality and human disturbance, with weak effects of pool size.

Consistent with our first prediction, the regional species pool strongly and positively affected local species richness, in line with previous studies on Chinese mountain plant richness (Wang et al., 2012), European beech forests (Jiménez-Alfaro et al., 2018), and global fern and lycophyte richness (Weigand et al., 2020). Meanwhile, our results also provide quantitative evidence of the positive linkage between species rarity and regional species pool size, which has been rarely considered. When regional species pool was considered, the total variance explained increased slightly, particularly in species rarity, supporting the weaker effects of regional species pool size and stronger effects of local climatic variation on local species rarity. These findings confirm that the regional species pool serves as a key mediator linking regional and local effects on local richness and rarity.

Our results indicate that regional species pool size contributes to local species rarity, although the effect is relative weak. In Chinese subtropical and tropical forests, longitudinal disparities in regional species pool size, separated by Mt. Xuefeng (~110°E; Fig. S2), are associated with local rarity and fewer endemic or threatened species in eastern forests. In contrast, southwestern regions (e.g., Yunnan and Sichuan), harbor over half of China's vascular plants (Hong and Blackmore, 2015) and characterized by paleoendemic species rarity (Tang et al., 2018). The relatively weak effect suggests that regional species pool size may primarily define the potential upper limit of species richness, rather than directly regulating the occurrence and persistence of local rare species via the absolute size of the regional pool (Cornell and Harrison, 2014). Rare species typically represent a non-random subset of the regional species pool, often characterized by narrow climatic tolerances, and limited dispersal ability, which may reduce their dependence on the size of regional species pool (McGill et al., 2007). After considering regional species pool, local climatic variation directly influences physiological performance, demographic stability, and range occupancy (Thuiller et al., 2005), thereby exerting a more immediate and stronger control over species rarity (Fig. 3 vs. Fig. 4).

As predicted, regional environmental variables shape local tree richness and rarity through regional species pool. The regional species pool is expected to be large when habitat types within grid cells have been more abundant (Zobel et al., 2011, 2016). In Chinese subtropical and tropical forests, warmer regional temperature, greater elevational heterogeneity, and high paleoclimate stability increased regional species pool size, which in turn enhanced local species richness and rarity (Figs. 3 and 4). The uplift of the Himalayas and Qinghai-Xizang Plateau created distinct monsoon systems (Indian monsoon in west vs. East Asian monsoon in east) and contrasting habitat heterogeneity (Chinese second-step topography with the average elevation of 1000~2000 m in west vs. third step with the average elevation < 500 m in east) (Deng et al., 2020). In the western regions, the crisscrossing of mountains generates diverse topographies (Song et al., 2021), accordingly, creating a great variety of habitats and stable climatic conditions to large regional species richness and support rare and relict species' persistence (Hong and Blackmore, 2015). Moreover, paleoclimatic legacies have further reinforced these patterns. While late Cenozoic cooling and Quaternary glaciations drove widespread extinctions of subtropical lineages (e.g., Lauraceae, Fagaceae) in Europe and North America (Martinetto, 2017), southwestern Chinese complex terrain (e.g., Yunnan–Guizhou Plateau, Sichuan Basin) provided long-term refugia for Paleogene-Neogene relict plants (such as Ginkgo biloba, Davidia involucrata, and Metasequoia glyptostroboides), high species richness and endemic species (1429 endemic species in this study, accounting for 43% of the total species, Song et al., 2013; Feng et al., 2016). Consequently, regional climate, habitat heterogeneity, and paleoclimatic stability strongly shaped the regional species pool, increasing local species richness and rarity in subtropical and tropical forests, in line with our second prediction.

In this study, human disturbance reduced rarity more strongly than richness, consistent with the prediction that rare species were disproportionately sensitive to anthropogenic pressure (Xu et al., 2019a, 2019b; Song et al., 2021). Rare or endemic species with small populations were particularly vulnerable in eastern regions, after centuries of agricultural cultivation and rapid urbanization (Song et al., 2021). For example, the average population of eastern regions are 2.8 times that of the western regions, particularly in the regions in Yangtze River Delta and Pearl River Delta Megalopolis. This supports the expectation of second prediction that rarity declines with increasing human activity. As human activity ramps up in the west, more rare or endemic species are facing threatened risk in the Anthropocene. This contrasts with a global study by Enquist et al. (2019) that reported that the regions with rare species are currently characterized by higher human impact. Discrepancy may arise from differences in spatial scale and the ways to measure the rarity. While Enquist et al. (2019) evaluated rarity at the global scale using one-degree spatial grids, our study assessed local rarity at the plot level, with one-degree grids as the regional scale. Furthermore, we calculated rarity with both species' abundance and presence-absence data, while only presence-absence data was used in Enquist et al. (2019). Thus, our measure of local rarity captured not only the distribution range of species occurrence but also species relative abundance that co-occur in the local community.

We documented that rare species were more frequent in areas with relatively low temperatures but high precipitation seasonality, suggesting that they were more sensitive to local climatic fluctuations. This confirmed our third prediction that low local temperature seasonality promote high species diversity, and it affects species rarity more strongly. For the evolution of Chinese subtropical and tropical floras, the East Asian and Indian monsoon circulations create substantial variations of local seasonal temperature and precipitation (Spicer, 2017) and heterogeneous development of evergreen broad-leaved forests (Zhao et al., 2025). To some extent, rare or endemic species are the product of local climatic stability and topographical complexity, such as the previous study on highly rare species of the genus Rhododendron in south-west China (Shrestha et al., 2018). This likely reflects the buffering role of microrefugia in heterogeneous mountains (Dobrowski, 2011). Microrefugia are thought to have allowed rare species to persist, protecting them from perturbation and buffering against seasonal climatic fluctuations, consequently to host a higher frequency of rare species (Morelli et al., 2016). In this study, the local climate conditions were at 1 km spatial resolution, which was beyond the level of vegetation survey. Nevertheless, considering the finer-scale climatic data are currently unavailable in most mountain areas and following the spatial scale used as local climate in previous macroecological studies based on vegetation plots from the global sPlot database (e.g., Bruelheide et al., 2018; Sabatini et al., 2022), the 30 arc-seconds (~1 km) as local climates were well-suited for the current study and provide good performance to explain the local species richness and rarity.

5. Conclusions

This study provides empirical evidence to understand the underlying mechanisms by which large-scale effects shape local species richness and rarity across large spatial scales, particularly in exploring the effects of regional species pool on local species richness and rarity in Chinese subtropical and tropical forests. We conclude that large-scale variations of local tree species richness and rarity can be well explained by regional environmental variables, which in turn affect the size of regional species pool, supporting the view of using regional species pool to explain the determinants of local plant diversity (Jiménez-Alfaro et al., 2018). Furthermore, we document that rare tree species in these forests are associated with diverse regional species pools and low anthropogenic pressures, highlighting the conservation prioritization of rare species in disturbed forest landscapes. We further emphasize that human disturbance intensity at both local and regional scales should be considered and quantitively measured as a key indicator in biodiversity conservation (Ellis et al., 2013), especially for rare and endemic species in highly fragmented subtropical and tropical forests.

Acknowledgements

This work was supported by the National Natural Science Foundation of China (32301401 & 32071538), the Innovation Program of Shanghai Municipal Education Commission (2023ZKZD36), the VILLUM Investigator project "Biodiversity Dynamics in a Changing World" funded by Villum Fonden (grant 16549), the Center for Ecological Dynamics in a Novel Biosphere (ECONOVO) funded by Danish National Research Foundation (grant DNRF173), and the Fundamental Research Program of Shanxi Province (202303021222281).

CRediT authorship contribution statement

Houjuan Song: Conceptualization (equal); Formal analysis (equal); Methodology (equal); Validation (equal); Writing – original draft (equal); Writing – review and editing (equal); Funding acquisition (equal). Borja Jiménez-Alfaro, Yongchang Song, Jens-Christian Svenning, Alejandro Ordonez: Conceptualization (equal); Formal analysis (supporting); Methodology (supporting); Writing – original draft (equal); Writing – review and editing (equal). Oukai Zhang, Xihua Wang, Enrong Yan, KunSong, Luxiang Lin, Shengbin Chen, Qingpei Yang, Buhang Li, Chuping Wu, Bo Jiang, Chao Jin, Zhiming Zhang, Yi Ding, Huilin Wan, Kankan Shang, Kunfang Cao, Wei Shi, Xiaoran Wang, Pengcheng Liu: Data acquisition (equal); Writing – review and editing (supporting). Xin Wang: Data acquisition (equal); Writing – review and editing (supporting); Funding acquisition (supporting). Jian Zhang: Conceptualization (equal); Funding acquisition (equal); Methodology (equal); Writing – original draft (equal); Writing – review and editing (equal); Funding acquisition (equal).

Data availability statement

The data of all studied vegetation plots are available at: https://doi.org/10.6084/m9.figshare.31361269.

Declaration of competing interest

The author Jian Zhang is an Editor for Plant Diversity and was not involved in the editorial review or the decision to publish this article. The other authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.pld.2026.02.002.