Lishunan Yang, Daniel J.Johnson, Zhihun Yang, Xiaohao Yang, Qiulong Yin,Ying Luo, Zhanqing Hao, Shihong Jia,＊

a School of Ecology and Environment, Northwestern Polytechnical University, Xi’an, Shaanxi, 710072, China

b Shaanxi Key Laboratory of Qinling Ecological Intelligent Monitoring and Protection, Northwestern Polytechnical University, Xi’an, Shaanxi, 710072, China

c School of Forest, Fisheries, and Geomatics Sciences, University of Florida, Gainesville, FL, USA

d School of Geography and Tourism, Shaanxi Normal University, Xi’an, Shaanxi, 710062, China

Keywords:Biodiversity Conspecific negative density dependence Dispersal Replicated point patterns Temperate forest Topographic habitat

ABSTRACT Conspecific negative density dependence (CNDD) is a potentially important mechanism in maintaining species diversity.While previous evidence showed habitat heterogeneity and species’dispersal modes affect the strength of CNDD at early life stages of trees (e.g., seedlings), it remains unclear how they affect the strength of CNDD at later life stages.We examined the degree of spatial aggregation between saplings and trees for species dispersed by wind and gravity in four topographic habitats within a 25-ha temperate forest dynamic plot in the Qinling Mountains of central China.We used the replicated spatial point pattern (RSPP) analysis and bivariate paircorrelation function (PCF) to detect the spatial distribution of saplings around trees at two scales, 15 and 50 m, respectively.Although the signal was not apparent across the whole study region (or 25-ha), it is distinct on isolated areas with specific characteristics, suggesting that these characteristics could be important factors in CNDD.Further, we found that the gravity-dispersed tree species experienced CNDD across habitats, while for wind-dispersed species CNDD was found in gully, terrace and low-ridge habitats.Our study suggests that neglecting the habitat heterogeneity and dispersal mode can distort the signal of CNDD and community assembly in temperate forests.

1.Introduction

One of the central questions in community ecology is to understand the processes that promote the plant diversity at the small spatial scales(Sutherland et al., 2013; Wright, 2002).One important mechanism,known as the conspecific negative density dependence(CNND),suggests that plant performance would decline as the density of surrounding conspecific plants increases.The negative effects of conspecifics encourage the raising of rare species,facilitating the species coexistence(Comita and Hubbell,2009;Comita et al.,2010;Hulsmann et al.,2021).There is growing evidence that density-dependent mortality has the potential to stabilize diversity by reducing both seedling and sapling recruitment and survival around conspecific trees through specialized natural enemies (Janzen-Connell hypothesis) (Connell, 1971; Janzen,1970) and intraspecific competition in tropical (Bagchi et al., 2011;Comita et al.,2010)and temperate forest communities(Jia et al.,2020;Murphy et al., 2020).Yet, the strength of ecological processes (e.g.,habitat heterogeneity,dispersal mode,abiotic factors)that hide the true strength of CNDD remains unclear.

The strength of CNDD can vary widely among different species, life histories,and dispersal modes(Johnson et al.,2018;Xu et al.,2022;Zhu et al., 2018).It is found that wind-dispersed species create large spatial clusters compared to gravity-dispersed species(Horn et al.,2001;Savage et al., 2014; Seidler and Plotkin, 2006).Therefore, gravity-dispersed species could suffer strong CNDD due to frequent attack by species specific natural enemies than the wind-dispersed species (Muller-Landau and Adler, 2007; Stump and Comita, 2018; Xu et al., 2022).However,studies on dispersal affects primarily focused on seed and seedings (Bai et al.,2012;Marteinsdottir et al.,2018), and whether species’dispersal mode affects the strength of CNDD at later stages has rarely been examined.

The strength of CNDD can also vary with abiotic factors,which may regulate CNDD effects by changing intraspecific competition or the pressure from natural enemies(LaManna et al.,2016;Song et al.,2020).While most studies focus on abiotic resource availability (Hulsmann et al.,2021;LaManna et al.,2016;Wright,2002),habitat heterogeneity may affect the spatial distribution of available resources within specific topographic habitats(Bagchi et al.,2011;Johnson et al.,2017;Murrell,2009; Pu et al., 2017).Recent studies showed that plants at the sapling stage might have different habitat preference and environmental regulation than other life history stages (Brown et al., 2021; Zheng et al.,2020).In addition,some specific topographic factors,such as elevation,slope and convexity,could also affect the strength of CNDD and maintain plant species coexistence(Song et al.,2020;Xu and Yu,2014;Yang et al.,2022).However, habitats within the same forest affect the strength of CNDD among later life stages(i.e.,saplings and trees)are rarely tested.

Moreover, habitat heterogeneity and species’ dispersal modes can each influence the process of CNDD.For example, topographic factors have reported to affect tree seedling survival (Song et al., 2020), which might cause the variation of CNDD.Other studies have found that gravity-dispersed species have stronger CNDD than wind-dispersed species(Bai et al.,2012;Xu et al.,2022).Yet,the strength of CNDD can vary across different topographic habitats between and within species’dispersal modes.However, studies that simultaneously investigate the effects of species’dispersal mode and topographic habitats on CNDD are rare.

Experiments using pesticides or fences are essential to explore the mechanisms of CNDD and provide a valuable approach for understanding the processes shaping plant population- and community-scale patterns(Bagchi et al., 2014; Jia et al., 2020; Murphy et al., 2020).However,these manipulation experiments were conducted over periods of less than five years,which generally failed to reveal a long-term spatial dynamic,especially in tree species.The analysis of ecological point patterns provides clues of the past process via measuring the degree of spatial aggregation between the offspring(seedlings or saplings)and trees(Bagchi et al., 2011; Getzin et al., 2008; Johnson et al., 2014; Zhu et al., 2013).For example, by analyzing the degree of spatial aggregation between trees and saplings, Bagchi et al.(2011) showed that the spatial point pattern analysis is an important approach to uncover the presence of local CNDD for trees.In addition,the spatial distribution of the focal species in the same habitat can be regarded as the replicates of the same process.Therefore,analyzing replicated point patterns can be used to investigate the degree of spatial aggregation of the same species among different habitats(Bagchi et al.,2015).

Although topographical habitats and species’ dispersal modes could affect the strength of CNDD,they have less been tested simultaneously in previous studies, particularly within the same forest.In this study, we examined the degree of spatial aggregation of saplings and trees, and spatial aggregation of saplings around trees in six different topographic habitats(i.e.,valley,low-ridge,slope,gully,high-ridge,terrace)in terms of four topographic factors (elevation, slope, aspect and convexity) in a warm temperate forest by replicated spatial point pattern (RSPP) analysis.We classified each tree species to either gravity-dispersal or winddispersal categories.We examined, whether species spatial distribution can signal the dispersal trait.Also, we examined, whether species’dispersal trait and species-topographic habitat association limit the spatial patterns.We tested three hypotheses: (1) Habitat heterogeneity will mask the signal of CNDD at the community-scale due to speciestopographic habitats association.(2) The strength of CNDD varies between species, but consistent within different dispersal modes.(3) The local spatial patterns of species,their saplings,trees and saplings around trees are critically related to species-topographic habitat association and dispersal traits.

2.Material and methods

2.1.Study site and data collection

Our study was conducted at the Qin-Ling Huang-Guan Forest Dynamics Plot(QLHG FDP)within the Changqing National Nature Reserve(33°32′21′′N, 108°22′26′′E).This reserve is located on the southern slope of Qinling Mountains in central China.The average annual temperature of the study area is 12.3°C, the annual precipitation is 908.0 mm,mostly as rain from July to September.The soil type is brown loamy soil.And the mean pH of the soil is 5.94.The vegetation is dominated by the warm temperate deciduous broad-leaved forest.Dominant trees include Quercus aliena var.acutiserrata, Fraxinus chinensis, Carpinus turczaninowii and Cornus kousa subsp.chinensis.

The QLHG FDP was established in 2019.Following the Forest Global Earth Observatory (ForestGEO) census protocol (Condit, 1998), we divided the QLHG FDP into 625 subplots of 20 m × 20 m using the Electronic Total Station (South Surveying & Mapping Instrument Co.,Ltd., NTS-352R8).All woody plant individuals with diameter at breast height (DBH) ≥1 cm in every subplot were tagged, mapped, and identified to species.We also recorded the DBH for every individual(He et al.,2022).

In this study, we classified every tree individual into either winddispersed or gravity-dispersed species category according to the description of the online database Flora of China (Institute of Botany,Chinese Academy of Sciences, 2008).We selected two most dominant species from the wind-dispersed species (i.e., Fraxinus chinensis and Carpinus turczaninowii) and the gravity-dispersed species (i.e., Quercus aliena var.acutiserrata and Cornus kousa subsp.chinensis).We limit our species-level analysis to four widespread distributed species(two gravity and two wind dispersed species) because the other species are habitat specialist, and their spatial distribution is limited to one or two specific habitats.

2.2.Habitat division

Within the 25-ha QLHG plot,we used the Electronic Total Station to measure the elevation of the four corners at a scale of 20 m × 20 m subplot.Based on the elevation data, the mean elevation, slope, convexity,and aspect were calculated at the 20 m×20 m scale.We defined the elevation as the mean elevation across four corners for each subplot(Valencia et al.,2004).We quantified convexity as the elevation of a focal subplot minus the mean elevation of the eight surrounding subplots(Song et al.,2020).We calculated the slope for each subplot as the mean angle that each of the four triangular planes created by connecting three of its adjacent corners deviates from the horizontal(Harms et al.,2001).We then quantified the average value of the angles among these four planes and the projection plane of the plot as the slope (Harms et al.,2001), and the aspect was calculated from the average of the angles between these four planes and the due north direction (Zuleta et al.,2020).For subplots at the edge of the 25-ha plot, we calculated the convexity as the elevation of the center point minus the average of four corners(Valencia et al.,2004).

We classified the 20 m×20 m subplots according to their topographic characteristics (hereafter called “topographic habitat”).A common approach is to perform hierarchical clustering through topographic factors and divide habitats according to clustering tree (Altman and Krzywinski,2017).We used Ward’s minimum variance method(Zuleta et al.,2020) of hierarchical clustering to divide all subplots into six habitats(Fig.1b).

2.3.Point pattern analysis

We classified each stem as either tree or sapling according to its DBH(Table S1).In addition,saplings were divided into three categories(i.e.,large, medium and small) according to the DBH distribution of species(see Supplementary Materials for details).

The method of pattern analysis has been widely used in seeking ecological processes, among which the most widely used is Ripley’s Kfunction and pair-correlation function(PCF)(Bagchi et al.,2011;Diggle,2013; Ramón et al., 2016; Ripley, 1976, 1977; Wiegand and Moloney,2004).The accumulative K-function detects aggregation or dispersion

Fig.1.Topography and the subdividing habitats within the Qin-Ling Huang-Guan (QLHG) plot.(a) Three-dimension topographic map of the QLHG plot; (b) Six habitats within the QLHG plot which were classified via the hierarchical clustering.Gray lines and numbers in the graph are elevations.

within circles of a given radius r (Ripley, 1976, 1977; Wiegand and Moloney,2004),while replacing circles with rings in Ripley’s K-function results in the PCF (Ripley, 1981; Stoyan and Stoyan, 1994).The K-function is cumulative and retains some small-scale effects at larger scales(Condit et al.,2000),however,using rings in the PCF allows for the isolation of specific distance classes (Wiegand and Moloney, 2004).We used the method for analyzing replicated point patterns with the isotropic edge correction method (Bagchi et al., 2015; Ramón et al.,2016).We used Ripley’s K-function (Ripley, 1976), and bivariate pair-correlation function to calculate second-order spatial point process,which were widely used for species spatial distribution analysis(Bagchi et al., 2011; Brown et al., 2011; Ramón et al., 2016; Wiegand et al.,2007).The K-function is usually simplified to:

which is a standardized version of K-function(Besag, 1977), where L(r)= 0 indicates the pattern follows spatial randomness (CSR) within distance r,L(r)＞0 indicates aggregation and L(r)＜0 indicates regularity.

The PCF is a derivative of Ripley’s K-function (i.e.,(Diggle,2013;Illian et al.,2008).Compared with the Ripley’s K-function,PCF is a non-cumulative function,which is convenient for the choice of null model (Stoyan and Stoyan, 1996).We used the bivariate PCF to represent the spatial relationship between conspecific trees and saplings.When g12(r) = 1, it means that the spatial distribution of the saplings(denoted by 2) around the adults (denoted by 1) follows the complete spatial randomness (CSR).g12(r) ＜1 indicates mutual inhibition of saplings around trees, and g12(r) ＞1 indicates clustering of saplings around trees.

To evaluate the attraction or inhibition relation between trees and saplings using PCF,we used an Antecedent Conditions(AC)model(i.e.,the locations of saplings can be randomly generated while the locations of trees are fixed)and calculated the null model from the fifth-lowest and fifth-highest values of 99 simulations(Wiegand and Moloney,2004).We used a common distance of 50 m for the pair correlation function(Johnson et al.,2018;Wiegand et al.,2007).Randomization of saplings uses likelihood cross-validation to select a smoothing bandwidth for the kernel estimation of point process intensity(Loader,1999).Specifically,we examined the spatial patterns of saplings around the trees of two dispersal categories.Further,we adopted the same approach for the two dominant species in each category.To verify whether the spatial pattern of the two categories is dominated by the two dominant species, we excluded the two dominant species from each category and redid the analyses.

2.4.Replicated point pattern analysis

Replicated point pattern analysis is like the single point pattern approach, in addition to requiring multiple plots to provide independent replicates.Compared with the single point pattern approach, the replicated point pattern analysis considers the distribution of pair distance among multiple patterns.Therefore, although the replicate of two individuals only provides a pair of distances,the single pair distances can be combined with other point patterns for useful inference (Bagchi et al.,2018).This allows small regions to be included in the analysis, while reducing their impact on the overall inference relative to having more replicate data(Bagchi et al.,2015).Therefore,several subplots can provide information like that of a single large plot.In addition, replicated point pattern analysis can be used to analyze inhomogeneous processes,where the intensity of points is different throughout the study region.A single point pattern analysis that regards the process as homogeneous will not distinguish between clustering and inhomogeneous (Diggle, 2013; Law et al., 2009).If the pattern is divided into multiple sub-regions and analyzed separately,a local spatial structure independent of habitat-scale heterogeneity can be obtained(Illian et al.,2008;Law et al.,2009).

Replicated point pattern analysis allows each sample to contain fewer points than the single point pattern analysis(Bagchi et al.,2015;Diggle et al.,2000).Therefore,uncertainly of the results is high.However,as the number of points increases, the pooled function becomes smoother and the width of the confidence interval decreases(i.e.,results become more robust)(Bagchi et al.,2015).We initially ensured an approximately equal or similar number of replicates in each habitat and then established a minimum requirement of 8 individuals per replicate for the analysis.A 20 m × 20 m replicate accommodates too few tree individuals.The replicate of 80 m×80 m contains more individuals,but it lacks sufficient replicates per habitat for the meaningful analysis (Table S2).While results are consistent between the 60 m×60 m and four randomly selected 40 m×40 m in each habitat(Figs.S1,S2 and S4),the later includes less tree individuals.Taken together,we ultimately selected the 60 m×60 m for analysis as they offer sufficient number of tree individuals and replicates.

We used the L-function(Eq.1)to analyze RSPPs,which included two dispersal mode categories and two dominant tree species,and the pooled of other tree species.We chose four replicates in each topographic habitat and analyzed the spatial pattern at a distance of 0-15 m,to focus on the most sensitive scales of CNDD in saplings (Bagchi et al., 2018; Hubbell et al.,2001).We used bootstrapping to simulate 999 K-functions for the null model, because the semi-parametric bootstrapping is a suitable method to estimate confidence intervals for parameter estimation and prediction(Bagchi et al.,2015;Diggle et al.,1991;Landau et al.,2004).

To detect whether the degree of spatial aggregation of saplings decrease with increasing of DBH,we divided the saplings into three DBH classes: large, medium and small.We analyzed the degree of spatial aggregation of three DBH class saplings using replicated point pattern analysis.We used the bootstrapping to resample the large trees to calculate the confidence interval.To test the interaction between the dispersal mode and the habitat,we used two-way ANOVA-like method to analyze replicated point patterns(Ramón et al.,2016).

All spatial analyses, simulations and statistical analyses were done using the “spatstat” package (Baddeley and Turner, 2005) and “replicatedpp2w” package (Ramón et al., 2016), and the plotting has done using the “ggplot2” package (Wickham, 2016) in the R 4.1.0 (R Development Core Team,2021).

3.Results

3.1.PCF for the overall study area

Across the whole study area,there was no evidence for an interaction between trees and saplings for gravity-dispersed species(Fig.2a).Among the two dominant species of gravity-dispersed, Quercus aliena var.acutiserrata had an aggregated distribution between trees and saplings at small scale (4-7 m) (Fig.2b), and Cornus kousa subsp.chinensis had an aggregated distribution at distances under 4 and 5-6 m (Fig.2c).After removing the two dominant species, the gravity dispersal category was still clustered at small scales (0-2 m) (Fig.2d).Wind-dispersed species had no interaction between trees and saplings (Fig.2e).However, the two dominant species are aggrgated at small scales (i.e., 0-5 m for Fraxinus chinensis and 0-11 m for Carpinus turczaninowii, Fig.2f and g).After removing these two dominant species, all other wind-dispersed species was still present in aggregations at small scales(3-5 m)(Fig.2h).

3.2.RSPP between species’ dispersal categories

Consistent with the prediction of CNDD,the degree of spatial aggregation of trees was generally lower than that of saplings for gravitydispersed species.Notably, these patterns were similar across all four topographic habitats(Fig.3).For wind-dispersed species,the pattern that significantly lowers degree of spatial aggregation of trees than saplings was only observed in three habitats (i.e., gully, terrace and low-ridge)(Fig.3).In slope habitat, however, trees were generally no evidence of difference compared to saplings.In addition,sensitivity analysis showed that gravity-dispersed species were still aggregated after removing two dominant species(except for the low-ridge habitat, Fig.S6).

3.3.RSPP at the species level

Fig.2.Bivariate intraspecies analysis of the two categories by the antecedent conditions (AC) null model for gravity-dispersed species (left column) and winddispersed species (right column), respectively.The pattern observed outside the envelope represents a significant deviation from the AC model.The dashed lines indicate the intersection of the value of g12(r) with the envelope.

Fig.3.The L-functions summarizing the degree of spatial aggregation patterns between trees and saplings for the gravity-dispersal and wind-dispersal among four habitats (i.e., gully, low-ridge, slope and terrace).The spatial pattern of gravity dispersal (top row) and wind dispersal (bottom row) at the scale of 60 m × 60 m,respectively.The red lines represent the L-function of the trees.The gray 95% confidence interval is calculated by re-sampling the saplings.Light yellow represents spatially aggregated, while light green represents spatially dispersed.

For the four dominant species, the results of RSPPs analysis for saplings around trees was generally inconsistent with the CNDD process(Figs.S3 and S5).In low-ridge habitat, only the Cornus kousa subsp.chinensis was consistent with the spatial aggregation process of CNDD(i.e., saplings were generally more aggregated than trees) (Fig.S3f),whereas the opposite was true for the other three dominant species(Figs.S3 and S5).In slope habitat,only Fraxinus chinensis was consistent with the spatial aggregation process of CNDD (Fig.S5c), while the opposite was true for the other three dominant species(Figs.S3 and S5).In the gully habitat,the four dominant species generally did not exhibit CNDD(Figs.S3 and S5).And in terrace habitat,only Quercus aliena var.acutiserrata was consistent with spatial aggregation of CNDD.In addition,after removing dominant species, we found other gravity-dispersed species showed the signal of CNDD in three habitats (e.g., low-ridge,slope, and terrace).However, the spatial pattern of other winddispersed species was only compatible with CNDD in the terrace habitat(Fig.S6).

3.4.Variation of RSPP across DBH classes

In the gravity-dispersed species,small saplings had the highest spatial aggregation at overall distance in gully and slope (Fig.4a and c).However, there was no obvious decreasing trend of spatial aggregation with the increase of DBH in wind-dispersal category (Fig.4).Spatial aggregation of small saplings was generally higher than trees in the four habitats(Fig.4a-h).

3.5.Interaction between dispersal mode and habitat

The replicated point pattern analysis showed that there is no interaction between the dispersal mode and the habitat in trees and saplings(Tables 1 and S3),but the wind-dispersed species had stronger clustering patterns than the gravity-dispersed species (Figs.5 and S7).The winddispersed species showed clustered patterns in all four focal habitats,but the gravity-dispersed species were relatively randomly distributed(Fig.5).

4.Discussion

Fig.4.The L-functions summarizing the degree of spatial aggregation patterns between trees and saplings for the gravity-dispersal and wind-dispersal among four habitats (i.e., gully, low-ridge, slope and terrace).The spatial pattern of gravity dispersal (top row) and wind dispersal (bottom row) at the scale of 60 m × 60 m,respectively.The dark-gray, red, cyan, and blue lines represent the L-function of all, large, medium and small DBH classes of saplings, respectively.The gray 95%confidence interval is calculated by re-sampling the trees.Light yellow represents spatially aggregated,while light green represents spatially dispersed.Note that here the envelope was calculated from trees and the value of the L-function was calculated from saplings with three size classes.

Table 1Replicated point pattern analysis of dispersal modes, habitats and interactions between dispersal modes and habitats for trees.BTSS: sum of squared differences.P-value simulates 999 K-functions by bootstrapping of the residual functions.To calculate the BTSS,we used K(r)functions estimated from r=0 to r=15 m, at intervals of 0.1 m.

The strength of CNDD can vary greatly with environmental heterogeneity and species’ dispersal mode.The spatial analysis is a common approach to investigate the signal of CNDD via checking the degree of spatial aggregation (lower degree of spatial aggregation of trees compared to saplings)(Bagchi et al.,2011).We show that,at the scale of the plot, both gravity-dispersed and wind-dispersed tree species were clustered spatially at short distances (＜5 m).However, at the scale of topographic habitat,the degree of spatial aggregation between trees and saplings were generally consistent with process of CNDD for both dispersal modes across habitats.In addition, the gravity-dispersed trees suffered strong CNDD than the wind-dispersed trees (Figs.3 and 5),which is consistent with a previous study in another temperate forest(Bai et al., 2012).These findings potentially confirmed the critical role of CNDD in maintaining coexistence of species.

Widespread evidence shows that CNDD exists for plants at the seedling stage via observing the reduction of plant performance near high densities of conspecific trees (Bai et al., 2012; Jevon et al., 2022).Although these studies showed that CNDD can affect the dynamic of many plant species or populations (Brown et al., 2019; Jansen et al.,2014; Jia et al., 2020), the strength of CNDD varies greatly among different life stages (LaManna et al., 2016; Zhu et al., 2018).While a previous study investigated spatial patterns of saplings and juveniles,they did not statistically test the differences between different life stages(Piao et al.,2013;Yao et al.,2020).In this study,we found larger saplings generally showed a weaker degree of spatial aggregation than smaller ones (Fig.4), which is consistent with the process of CNDD.The earlier stages of plants(e.g.,smaller saplings)may be more sensitive to pressure of natural enemies (Hulsmann et al., 2021; Zhu et al., 2018) or competition for resources (Comita and Hubbell, 2009; Wright, 2002)than those at later stages (e.g., larger saplings), which potentially generate the pattern we observed here.This study highlights that the stage of saplings is also a critical period for recruitment and future forest community structure.

Recent studies have found that the strength of CNDD can also vary greatly among species with different dispersal modes (Lu et al., 2015).Although we find there is no interaction between the dispersal mode and habitat, our results indicated that the strength CNDD for gravity-dispersed species could be stronger than wind-dispersed species,which is in line with some recent studies (Xu et al., 2022; Zheng et al.,2020).In addition,our results suggest that CNDD processes become more complex in forests with a higher degree of heterogeneity, which is consistent with a recent study showing that the strength of CNDD varies across habitats(Song et al., 2020).Additionally, our findings show that two dispersal categories experienced different degrees of CNDD in different topographic habitats, possibly due to differences in the characteristics of habitats may have led to variations in the seed dispersal patterns(Parciak,2002),which ultimately result in different strength of CNDD.Interestingly, the two most dominant species were unlikely to drive these overall spatial patterns (Figs.3, S3, S5 and S6).Our study suggests that tree species’ dispersal mode may have a long-term impact on the spatial distribution and community structure.We propose that more individuals included in the spatial analysis for pooling all the same dispersed species could potentially increase the ability to detect the signal of CNDD(i.e.,the higher degree of spatial aggregation for saplings compared to trees) (Bagchi et al., 2015).In addition, spatial analysis is widely used to detect CNDD, however, we also acknowledge that such spatial patterns can also generate due to other processes (e.g., dispersal limitation) (Zhang et al., 2020).Indeed, such observational approaches should be combined with manipulation experiments(Bagchi et al.,2014;Jia et al., 2020) to further explain the mechanisms that cause CNDD.

Fig.5.The degree of spatial aggregation of trees in the four habitats due to gravity-dispersed and wind-dispersed tree species at the scale of 60 m × 60 m.The Lfunction values were calculated from four replicated plots in each habitat, for r = 0 to r = 15, with 0.1 m intervals.Error bars indicate standard errors, and nonparametric comparison of different values is represented by asterisks (＊＊＊P ＜0.001).Light yellow represents spatially aggregated, while light green represents spatially dispersed.

Recent studies found that the strength of CNDD varied among habitats (Johnson et al., 2017; Song et al., 2020).In this study, the gravity-dispersed and wind-dispersed species showed no evidence of CNDD at the whole study area.After dividing the whole plot into different topographic habitats, we found CNDD existed both gravity-dispersed and wind-dispersed species in specific habitats (e.g.,gully, low-ridge and terrace habitats), but not in the other habitat(Fig.3).Together, these results suggest that including habitat characteristics is critical to reveal the real spatial patterns(Bagchi et al.,2011;Jara-Guerrero et al., 2015).Notably, gravity-dispersed species showed CNDD across all habitats, while CNDD existed in gully, low-ridge and terrace habitats for wind-dispersed species.We suspected that plants might suffer stronger intraspecific competition in the gully habitat and low-ridge habitats because these habitats potentially have abundant resources(LaManna et al.,2016),although the pattern is consistent across habitats for gravity-dispersed species.According to the storage effect,intraspecific competition becomes more intense when the environment favors the focal species (Chesson, 2000).In this case, resource-rich habitats often promote intraspecific competition, resulting in CNDD.Despite we found the strength of CNDD varied among topographic habitats, other habitat-associated variables, such as, light conditions and micro-climate may be also important in regulating the strength of CNDD(Song et al.,2020;Xu et al.,2022;Yao et al.,2020).While more similar studies should be conducted in other forests, our study highlights that considering the fine-scale habitat heterogeneity in mediating the strength of CNDD is important in natural forests,particularly for species with distinct dispersal modes.

In this study, both wind- and gravity-dispersed tree species exhibit apparent CNDD in specific topographic habitats.Notably,wind-dispersed tree species show a more pronounced clustering pattern than gravitydispersed species.These findings will be informative for forest managers or owners who want to improve the regeneration specific tree species by considering the seed-dispersed type and the topographic characteristics.Specifically, to allow more saplings to establish in the forest,the logging intensity of conspecific adult trees should vary across species with different seed-dispersed modes and among distinct topographic habitats.

5.Conclusion

While the traditional single-point pattern analysis conducted at a whole community-scale plot is widely used to detect the signal of CNDD,this approach ignores the variation within the plot(Condit et al.,2000),particularly in the montane forest covered on a heterogeneous landscape.Using the recently developed replicated point pattern analysis (Bagchi et al., 2018; Ramón et al., 2016), our study suggests that the habitat heterogeneity within a forest should take into consideration in predicting the strength of spatial patterns and CNDD.Meanwhile, results also showed that the strength of CNDD may vary with species’dispersal mode and life stages.Overall, our study highlights that considering habitat heterogeneity and species’dispersal mode is critical in understanding the spatial patterns and CNDD processes of plants in natural forests.

Authors’contributions

Shihong Jia and Lishunan Yang conceived the idea and designed the research;Shihong Jia,Lishunan Yang,Zhanqing Hao,Xiaochao Yang and Qiulong Yin collected the data; Lishunan Yang and Zhichun Yang conducted the data analyses; Lishunan Yang, Shihong Jia, and Daniel J.Johnson wrote the first draft.All authors contributed to the final manuscript.

Availability of date and materials

The datasets used and generated from this study are available from the corresponding author on reasonable request.

Competing interests

The authors declare no conflict of interest.

Acknowledgments

We thank the field workers who collected data in the Qin-Ling Huang-Guan 25-ha forest dynamics plot.We are grateful to Zikun Mao for his valuable comments and suggestions in data analysis.Shihong Jia was financially supported by the National Natural Science Foundation of China (Grant No.32001120), and the Fundamental Research Funds for the Central Universities (Grant No.31020200QD026).Qiulong Yin was supported by the National Natural Science Foundation of China (Grant No.32001171).Ying Luo was supported by the Innovation Capability Support Program of Shaanxi(Grant No.2022KRM090).

Appendix A.Supplementary data

Supplementary data to this article can be found online at https://doi.i.org/10.1016/j.fecs.2023.100139.