Fluctuations in Species Diversity in Evergreen Broad-Leaved Forests and Changes in Their Co-Occurrence Network

Abstract

Understanding the population dynamics and interspecific interactions in subtropical forests is crucial for uncovering the underlying mechanisms of species coexistence and community stability. Two censuses were conducted between 2018 and 2023 in a 9.6 ha subtropical evergreen broad-leaved forest dynamics plot situated in Mount Wuyi, southeastern China. Utilizing co-occurrence networks and long-term data, we examined the relationship between species interactions and their contributions to community assembly. Our findings reveal that high mortality rates among small-diameter individuals have created ecological niches, facilitating the establishment of 12 new species between 2018 and 2023. A generalized linear mixed-effects model showed positive relationships between sapling abundance and conspecific neighbor density. Co-occurrence networks demonstrated a shift toward higher positive interactions but reduced modularity, indicating a more integrated yet less stable community structure. Despite their low abundance, rare species demonstrated significant roles in network connectivity and stability, underscoring their status as keystone species. Additionally, the significant correlations between topographic factors and species richness highlighted the role of environmental filtering in shaping community composition. Our findings contribute to a deeper understanding of subtropical forest community dynamics, emphasizing the importance of long-term monitoring to unravel the complex interactions between populations and their environmental conditions. This study represents the first long-term observational experiment conducted in a subtropical secondary forest, providing valuable insights into the dynamics of forest community assembly in this region.

1. Introduction

One of the key objectives in community ecology is to understand how to prevent competitive exclusion and drift among species [1]. These two ecological processes, which weaken species coexistence, drive changes in communities over time and space, ultimately leading to dominance by only one or two species [2,3]. However, low-latitude forest ecosystems continue to maintain high species diversity and stable coexistence [4]. One explanation is that species dispersal, migration, and speciation maintain tree richness in neotropical forests [5,6]. Indicators of these neutral processes require long-term monitoring to be accurately obtained. Therefore, it is crucial to conduct long-term forest community observations to examine population fluctuations within forest species and to effectively characterize the dynamic behaviors of complex low-latitude forest ecosystems. In this study, we demonstrate the importance of large sample sizes at the local scale and in the short term for detecting fluctuations in populations.

Demographic and environmental variance are considered two critical factors that drive changes in local-scale population abundance, ultimately influencing community assembly [7,8]. Mortality rates, recruitment rates, and population fluctuations are direct indicators of population dynamics. Long-term monitoring in forest dynamics plots provides valuable insights into these dynamics, enabling predictions about future community trajectories. A study based on ForestGEO data demonstrated that the dynamics of rare species are predominantly governed by demographic stochasticity, while those of common species are largely shaped by environmental variability [9]. Rare species, characterized by low genetic diversity, are more susceptible to drift [10]. In contrast, common species, due to their larger population sizes, are more likely to be influenced by environmental factors and specialized pathogens [11,12]. Under stable environmental conditions, a widely accepted consensus is that conspecific negative density dependence (CNDD) limits the population growth of common species, thereby allowing rare species to gain a competitive edge [13]. However, during periods of extreme heat, such as from 2018 to 2023 [14], CNDD may not be the sole significant factor affecting population fluctuations. Storage effects or temporal niche partitioning may also play critical roles, potentially leading to a lagged response in population changes and enabling prolonged coexistence within the community [15]. For example, soil moisture content mediates stronger CNDD effects on seedlings, exerting a lasting impact on seedling diversity [16]. However, CNDD tends to weaken under drier conditions [17]. Therefore, in the context of global warming, the long-term monitoring of population dynamics is crucial for advancing our understanding of community assembly processes. Given the CNDD framework, we hypothesize that saplings of more abundant species have a higher number of conspecific neighbors (H1). This is because tree species with higher abundance tend to produce a larger number of seeds, leading to dense seedling clusters around parent trees, and thereby resulting in a higher density of conspecific neighbors among their seedlings [13].

The influence of one plant individual on another, whether of the same or a different species, is considered plant interaction [18]. Interactions among species are considered widespread in natural communities, driven by the relative scarcity of space and resources [19]. For instance, a decline in the network-forming potential of ectomycorrhizal (EcM) fungi and increased competition from non-EcM plants can alter the survival rates and nutritional status of neighboring pine and spruce seedlings [20]. Conversely, mutualistic dependencies are also observed, such as leguminous plants benefiting from symbiosis with nitrogen-fixing rhizobia, which are strongly promoted by heterospecific legume genotypes. However, this symbiosis may be reversed by silicon content [21]. These examples suggest that relationships of both cooperation and competition likely exist at local scales. Identifying and understanding such relationships are crucial for unravelling the mechanisms of stable coexistence within communities. Various network-based approaches have been developed and extensively applied in microbiome biology to uncover interactions among organisms, including fungi and bacteria [22]. Unlike microbial interactions, which can be directly inferred through high-throughput sequencing with co-occurrence patterns, plant ecological networks are influenced by multiple, often unobservable, factors, such as soil composition, herbivory, and seasonal climate variations. Furthermore, many species interactions—including competition, mutualism, and symbiosis—do not always leave distinct spatiotemporal signals, making network inference particularly challenging [23]. However, community-scale network studies in large plant and animal ecological systems remain exceptionally rare. This gap significantly hinders our understanding of subtropical secondary forest community stability. In subtropical regions, ectomycorrhizal (EcM) species have been demonstrated to dominate evergreen broad-leaved forests. However, species richness is low compared to arbuscular mycorrhizal (AM) species (EcM/AM ratio of 17:145 in 2023) [24,25]. Considering that ectomycorrhizal (EcM) fungi establish stable symbiotic relationships with plants, they serve as key guardians of terrestrial ecosystems [26]. Thus, we hypothesize that mutualistic interactions play a crucial role in maintaining the stability of subtropical evergreen broad-leaved forest communities. This hypothesis is reflected in co-occurrence networks, where the abundance relationships between most EcM species exhibit positive correlations (H2).

“Everything is everywhere, but the environment selects” is one of the classic theories in ecology. While species possess broad potential dispersal capacities, environmental variance determines which species can adapt to specific local conditions and sustain their populations. Environmental selection not only promotes species coexistence by enhancing ecological niche differentiation but also influences regional plant assembly and diversity through processes like environmental filtering, such as soil acidification-driven long-term vegetation succession [27]. A study on the elevational gradient in subtropical forests revealed that topography and microclimatic variables are the primary factors driving phylogenetic structure variation [28]. Meanwhile, local species abundance mediates deterministic processes that influence species composition along the altitudinal gradient [28]. Since conspecific individuals exhibit similar responses to environmental factors, such as extreme climate, water availability, and resource fluctuations, environmental variation is correlated among individuals within a species. As a result, environmental variation can have significant impacts on both large and small populations [9]. However, our understanding of how microhabitats shape species population structures remains limited. While this hypothesis is theoretically supported, there is a lack of large-scale empirical studies systematically investigating the relationship between species population abundance and microhabitat heterogeneity. This gap limits our understanding of how species populations respond to environmental heterogeneity and hinders deeper exploration of the mechanisms underpinning stability in more complex ecosystems. Consequently, we further hypothesize that dominant species in subtropical secondary forests are more susceptible to selective processes, which are accompanied by more pronounced population fluctuations in these dominant species (H3).

Mt. Wuyi, located in southeastern China, harbors a representative mid-subtropical forest ecosystem with distinct vertical vegetation zonation, encompassing nearly all vegetation types found in China’s mid-subtropical region. Benefiting from its exceptional ecological environment and unique geographical location, Mt. Wuyi has served as a natural refuge for numerous plant and animal species during geological evolution, making it a hotspot for biodiversity with extraordinarily rich species resources [29]. Recognized as one of China’s biodiversity hotspots, Mt. Wuyi offers invaluable samples for diversity research and serves as a key area for exploring ecosystem stability and species coexistence mechanisms. Conducting long-term monitoring and research in this region, particularly systematic observations of dynamic changes in forest ecosystems, is thus of significant importance. Building on this, we carried out a decade-long study in the subtropical secondary forests of Mt. Wuyi. This study primarily aimed to identify the long-term dynamics of forest community and explore the potential mechanisms driving interspecific interactions, providing theoretical insights into the ecological processes and species maintenance mechanisms of mid-subtropical forests.

2. Materials and Methods

2.1. Site Description

Mt. Wuyi harbors the most complete, typical, and largest subtropical forest ecosystem at this latitude, globally [24]. It is a UNESCO World Natural and Cultural Heritage Site [24]. In 2021, the Chinese government established Wuyishan National Park to protect the regional biodiversity. The area has a mid-subtropical monsoon climate. Between 1961 and 2018, the average annual temperature ranged from 18.4 °C (1984) to 20.3 °C (1998), with a multi-year mean of 19 °C [30]. Annual precipitation varies between 1690 and 2630 mm, with localized areas receiving over 3000 mm [31].

2.2. Plot Survey

Following the survey guidelines established by the Smithsonian Tropical Research Institute’s Center for Tropical Forest Science (CTFS) [32], we established a 9.6 ha forest dynamics plot (WYS, 27°35’24″ N, 117°45’55″ E) in Wuyishan National Park in 2013 (Figure S1) [24]. The plot, measuring 400 m × 240 m, was surveyed using a total station and divided into 240 quadrats of 20 m × 20 m each. Each 400 m² quadrat was further subdivided into 16 subplots of 25 m² [33]. Permanent concrete markers were installed at the 20 m intersection points, and PCR tubes were placed at the 10 m line intersections. During the initial census, the boundaries of the 20 m × 20 m quadrats were permanently marked. Grid lines were stretched to delineate these boundaries based on the permanent concrete markers. Each 20 m × 20 m quadrat was further subdivided into 5 m × 5 m smaller plots using midpoint markers on the 20 m and 10 m grid lines. Elevation was recorded at the 5 m intersection points, and slope, aspect, and convexity were derived from the elevation data [34].

We conducted a comprehensive survey of WYS from September to December 2013. Two censuses were conducted in 2018 and 2023 to remeasure the plot. The plant species recorded during the survey were cross-referenced with the List of Plant Species in China (2023 edition) to ensure accurate identification. This validation process was conducted using the LPSC R package [35]. All trees with a diameter at breast height (DBH) ≥ 1 cm within the plot were identified, tagged, mapped, and measured for DBH. Every five years, marked trees were re-measured to assess their survival and growth. Additionally, all newly recruited individuals with a DBH ≥ 1 cm were identified, tagged, mapped, and measured [36]. To ensure that the DBH measurements were consistent across surveys, paint was applied at the measurement points.

2.3. Division of Life Stage

To test the CNDD experienced by sapling trees, we categorized species within the plot based on their DBH (diameter at breast height) to reflect different life stages. The DBH is a direct proxy for tree age, as larger DBH values generally correspond to older individuals within the same species. In this study, the DBH was used to define age classes, a common approach in community ecology research at local scales [37,38,39]. For each species, we classified individuals using the 99th percentile of the DBH (DBH₉₉): individuals with a DBH ≤ DBH₉₉½ were defined as juvenile trees, those with a DBH ≥ DBH₉₉⅔ were considered adult trees, and individuals with a DBH₉₉½ < DBH < DBH₉₉⅔ were not further subdivided to ensure accuracy in size class delineation [40]. Species with fewer than 10 individuals in the plot were excluded from age class categorization. This approach accounts for interspecies variation in the maximum DBH during growth, reducing the risk of misclassifying species’ life stages.

2.4. Data Analysis

To identify dominant species and their dynamics in the WYS plot, we calculated importance values (IVs) using the following formula: IV (%) = (Relative Abundance + Relative Frequency + Relative Dominance)/3. For the dominant species identified in 2023, we calculated their recruitment rate (R), mortality rate (M), and population size change rate (λ) as follows:

$$M={\displaystyle \frac{\ln n_0-\ln S_t}{T}}$$

(1)

$$R={\displaystyle \frac{\ln n_t-\ln S_t}{T}}$$

(2)

$$\lambda ={\displaystyle \frac{\ln n_t-\ln n_0}{T}}$$

(3)

Here, n₀ and n_t are the population sizes at the first and the t-th census, S_t is the number of surviving individuals at the t-th census, and T is the time interval between two censuses [41].

To compare species diversity across 2013, 2018, and 2023 under a unified framework, we employed Hill numbers, which incorporate species richness, diversity indices, and evenness into a single formula, providing a flexible perspective for diversity measurement:

$${}^{\mathrm{q}}D={\left({\displaystyle \sum _{i=1}^S{p}_i^q}\right)}^{1/(1-q)}$$

(4)

where ^qD is the Hill number of order q, S is the total number of species (richness), p_i is the relative abundance of species i, and q adjusts the sensitivity to rare and common species. These calculations were performed using the SpadeR package [42]. One main advantage of the Hill number is that it follows the replication principle, allowing for direct comparisons of different diversity metrics at the same scale [43]. The sensitivity of Hill numbers to rare and dominant species can be adjusted by the value of q, allowing the dynamic changes of rare and dominant species to be captured at different time points.

To test the conspecific negative density dependence (CNDD) experienced by sapling trees, we applied a Gamma-distributed generalized linear mixed-effects model (GLME) with a log link function to evaluate the relationship between sapling abundance and conspecific neighbor density (CI). The reason for using a Gamma distribution is that our data are positive, continuous, and exhibit a long-tailed distribution. The conspecific neighbor index (CI) was calculated based on Equation (5), and the sapling abundance for each species was quantified. Species with more than 96 individuals (density > 10 individuals/ha) were included in the GLME analysis, with the quadrats and species identities treated as random effects. The formula for calculating the CI is as follows [44]:

$$CI={\displaystyle {\sum }_i^N\frac{B{A}_i}{Di{s}_i}}$$

(5)

Here, N represents the number of conspecific neighbors within a specified radius, BA_i is the basal area of the conspecific neighbor i (calculated from the DBH measured during the 2013, 2018, and 2023 surveys), and Dis is the distance from the conspecific neighbor i to the target individual. In this study, we considered conspecific neighbors within a 15 m radius. Zhang et al. indicated that the negative conspecific density dependence (CNDD) within a 15 m range played a critical role in shaping the aggregated diversity structure of heterospecific saplings during the early life stages from 2013 to 2018 [45]. This suggests that density-dependent processes are fundamental to early-stage trees [45]. To ensure the reliability of the GLME, we conducted model validation and diagnostic checks. Overdispersion was assessed using the Pearson statistic ratio, implemented in the performance R package [46]. To evaluate the contribution of random effects (species and quadrat) to model variance, we used the VarCorr() function from the lme4 R package [47].

To explore potential plant interactions, we conducted co-occurrence network analysis to reveal the community assembly of WYS. Spearman correlation coefficients among species were calculated using the Hmisc package. Valid associations were defined by an absolute correlation coefficient (r) greater than 0.29 and a p-value less than 0.01. The resulting network topology was visualized using the ggraph package, and network attributes were calculated. The temporal dynamics of the community were assessed by comparing the topological structures of co-occurrence networks across different time periods. These analyses were implemented using the microeco package [48].

To identify keystone species that may play critical roles in maintaining the stability of the WYS community structure, we used within-module connectivity (Zi) and among-module connectivity (Pi) to identify key nodes. Additionally, we considered the relationship between the genus-level taxonomy of these species and topographic factors to determine the impact of environmental filtering on population sizes using redundancy analysis (RDA). All analyses were conducted in R_4.3.3.

3. Results

3.1. Plot Census

In 2023, a total of 184 species were identified within the WYS plot, including 12 species recorded for the first time. Skimmia reevesiana, Trema tomentosa, Dichroa febrifuga, and Neoshirakia japonica, which were documented during the 2013–2018 surveys, had completely died out by 2023. Toxicodendron sylvestre and Eurya loquaiana, first observed in 2018, persisted into 2023. Despite significant changes in species richness, the dominant species of the WYS community remained relatively stable, with Castanopsis carlesii continuing to have the highest importance value (IV). Notably, Eurya muricata appeared in the top ten IV rankings for the first time (

Table 1. Changes in top 10 species with importance values in Mt. Wuyi Forest dynamics plot. IV, important value.

2013 Species	IV	2018 Species	IV	2023 Species	IV
Castanopsis carlesii	6.88%	Castanopsis carlesii	7.16%	Castanopsis carlesii	7.71%
Castanopsis fordii	4.61%	Castanopsis fordii	4.63%	Castanopsis fordii	4.54%
Castanopsis eyrei	4.60%	Castanopsis eyrei	4.33%	Engelhardtia fenzelii	3.83%
Engelhardtia fenzelii	3.95%	Engelhardtia fenzelii	3.78%	Castanopsis eyrei	3.80%
Schima superba	2.91%	Syzygium buxifolium	3.04%	Syzygium buxifolium	3.41%
Syzygium buxifolium	2.86%	Schima superba	2.82%	Altingia gracilipes	2.90%
Castanopsis faberi	2.73%	Castanopsis faberi	2.62%	Schima superba	2.80%
Altingia gracilipes	2.51%	Altingia gracilipes	2.62%	Rhododendron henryi	2.74%
Rhododendron henryi	2.48%	Rhododendron henryi	2.54%	Eurya muricata	2.69%
Lithocarpus harlandii	2.42%	Elaeocarpus japonicus	2.47%	Elaeocarpus japonicus	2.63%

). Among the top ten species ranked by dominance in 2023, seven exhibited recruitment rates lower than mortality rates, resulting in reduced abundance and negative population changes within the plot (Table S1).

The diameter at breast height (DBH) distribution in 2023 maintained a J-shaped pattern consistent with the 2013–2018 distributions. However, the number of species in the small-diameter class (1–10 cm) was significantly lower than in 2013 and 2018 (Figure 1). No significant changes were observed in the abundance of species with a DBH ≥ 21 cm. The locations of the recruited individuals generally coincide with those of deceased individuals (Figure S2).

Species diversity analyses based on the Hill numbers revealed contrasting trends. While species richness (q = 0) significantly increased in 2023, diversity indices accounting for Shannon diversity (q = 1) and dominant species contributions (q = 2, Simpson diversity) were lower compared to 2013 and 2018 (Figure 2). Our results provide a clear quantification of the changes in population and community species diversity from 2013 to 2023. To further investigate the reasons behind the negative growth of dominant populations, we examined the relationship between population density and the conspecific neighbor index.

From 2013 to 2023, a total of 90, 91, and 83 species, respectively, were included in the generalized linear mixed-effects model (GLME) analysis. The conspecific neighbor index was primarily distributed below 1000 cm²/m, with the majority concentrated between 0 and 400 cm²/m (Figure 3a–c). The GLME results indicate a positive correlation between the conspecific neighbor index and the absolute abundance of saplings, with the mean positive coefficient increasing over the years (Figure 3d). The dispersion ratios of the three models ranged from 0.059 to 0.185, with a slight overdispersion detected in 2013. However, all p-values were greater than 0.4, indicating that the level of dispersion remained within an acceptable range. The variance contribution of the quadrat random effect ranged from 0.52 to 0.62, while the variance contribution of the species random effect ranged from 1.25 to 1.42.

3.2. Co-Occurrence Network Analysis of the WYS Plant Community

We constructed species-level co-occurrence networks for the WYS plant community (Figures S3–S5). The results reveal that the proportion of positive correlations among species richness was highest in 2023, but the modularity coefficient was the lowest across the three survey years (

Table 2. Topological characteristics of WYS plant community networks from different species.

Attribution	2013 Value	2018 Value	2023 Value
Vertex	84.00	79.00	94.00
Edge	302.00	276.00	310.00
Average degree	7.19	6.99	6.60
Average path length	0.95	0.92	0.93
Network diameter	2.00	2.00	3.00
Clustering coefficient	0.41	0.42	0.42
Density	0.09	0.09	0.07
Heterogeneity	0.83	0.92	1.03
Centralization	0.24	0.27	0.25
Modularity	0.44	0.38	0.37
Positive	81.79%	87.32%	87.74%
Negative	18.21%	12.68%	12.26%

). In 2013, species of the Castanopsis genus (all EcM species) exhibited significant associations with 60 plant species, with a positive correlation rate of 93%. In 2018, Castanopsis species showed significant associations with 56 plant species, with a positive correlation rate of 91.07%. By 2023, Castanopsis species maintained significant associations with 60 plant species, with a positive correlation rate of 90% (Figure 4). In 2023, the network density and average degree were at their lowest, while the network diameter reached its highest value (Table 2).

Zi-Pi analysis indicated that module hubs played a pivotal role in maintaining plant community stability from 2013 to 2023, and most of these were rare species (Figure 4). Among the dominant species, only Castanopsis eyrei was identified as a module hub in 2018. While two species acted as connectors during 2013–2018, they shifted to peripheral nodes by 2023 (Figure 4). Interestingly, from 2013 to 2023, dominant species—particularly those with high importance values—were consistently categorized as peripheral nodes (Figures S6–S8). The results of the co-occurrence networks clearly demonstrate the changes in community stability over the decade from 2013 to 2023 while highlighting the relatively minor role of dominant species in community succession.

The redundancy analysis revealed that the average elevation, convexity, slope, and aspect collectively explained 84.9% of the total variation along the first axis (Figure 5, 999 permutations: p < 0.01). Among these, convexity and average elevation were identified as the primary driving factors. Additionally, species exhibited varying responses to environmental factors: some species (from, e.g., the Eurya and Castanopsis genera) were more adapted to areas with high convexity and elevation, while others (e.g., Camellia and Michelia) were more influenced by the slope aspect. We also examined the correlations between species richness and topographic factors. Spearman’s correlation analysis showed that the mean elevation and convexity within each 20 m × 20 m quadrat were significantly correlated with species richness. Notably, genera with higher abundance, such as Castanopsis, exhibited significant positive correlations with both the mean elevation and convexity. However, no significant relationships were detected between the species richness and the slope aspect or gradient (Figure S9).

4. Discussion

4.1. Changes in Species Diversity in Mt. Wuyi

Community changes are often driven by shifts in population dynamics at the local scale [9,15]. Our results reveal that the population size of C. carlesii expanded over the decade (Table 1). However, not all populations exhibited consistent trends. For example, the importance value (IV) of C. faberi declined from 2.72% in 2013 to 2.62% in 2018, and by 2023, the species no longer ranked in the top 10 IV species. This decline in Castanopsis faberi from 2013 to 2018 could be attributed to conspecific negative density dependence (CNDD). The study by Zhang et al. confirmed this pattern, showing that from 2013 to 2018, the conspecific neighbor density and conspecific neighbor index consistently had a negative impact on sapling survival [49]. The peak population density of C. faberi in 2013 may have exceeded the environmental carrying capacity, leading to a contraction over five years due to resource limitations. However, the CNDD alone is unlikely to have caused such a dramatic decline in the following five years. In 2023, we experienced the hottest year in the past 2000 years, with surface temperatures rising by 0.23 °C between 2016 and 2023 [14]. Additionally, our 2023 census found that while the importance values of 7 out of the top 10 species increased, their population sizes exhibited a negative trend (Table 1 and Table S1). This suggests that climate warming may be a primary driver of population fluctuations in the Mt. Wuyi community. Although we lack more direct evidence to confirm the impact of drought events on community assembly, an observational study has suggested that CNDD is more likely to affect seedling-stage species during dry seasons, thereby reducing understory recruitment rates [50]. Therefore, future studies should establish broader climate monitoring systems to test whether high temperatures and droughts also serve as key drivers of population fluctuations in subtropical secondary forests.

Another notable change is the significant decline in the abundance of small-diameter (1–5 cm DBH) individuals over the three censuses, contrasting with a steady increase in the abundance of large-diameter individuals (>21 cm DBH). This suggests that the high mortality rate of small-diameter individuals has outweighed the growth of larger-diameter trees, potentially driving the sharp decline in overall individual numbers. Similar results were observed in the Gutianshan forest dynamics plot following ice and snow disturbances [51]. Additionally, the findings from the 2013–2018 study revealed that saplings of 32 dominant species within the plot were negatively affected by the conspecific neighbor index, and non-random mortality events were more likely to occur during the early life stages of these saplings [49]. However, unlike ice and snow disturbances, high temperatures may disproportionately harm smaller individuals compared to larger ones. This could be due to life history differences, as small-diameter individuals tend to allocate more resources to growth at the expense of defense against warmth stress [52,53]. The large-scale mortality of small-diameter individuals between 2018 and 2023 created ecological niches for colonization. In the 2023 census, 12 new species were recorded within the plot. While we did not assess whether these species are rare at the regional scale, the observed ecological niche availability within the plot may provide opportunities for rare species to establish. To verify whether rare species advantages exist in the WYS region, our GLME results reveal a positive correlation between sapling abundance and conspecific neighbor density. This finding indicates that species with lower abundance experience weaker self-limitation (H1). We explained that this phenomenon may be attributed to the lower conspecific encounter rate for rare species compared to common species at the local scale [1]. Rare species may be rare because they experience stronger negative effects when their local density is high, whereas abundant species exhibit the opposite pattern. As a result, species abundance effects in the community show a positive correlation [54].

Between 2018 and 2023, the increase in species richness led to greater unevenness in species distributions within the WYS plot (Figure 2). This indicates that the WYS community may be undergoing a new round of species turnover. The high mortality of small-diameter species during this period provided substantial ecological space for the establishment of new species. However, it is worth noting that the weightings of rare species (q = 1) and common species (q = 2) in 2023 were lower than in 2013–2018. This suggests that while new species have entered the plot, they have yet to establish stable coexistence with other species. Additionally, this result suggests that newly recruited species have low abundance and contribute little to overall evenness. Neutral theory highlights the importance of species immigration in suppressing niche differentiation, which can act as a critical driver of diversity and community stability [6]. Therefore, future research should aim to quantify the effects of neutral processes, such as species migration and dispersal, on the WYS community. While we emphasize the potential role of neutral theory in subtropical forests of southeastern China, our plot represents only a subset of the regional species pool. Broader niche differentiation may play a more significant role in maintaining community stability at the landscape scale.

4.2. Co-Occurrence Networks Reveal Structural Changes in the Mt. Wuyi Community

Traditional community ecology often focuses on pairwise interactions between species to infer broader community dynamics [55,56]. However, higher-order interactions repeatedly highlight the presence of multi-species interactions. Co-occurrence networks provide a means to uncover the complex competitive and mutualistic relationships among species. Across all three censuses, over 80% of species relationships exhibited positive correlations, indicating that mutualistic relationships dominate in the WYS region, while competition plays a secondary role (supporting H2). This phenomenon may be attributed to changes in species composition, particularly the introduction of 12 species newly recorded within the plot during the 2023 census, which likely altered the intensity of competition. The increase in nodes and edges suggests that niche expansion and mechanisms promoting species coexistence may have been reinforced. This finding aligns with results from studies conducted in the Huangshan region [25], though there are notable differences. In WYS, the species playing key linkage roles in the community are primarily arbuscular mycorrhizal (AM) species, which are also common species. This highlights the critical role of AM species in maintaining the structural stability of the WYS community. Compared to EcM species, AM species exhibit a significantly higher pathogen accumulation rate, indicating that AM species experience stronger conspecific negative density dependence (CNDD) than EcM species [57]. The potential effects of CNDD on numerous AM species may contribute to the maintenance of species diversity in the WYS forest community. From the perspective of network topology metrics, species interaction coefficients were lowest in 2018 but increased slightly in 2023, likely due to the influx of new species following periods of extreme heat, which enhanced the likelihood of species interactions. However, the network stability index for 2023 was the lowest across all three censuses, suggesting that community assembly had become less stable. The observed changes in network topology reflect the dynamics of WYS communities (Table 2). Network stability is quantified using the modularity index, and network structures with high modularity are generally considered to enhance ecosystem stability [58,59]. The increase in vertices and edges, alongside reduced modularity and density, suggests that the community became more integrated and cohesive over time, potentially due to the dominance of positive interactions. Meanwhile, the rise in heterogeneity and centralization highlights an uneven distribution of interactions, potentially driven by the emergence of dominant species or functional groups. The clustering coefficient showed a minor increase from 0.41 in 2013 to 0.42 in 2023, suggesting stable local clustering in species interactions. Modularity decreased slightly over time (from 0.44 to 0.37), reflecting weaker subdivision into discrete modules. This trend implies a reduction in the functional or spatial segregation of species groups, potentially indicating a more integrated community structure [25]. These results suggest that while high temperatures promote community turnover and the introduction of new species, they also create a platform for increased competitive exclusion, leading to community fluctuations.

The Zi-Pi analysis revealed that only Castanopsis eyrei played a critical role in maintaining community stability from 2013 to 2018. A previous study linking species-genetic diversity in WYS showed a positive correlation between the genetic diversity of C. eyrei and the species diversity of its associated community [60]. Species-genetic diversity correlations (SGDCs) often reflect intra- and interspecific relationships. Positive SGDCs suggest that high genetic diversity enhances species-specific plant–soil feedback (PSF), promoting stable coexistence and increased species richness within the community [61]. This explains why C. eyrei acted as a critical connector species between 2013 and 2018. However, it is concerning that, in 2023, most dominant species, including Castanopsis, were categorized as peripheral nodes (Figure 4). This could be due to the dominance of single genotypes within these species over the five-year period, driven by superior intra-species competition and enhanced disease resistance compared to other genotypes. This result implies that dominant species contribute less to community assembly in mid-subtropical forests, while attention should instead be given to AM species, which play a more critical role in community assembly. This finding contrasts sharply with studies conducted in Huangshan [25]. Interestingly, our results show a significant positive correlation between the richness of dominant genera and elevation. This supports the notion that elevation may be a key process influencing dominant species population fluctuations. The RDA results clearly demonstrate that environmental filtering significantly shapes the composition of the subtropical forest community. Key environmental factors, such as convexity and average elevation, appear to exert strong selective pressures, favoring species like Castanopsis and Eurya, which are better adapted to high-convexity and high-altitude conditions. This highlights the need for further research into the role of environmental selection in shaping genetic diversity and its implications for community assembly and stability.

5. Conclusions

Our study reveals that the subtropical evergreen broad-leaved forest in Wuyi Mountain has remained relatively stable over the past decade, mainly due to the persistence of dominant species. However, recent warming has increased mortality among small-diameter individuals of dominant species, facilitating new species colonization. We found that positive interspecific interactions outweighed competition, highlighting their role in maintaining community stability. Despite their low abundance, rare species were crucial for biodiversity maintenance and ecosystem resilience. Beyond species interactions, climate and topography likely influence community assembly, but limited census and climate data constrained our ability to quantify these effects. Future research should experimentally test CNDD effects through conspecific neighbor removal and monitor long-term climate impacts on species coexistence and CNDD dynamics.