2.1 Study Area and Data Collection

The study area is located in northeastern China (39°42′48″–53°19′21″ N; 119°48′12″–134°01′01″ E), spanning parts of Jilin, Heilongjiang, and Liaoning Provinces, as well as the Inner Mongolia Autonomous Region. Covering approximately 700,000 km², the region features a temperate continental climate, with annual precipitation ranging from 363.8 to 1073.7 mm and mean annual temperatures between −5.6°C and 9.8°C (Fick and Hijmans 2017).

In the summer of 2017, a network of 456 permanent circular plots (0.1 ha each; radius = 17.84 m) was established across the study area (Figure 1). Within each plot, all trees with a diameter at breast height (DBH) ≥ 5 cm were identified, measured, and mapped (Table S1). Increment cores were extracted from the north side of each tree at a height of 1.3 m using a 5.15 mm diameter increment borer. These cores were preserved and later analyzed in the laboratory to measure annual ring width increments.

2.2 Metacommunities

Following Qiao et al. (2022), we aggregated neighboring plots into metacommunities to account for spatial variation in our analyses. Because the plots were unevenly distributed, with some areas sampled more densely than others, this spatial heterogeneity could introduce bias in the analysis of β scale patterns. Specifically, metacommunities in densely sampled areas would cover significantly smaller spatial extents than those in sparsely sampled areas. To minimize this bias, we selected 262 of the 456 plots based on their relatively uniform spatial distribution across the study region. These 262 plots, previously used in studies on β diversity (Zhang et al. 2020) and ecosystem stability (Qiao et al. 2023a, 2023b), are representative of the region.

Each plot was designed as a focal plot, and the n closest plots were identified and aggregated with a focal plot to form a metacommunity of size N = n + 1. To ensure the robustness of our study results, we considered multiple values of N (5, 6, 7, 8, 9, and 10). To assess the spatial correlation of community composition, we performed a Mantel correlogram analysis, which indicated that neighboring communities exhibited similar compositions when the spatial extent of metacommunities was less than 200 km (Figure S2). We also calculated the distances among local plots within each metacommunity for different values of N, confirming that the selected spatial extents were appropriate for analysis (Figure S3).

2.3 Stability and Related Metrics

For each metacommunity, we calculated the following stability-related metrics based on aboveground carbon increment: α stability (α_s), γ stability (γ_s), spatial asynchrony (φ), species asynchrony (φ_sp), and species stability (Sp_s).

Aboveground carbon increment was used as a proxy for forest carbon sequestration. We calculated this metric across three time periods: 2005–2009, 2009–2013, and 2013–2017. The use of multi-year measurement intervals to estimate forest carbon increment reduces the influence of short-term anomalies caused by extreme climatic events and helps smooth annual fluctuations (Ouyang et al. 2021; Zhou et al. 2022). Detailed calculation methods are provided in Supporting Information S1.

To remove directional trends in aboveground carbon increment over time, we applied a detrending method (Tilman et al. 2006). Stability-related metrics were then calculated following the approaches of Wang and Loreau (2014) and Wilcox et al. (2017).

Species asynchrony and species stability values were averaged across plots, weighted by total plot abundance, to obtain a single value for each metacommunity.

2.4 Biodiversity

Within each metacommunity, we calculated biodiversity metrics across multiple facets and scales using Hill numbers (Chao et al. 2014). Hill numbers adhere to the replication principle, enabling the decomposition of γ diversity into independent α and β diversity components.

To quantify taxonomic, functional, and phylogenetic diversity, we employed the Gini-Simpson index (q = 2), Walker's FAD (q = 0), and Faith's phylogenetic diversity (q = 0), respectively. Functional traits were measured following Cornelissen et al. (2003) and Hao et al. (2020). Specifically, we included seven plant functional traits related to carbon sequestration capacity in our functional diversity calculation: specific leaf area (SLA), leaf area (LA), wood density (WD), leaf nitrogen content (LNC), leaf carbon-to-nitrogen concentration (C:N), and maximum plant height (Hmax) (Table S3).

For each metacommunity, we calculated taxonomic, functional, and phylogenetic diversity at both the α and γ scales. Corresponding β diversity metrics were computed as γ diversity divided by α diversity. All diversity metrics were calculated using the hillR package in R software (Li 2018).

2.5 Climate Data

We obtained the monthly mean temperature (Peng 2019) and monthly precipitation (Peng 2020) data from 2000 to 2017 for each plot through the National Qinghai–Tibet Plateau Scientific Data Center (https://data.tpdc.ac.cn). Climate variability was quantified as the coefficient of variation of temperature and precipitation over this period (Table S4; Supporting Information S2).

Additionally, we extracted 19 bioclimatic variables from the WorldClim database (http://www.worldclim.org/) for the study period (Table S4). To evaluate the relative importance of climate variables in driving ecosystem stability across scales, we employed random forest models using the rfPermute package. The percentage increase in mean squared error (%IncMSE) was used to quantify the reduction in model accuracy when a given variable was randomly permuted, with higher %IncMSE values indicating stronger predictive power (Liaw and Wiener 2002). This metric is widely used in ecological studies due to its robustness against multicollinearity and nonlinear relationships (Qin et al. 2024; Silvestro et al. 2024). Our results indicated that temperature and precipitation variability exhibited the highest %IncMSE values, suggesting they were the strongest predictors of ecosystem stability across scales (Figure S4).

2.6 Statistical Analysis

We applied ordinary least-squares (OLS) regression to evaluate the impacts of climate variability, biodiversity (taxonomic, functional, and phylogenetic), and biotic stability mechanisms (species asynchrony, species stability, and spatial asynchrony) on ecosystem stability across spatial scales. In these models, stability was treated as the response variable, while the other factors were predictor variables. We assessed the individual effects of each predictor variable across all six spatial extents, standardizing all variables prior to analysis to facilitate comparisons of their relative importance.

To examine relationships between multiple facets of biodiversity and biotic stability mechanisms, we employed linear mixed-effects models (LMMs), including spatial extent as a random factor. The models were fitted using the lme function in the nlme package (Pinheiro et al. 2013). To meet normality assumptions, all variables were log₁₀-transformed prior to analysis. We also conducted simple linear regression to estimate standardized coefficients and assess how relationships between α diversity and species asynchrony, α diversity and species stability, and β diversity and spatial asynchrony changed with increasing metacommunity spatial extent.

We used piecewise structural equation modeling (SEM) to investigate the direct and indirect pathways linking climate variability, biodiversity facets, and ecosystem stability across scales. Metacommunity spatial extent was included as a random factor. SEMs were constructed using the lme function from the nlme package and the psem function from the piecewiseSEM package (Lefcheck 2016). The initial SEM was based on the conceptual model (Figure S1). Shipley's d-separation test was applied to identify any missing paths (Shipley 2009). Fisher's C-statistic (p > 0.05) indicated that there were no additional paths required, and the model fit well. The final SEM was determined by eliminating nonsignificant paths. To account for potential spatial autocorrelation, we incorporated Gaussian spatial autocorrelation structures into the SEM. Model fit was assessed using marginal R² and conditional R², while Akaike information criterion (AIC) was used for model selection. All variables were standardized before SEM construction to obtain standardized path coefficients. All statistical analyses were conducted in R v4.2.2 (R Core Team 2023).

3.1 Relationship Between Multidimensional Biodiversity and Stability Across Scales

Our results indicate that α diversity—including taxonomic α diversity (α_TD), functional α diversity (α_FD), and phylogenetic α diversity (α_PD)—was positively correlated with α stability across metacommunity spatial extents (N = 6–10) (Figure 3a; p < 0.001). Similarly, γ diversity (taxonomic, functional, and phylogenetic) was positively associated with γ stability (Figure 3b; p < 0.001). However, the diversity–stability relationship (DSR) was relatively weak at both local and larger spatial scales, with standardized coefficients ranging from 0.08 to 0.33. At the local scale, species asynchrony exhibited a stronger association with α stability than species stability. At the larger scales, spatial asynchrony was more strongly associated with γ stability than α stability. Temperature variability was negatively correlated with both α and γ stability, whereas precipitation variability exhibited the opposite pattern. The strength of these relationships increased with spatial extent (Figure 3).

3.2 Relationship Between Multidimensional Biodiversity and Biotic Stability Mechanisms Across Scales

We found that α_TD, α_FD, and α_PD were positively correlated with species asynchrony but negatively correlated with species stability (Figure 4a,b,d,e; Figure S5a,b). At the β diversity level, functional β diversity (β_FD) and taxonomic β diversity (β_TD) showed weak positive associations with spatial asynchrony (Figure 4c; Figure S5c). In contrast, phylogenetic β diversity (β_PD) exhibited a stronger positive relationship with spatial asynchrony, with the strength of the relationship increasing across spatial extents (Figure 4f).

3.3 Relationships Among Climate Variability, Multidimensional Biodiversity, and Stability Across Scales

SEM revealed that increased temperature variability (p < 0.001; standardized direct effects: −0.76 to −0.49) and precipitation variability (p < 0.001; standardized direct effects: −0.37 to −0.31) were negatively associated with α_TD, α_FD, and α_PD. These declines in α diversity were linked to reduced species asynchrony, which, in turn, corresponded to lower α stability (Figure 5a–c). We observed strong positive correlations between α diversity (α_TD, α_FD, and α_PD) and species asynchrony (p < 0.001; standardized direct effects: 0.50 to 0.81), which counterbalanced the negative correlations between α diversity and species stability (p < 0.001; standardized direct effects: −0.61 to −0.27). As a result, the overall net positive association between α diversity and α stability was primarily mediated by species asynchrony (Figure 5a–c; Figures S6 and S7).

Temperature variability exhibited a net negative association (both direct and indirect pathways) with α stability, whereas precipitation variability exhibited a net positive association with α stability (Table S5). At the β diversity level, β_TD, β_FD, and β_PD indirectly positively correlated with γ stability via spatial asynchrony (p < 0.001; standardized total effects: 0.05 to 0.18). However, β_PD and β_FD contributed more strongly to γ stability than β_TD (Figure 6). Increased temperature variability was negatively associated with β_TD and β_PD (p < 0.001; standardized direct effects: −0.59 to −0.14). Declines in β diversity were associated with reduced spatial asynchrony, which in turn corresponded to lower γ stability (Figure 5a–c). Conversely, increased precipitation variability was positively associated with β_TD, β_FD, and β_PD (p < 0.001; standardized direct effects: 0.21–0.46), promoting greater spatial asynchrony and higher γ stability (Figure 5a–c).

4.1 Consistent Positive Relationships Between Multidimensional Biodiversity and Stability Across Scales

Our findings support the first hypothesis that multiple facets of biodiversity are positively correlated with stability across spatial scales, from local (α) to larger (γ) scales, in natural temperate forests of northeastern China. Previous studies on DSRs have primarily focused on single biodiversity facets at local scales, with increasing evidence consensus suggesting that both functional and phylogenetic diversity enhance ecosystem stability (Craven et al. 2018). Variation in species' taxonomic composition, functional traits, and evolutionary histories facilitates niche differentiation over time, promoting asynchronous population fluctuations that enhance local ecosystem stability (Sasaki et al. 2019; Schnabel et al. 2021).

Additionally, our results revealed a positive correlation between γ diversity and γ stability, which may be attributed to local-scale DSRs contributing to patterns at broader scales. However, the R² in our diversity–stability models was relatively low, indicating that while biodiversity positively influences stability, the strength of this relationship remains weak in natural ecosystems—a finding consistent with previous findings (Houlahan et al. 2018). Furthermore, we observed that the relative importance of different biodiversity facets varied across spatial scales.

At the local scale, taxonomic and phylogenetic diversity were more strongly associated with α stability via species asynchrony than functional diversity. We speculate that unmeasured functional traits, such as subsurface structural traits (Da et al. 2023), phenology (Gu et al. 2022), and non-structural carbohydrate storage (O'Brien et al. 2014), may play a more significant role. However, due to the scarcity of data on these traits, further investigation is needed. Moreover, previous studies indicate that the relative importance of functional diversity varies depending on ecosystem functions. For example, functional diversity contributes more to biomass stability than taxonomic and phylogenetic diversity (Guo et al. 2021), whereas its contribution to productivity stability is comparatively weaker (Craven et al. 2018).

At larger spatial scales, phylogenetic diversity exhibited stronger associations with γ stability via spatial asynchrony compared to taxonomic diversity. This may be because ecosystems at broader scales encompass more complex environmental gradients and historical processes. Phylogenetic diversity better captures differences in species' evolutionary histories and response strategies to environmental changes, particularly in heterogeneous environments (Qian et al. 2024a, 2024b; Wieczynski et al. 2019). Consequently, studies that focus solely on taxonomic diversity may fail to account for interspecific genetic relationships, evolutionary histories, and diverse adaptive strategies to environmental variability (Prinzing et al. 2008; Zhang et al. 2023).

Given the interdependence among different biodiversity facets, we quantified their individual and combined contributions to stability. The combined contribution of TD, FD, and PD to species asynchrony at the α scale was 43.6%. After controlling for taxonomic α diversity, the combined explanatory power of functional and phylogenetic α diversity on species asynchrony was 9.4%. At the β scale, the three biodiversity facets collectively explained only 0.4% of spatial asynchrony. However, phylogenetic β diversity exhibited significantly greater explanatory power, uniquely contributing 7.7% to spatial asynchrony (Figure S8). These results suggest that functional diversity provides limited additional explanatory power at both α and β scales, while phylogenetic diversity plays a more critical role at larger spatial scales. This finding underscores the complementary roles of different biodiversity dimensions in understanding and predicting ecological stability. Importantly, no single biodiversity facet demonstrates substantial explanatory power in isolation, highlighting the necessity of integrating multiple biodiversity components for a comprehensive assessment of stability patterns.

4.2 Insurance Effects as a Key Biotic Mechanism for Maintaining Stability Across Scales

Consistent with our second hypothesis, insurance effects emerged as the primary mechanism for maintaining stability at both local and larger spatial scales.

At the local scale, species asynchrony was more strongly associated with α stability than species stability. This finding aligns with studies conducted in forests (Wu et al. 2022; Yuan et al. 2019) and grassland ecosystems (Isbell et al. 2009; Valencia et al. 2020; Xu et al. 2021). The core mechanism of species asynchrony is its role as the temporal manifestation of niche differentiation, which buffers local disturbances through compensatory effects such as resource-use differentiation. For example, the phenological complementarity between C3 and C4 plants enhances community stability by mitigating the impacts of interannual climatic fluctuations (Sasaki et al. 2024). In contrast, communities with high species stability may still experience substantial functional fluctuations if all species respond synchronously to disturbances (i.e., low asynchrony) (Zhao et al. 2022). These findings highlight the crucial role of species asynchrony in maintaining ecosystem stability.

Theoretical models suggest that compensatory population dynamics should decrease species stability while increasing community-level temporal stability by enhancing species asynchrony (Loreau and de Mazancourt 2013; Thibaut and Connolly 2013). However, Schnabel et al. (2021) observed that species stability exerted a stronger influence than species asynchrony on community productivity stability in the BEF-China experiment. They attributed this discrepancy to the slower growth cycles of forests and the shorter duration of the study. In contrast, our study, which spans a longer time period (> 10 years), detected stronger compensatory dynamics among species in temperate forests, further emphasizing the pivotal role of species asynchrony in stabilizing aboveground carbon increment at the local scale. One possible explanation is that communities with high species asynchrony can more efficiently optimize canopy occupancy among coexisting species, thereby enhancing overall ecosystem stability (Morin et al. 2014). Species asynchrony is primarily driven by interspecific differences in phenology, physiological traits, and resource utilization strategies. For example, in Northeast China, Larix gmelinii exhibits rapid photosynthesis as spring temperatures rise but enters dormancy during autumn cooling. In contrast, Pinus koraiensis maintains weak photosynthesis at low winter temperatures but is more susceptible to spring frosts (Ma et al. 2024). This phenological mismatch between Larix gmelinii and Pinus koraiensis reduces the sensitivity of community carbon sequestration to extreme temperature fluctuations (Ning et al. 2022).

Furthermore, differences in resource-use strategies contribute to ecosystem stability by increasing species asynchrony: fast-resource-acquisition species (e.g., Larix gmelinii) efficiently sequester carbon when resources are abundant, while resource-conservative species (e.g., Pinus koraiensis) maintain carbon sequestration under harsh conditions. Consequently, mixed-species forests exhibit higher stability than monocultures (Del Río et al. 2022). In mixed forests, species interactions mediated by biological traits—such as root architecture, soil microbial community composition, and belowground competitions—further enhance compensatory effects by influencing interspecies relationships and resource allocation (Herms et al. 2022; Veresoglou et al. 2024; Yu et al. 2021). These findings underscore how species diversity, by modulating species-specific responses to environmental change, enhances ecosystem stability and adaptability.

Our study also revealed that as spatial extent increased, the correlation between species asynchrony and α stability strengthened. A plausible explanation is that larger spatial extents incorporate a greater number of species, which enhances asynchronous responses to environmental fluctuations (Isbell et al. 2009; Liu et al. 2020).

At the larger scale, our results demonstrated that spatial asynchrony exerted a stronger influence on γ stability than α stability did on γ stability in large temperate forests. Similar findings have been reported in animal communities; for instance, Catano et al. (2020) found that spatial asynchrony explained three times more variation in regional stability of bird biomass than α stability. However, contrasting results have been observed in grassland ecosystems, where α stability was found to have a greater effect on γ stability than spatial asynchrony (Hautier et al. 2020; Wang et al. 2021; Wilcox et al. 2017). The same pattern was detected in grassland studies examining the effects of grazing or nitrogen addition (Liang et al. 2021; Zhang et al. 2019). These discrepancies may be due to limited spatial grain size and lower heterogeneity in grasslands (Wang et al. 2021). In relatively homogeneous landscapes, where communities respond similarly to environmental fluctuations, α stability becomes the primary determinant of gamma γ stability. For example, in agricultural ecosystems, crop stability is primarily related to species stability, and increasing the harvested area does not show greater heterogeneity or provide stronger insurance effects (Meng et al. 2024).

By contrast, in forest ecosystems, intrinsic heterogeneity and species turnover increase with spatial scale, leading to greater differentiation in community responses due to environmental drivers and demographic stochasticity. Consequently, spatial asynchrony plays a more dominant role at larger scales (Daleo et al. 2023). In our study, which spans a wide spatial gradient, we observed that as the metacommunity range N increased, the relationship between spatial asynchrony and γ stability strengthened, whereas the relationship between α stability and γ stability weakened slightly (Figure 3). This suggests that the relative importance of α stability and spatial asynchrony in shaping γ stability is scale-dependent (Qiao et al. 2022).

Moreover, β diversity was positively correlated with γ stability through its relationship with spatial asynchrony, consistent with previous theoretical (Wang et al. 2019; Wang and Loreau 2016) and empirical studies (Li et al. 2023; Wang et al. 2021). β diversity generally increases with geographic distance due to physical barriers and dispersal limitations (Qian et al. 2024a, 2024b), which in turn promotes asynchronous community responses to environmental changes. However, some studies have reported that β diversity is a weak predictor of spatial asynchrony (Wilcox et al. 2017; Yang et al. 2022). This discrepancy may arise from confounding factors masking the correlation, such as precipitation changes (Song et al. 2024), fertilization effects (Hautier et al. 2020), or stochastic population dynamics (Zhang et al. 2019).

Additionally, the plot selection strategies may influence the related analysis of β scale analyses. Small sampling areas or uneven plot distribution may fail to accurately reflect community compositional differences. To minimize spatial biases, we applied a systematic grid sampling approach for study analysis. We further conducted 1000 resampling iterations to address the robustness of our findings. The results across all resampling iterations were highly consistent with those presented in Figures S9–S11, confirming that plot selection had minimal influence on our conclusions.

4.3 Climate Variability Is Associated With Changes in Biodiversity and Stability Across Scales

Climate change is a major driver of biodiversity loss, affecting both terrestrial and aquatic ecosystems across scales, from local to global (Bellard et al. 2012). Our results indicate that temperature and precipitation variability are negatively and positively correlated with α stability, respectively, consistent with findings from grassland ecosystems (Han et al. 2023). These results support our hypothesis that α diversity (taxonomic, functional, and phylogenetic) declines with increased temperature and precipitation variability. Such declines in α diversity are associated with reduced species asynchrony, leading to diminished α stability (Boyce et al. 2006; Isbell et al. 2009).

Previous theoretical predictions suggest that the response of α diversity to climate variability is likely to follow a single-peaked curve—initially increasing with climate variability, peaking at an optimal point, and then declining as the risk of stochastic extinction exceeds competitive stability (Letten et al. 2013). However, we did not observe such a single-peak relationship between climate variability and α diversity (Figure S12). This discrepancy may be attributed to the relatively short duration of our study.

Furthermore, both temperature and precipitation variability exhibited indirect positive correlations with α stability through species stability. This suggests that temporal climatic variability enhances species stability by increasing the number of temporal ecological niches available within a given space (i.e., temporal niche differentiation), thereby stabilizing competitive interactions (Chesson 2000; Chesson and Huntly 1997; Loreau and de Mazancourt 2013). However, these positive correlations were weaker than the negative relationship between climatic variability and α stability through α diversity. Unexpectedly, precipitation variability showed a net positive correlation with α stability. We speculate that this positive relationship may reflect mechanisms such as diversification of resource availability and reduced competitive pressures among species, which could enhance community stability (Van Dyke et al. 2022). In grassland ecosystems, precipitation variability primarily enhances temporal community stability by increasing species asynchrony. In contrast, our study found only weak or nonsignificant positive correlations between precipitation variability and species asynchrony. This could be due to the slower growth period of forest communities compared to grassland ecosystems, leading to weaker asynchronous dynamics under precipitation variability (Zhang et al. 2023).

At the larger scale, there is limited evidence regarding whether temporal climate variability facilitates or limits spatial changes in diversity (Letten et al. 2013). Our results revealed that both β taxonomic and phylogenetic diversity decrease with increasing temperature variability. This suggests that temperature variability contributes to the homogenization of species composition and phylogenetic diversity across different regions, potentially reducing species turnover and phylogenetic potential among communities (Xu et al. 2023). However, temperature variability was positively correlated with functional β diversity, which may reflect adaptive mechanisms employed by species to cope with changing temperatures. As temperature variability intensifies, species may reduce competitive pressures and optimize resource allocation by adjusting resource utilization strategies, such as rates of resource acquisition (Wieczynski et al. 2019). Taxonomic, functional, and phylogenetic β diversity increased with precipitation variability. This could be attributed to the increased spatial heterogeneity at larger scales, where interannual precipitation variability allows species with diverse hydraulic characteristics to establish in different regions. This diversification of hydraulic traits likely promotes species turnover among communities, thereby increasing β diversity (Xu et al. 2023).

Our study found that species carbon sequestration stability is more sensitive to temperature variability than to precipitation variability (Jonas et al. 2015). Temperature and precipitation are crucial limiting factors for vegetation growth. Temperature influences plant metabolic rates (Michaletz 2018), leaf water use efficiency (Wang et al. 2023a, 2023b), and phenology (Prevéy et al. 2017). Precipitation, on the other hand, affects photosynthesis (Chen et al. 2009), competitive interactions (Van Dyke et al. 2022), and radial growth (Zeng et al. 2022), especially in regions with low latitudes. Northeastern China, located in the mid-latitudes, experiences relatively poor thermal and water conditions. Despite this, its relatively low annual evapotranspiration results in predominantly humid or semi-humid conditions (Su et al. 2022). There are clear patterns in how vegetation growth responds to precipitation: plants in arid regions exhibit higher positive sensitivity to precipitation variability than those in more humid regions (Li et al. 2024). In arid areas, plant growth is primarily limited by water availability (Currier and Sala 2022), whereas in humid areas, factors such as nutrients and light, along with temperature, limit growth. Temperature, for instance, is the main limiting factor for the radial growth of the dominant species Larix olgensis and Pinus koraiensis in the northeastern China forest region (Ning et al. 2022). These regional variations may explain why carbon sequestration in this area is more influenced by temperature variability (Zeng et al. 2022).

With global warming, plants in arid and humid subtropical regions are increasingly sensitive to precipitation variability rather than to temperature variability (Li et al. 2021, Specifically, vegetation in arid regions has shown an average 0.624% per year increase in sensitivity to precipitation variability (Zhang et al. 2022), with ecosystem stability tending to decrease as precipitation variability increases (Bernardino et al. 2025). This highlights the critical role of precipitation variability in influencing ecosystem stability in these regions, warranting further attention. Conversely, in tropical ecosystems, frequent extreme dry-wet cycles have resulted in a negative association between increased precipitation variability and ecosystem resilience (Yao et al. 2024). In contrast, in temperate ecosystems, temperature variability is the main factor reducing ecosystem resilience. This underscores the regional heterogeneity in the dominant factors influencing ecosystem resilience. Our study further confirms that temperature variability is negatively correlated with the temporal stability of temperate ecosystems, which has been similarly observed in temperate grassland ecosystems (Zhang et al. 2018). While the findings from our study are particularly relevant to temperate ecosystems, they may not be directly generalizable to other biomes or climatic zones due to significant regional differences in environmental heterogeneity and vegetation characteristics. Future research should focus on cross-regional studies to deepen our understanding of how different ecosystems mediate vegetation responses to global climate change.

Furthermore, climate variability affects not only biodiversity and stability across different scales but also the stability of ecosystem functioning by altering mechanisms such as species asynchrony and species stability (Garcia et al. 2014). Consequently, changes in diversity are not the primary cause of ecosystem stability but rather an intermediate response to global change drivers, such as climate change, nitrogen deposition, and land-use changes. These drivers may also affect ecosystem stability through alternative pathways (Blüthgen et al. 2016; Liu et al. 2023). In future research, it is crucial to gain a deeper understanding of the roles that these factors play in natural ecosystems to enhance our foundational comprehension of ecosystem stability. We propose that the multiple facets of biodiversity at different scales be conserved in the future to buffer the influence of climate variability on ecosystem functioning.

4.4 Limitations and Perspectives

In this study, we systematically analyzed the direct and indirect relationships among climate variability, biodiversity, and ecosystem carbon sequestration stability across different spatial scales. Unlike previous manipulative experiments conducted in grassland ecosystems, our research spans larger spatial scales and focuses on carbon sequestration stability in natural temperate forest communities—an area that has received relatively little attention in prior studies. Furthermore, we examined not only taxonomic diversity but also functional traits and phylogenetic diversity at broader scales, providing new perspectives and empirical insights for biodiversity conservation in the context of climate change.

Despite these contributions, our study has some limitations. First, our analyses did not account for other potentially influential factors, such as soil characteristics, resource heterogeneity (e.g., nutrient availability), and light availability. These factors are known to influence ecosystem stability and carbon dynamics, but their exclusion was due to limitations in data availability and research scope. At large scales, climate is the dominant driver of the DSR (García-Palacios et al. 2018); thus, the omission of these variables is unlikely to alter our primary conclusions regarding the importance of climate variability for ecosystem stability. However, future research should further investigate these factors, particularly the mechanisms underlying soil-climate interactions, to develop a more comprehensive understanding of the stability of ecosystem carbon sinks.

Second, we aggregated metacommunities after uniform sampling, which effectively mitigated sampling bias in β scale analyses and enhanced the robustness of our results. However, artificially aggregating localized communities into metacommunities may overlook environmental heterogeneity and dispersal effects among local communities. To address this, future studies could employ systematic sampling at larger spatial scales using nested structures, which would better capture ecological processes and patterns across different scales.

Finally, because this study is based on observational data, the correlations we identified do not eliminate the possibility of confounding factors. Therefore, future manipulative experiments will be essential to establish causal mechanisms and further validate our findings.