Consequences of agricultural deforestation and subsequent afforestation on soil biodiversity and ecosystem multifunctionality

Abstract

The Earth is currently in an era where massive deforestation and afforestation coexist. The impact of large-scale agricultural deforestation and the subsequent afforestation on soil biota multidiversity, ecosystem multifunctionality (EMF), and the relationship between soil biota multidiversity and EMF (BEF_m) remain unclear. Here, we investigate 405 paired plots along a 4000 km south-north transect, spanning tropical, subtropical, temperate, and boreal zones. We measure 19 ecological functions and sequence soil biota (including bacteria, fungi, archaea, viruses, protists, and invertebrates) and metagenomes. We find that agricultural deforestation reduces soil multitrophic biodiversity by 25% and EMF by 58%, and afforestation has partially restored them, but pristine levels have not been reached. Agricultural deforestation decouples the positive BEF_m relationship across four climatic zones, while afforestation restores the positive BEF_m relationship in tropics and subtropics but not in temperate and boreal zones. Afforestation in the warmer zone triggers potential multitrophic cascades to enhance EMF, thereby strengthening BEF_m relationship. The changes in biogeochemical-cycling genes induced by afforestation exert more significant driving effects on EMF in the warmer zone than the colder zone. Our study provides integrative evidence that climate modulates the recovery of BEF_m relationship and offers multitrophic and metagenomic insights into the mechanisms underlying EMF.

Large-scale deforestation and afforestation currently coexist. Along a 4,000 km transect, Yu et al. show that deforestation reduces soil multi-diversity by 25% and ecosystem multifunctionality by 58%, while afforestation partially restores both.

Introduction

Since 1990, the world has lost almost six million hectares of forests annually, primarily driven by land-use changes related to agricultural expansion¹. The widespread agricultural deforestation has raised global concerns about the consequent disruption of biodiversity and ecosystem functioning and services^2,3. To promote forest restoration, large-scale afforestation initiatives, such as the Trillion Trees Initiative, Bonn Challenge, and United Nations Decade on Ecosystem Restoration, have been proposed and set ambitious goals^4–6. Gradually, forests are expanding in East Asia, Europe, and North America, mainly due to afforestation efforts^7,8. Thus, the Earth is currently in an era where massive deforestation and afforestation coexist⁹. Understanding the impacts of large-scale agricultural deforestation and subsequent afforestation of those deforested lands on biodiversity, ecosystem functions, and their interrelationships, especially in the hidden belowground systems, is imperative for predicting ecosystem responses to human activities and developing sustainable afforestation strategies in the context of global change.

Forest soils harbor diverse organisms, including archaea, bacteria, viruses, and various eukaryotes (fungi, protists, and invertebrates), which drive and support various ecosystem functions, such as decomposition, biogeochemical cycles, and soil productivity^10,11. Moreover, soil organisms occupy various trophic levels and form complex food webs that mediate energy and nutrient fluxes within ecosystems^11,12. Agricultural deforestation and afforestation often cause fundamental changes in habitat and environment^13,14, which may alter soil biodiversity and the ecosystem’s ability to simultaneously provide multiple ecological functions and services, i.e., ecosystem multifunctionality (EMF)^15–17. However, it remains unclear how soil multitrophic biodiversity and their associated EMF respond to large-scale agricultural deforestation and subsequent afforestation, and whether afforestation can restore soil multitrophic biodiversity and EMF disrupted by agricultural deforestation.

The biodiversity-ecosystem functioning (BEF) relationship is foundational to understanding how biodiversity drives ecosystem functions and how biotic community responds to changes in environment and habitat¹⁸, and has sparked intense debate for over 50 years¹⁹. Although BEF studies have provided strong evidence to support positive associations between biodiversity and single ecological functions²⁰, BEF patterns are inconsistent when expanding ecosystem functionality to EMF²¹. A systematic review study found that, in naturally assembled communities, 57.7%, 30.6%, and 11.7% of case studies have observed neutral (uncorrelated), positive, and negative relationships between biodiversity and EMF, respectively²¹. For soil biota, the biodiversity-EMF relationships are expected to be highly complex due to the presence of diverse and interacting taxa across multiple trophic levels. However, most current research on the soil biodiversity-EMF relationships ignored the multitrophic levels and their interrelationships^22–24, likely underestimating the true effect of soil biodiversity on EMF²². Moreover, the soil biodiversity-EMF relationships may change with land-use and ecosystem dynamics, such as the current massive agricultural deforestation and afforestation. However, how soil multidiversity (i.e., the overall diversity of bacteria, fungi, archaea, viruses, protists, and invertebrates) relates to EMF across deforested and afforested ecosystems remains poorly understood, which limits our understanding of the BEF relationship and ecosystem response to human activities, such as land-use change.

To fill these knowledge gaps, we conduct a large-scale field investigation to explore the effects of agricultural deforestation and subsequent afforestation on soil multidiversity, soil functional genes, individual ecological functions, EMF, and BEF relationships along a 4000 km south-north transect in China, spanning tropical, subtropical, temperate, and boreal zones. This region has experienced extensive agricultural deforestation in history and is currently undergoing rapid forest restoration^7–9, providing a natural laboratory for testing ecosystem responses to land-use change. We sample 405 paired plots, sequence soil biota communities (including bacteria, fungi, archaea, viruses, protists, and invertebrates) and metagenomes, and measure 19 widely used surrogates of ecosystem functions grouped into six categories^15–17, i.e., aboveground plant productivity, soil C storage, soil water regulation, matter decomposition, nutrient cycling, and plant-soil mutualism. We hypothesize that: (1) agricultural deforestation reduced soil multitrophic biodiversity and EMF, while afforestation could restore them; (2) agricultural deforestation decoupled the soil multidiversity-EMF relationship (BEF_m), and afforestation restored it; (3) the recovery of BEF_m relationship following afforestation is modulated by climate and soil functional genes.

Here, we report that agricultural deforestation reduces soil multitrophic biodiversity by 25% and EMF by 58%, and afforestation has partially restored them, but pristine levels have not been reached. Agricultural deforestation decouples the positive BEF_m relationship across four climatic zones, while afforestation restores the positive BEF_m relationship in tropics and subtropics but not in temperate and boreal zones.

Results

Impacts of agricultural deforestation and afforestation on soil multidiversity and EMF

Overall, agricultural deforestation decreased soil multidiversity by 25% and EMF by 58%, while afforestation led to partial recovery of both, though neither reached the level of native forest (Fig. 1b, c, and Supplementary Fig. 1a, b). Agricultural deforestation reduced the diversity of most groups of soil organisms (i.e., bacteria, fungi, protists, and invertebrates; Supplementary Fig. 1c–h) and most individual category functions (i.e., soil C storage, soil water regulation, matter decomposition, nutrient cycling, and plant-soil mutualism; Supplementary Fig. 2b–f). The decline in soil multidiversity caused by agricultural deforestation further diminished EMF (Supplementary Figs. 3, 4, and Supplementary Table 1). In contrast, afforestation increased the diversity of most soil organism groups and most individual category functions compared to the paired croplands (Supplementary Figs. 1c–h and 2b–f). Notably, the effect of afforestation on EMF, as measured by the natural logarithm of response ratio (LRR) over planted forest and cropland, significantly increased with mean annual temperature (Supplementary Fig. 5a). Afforestation was identified as the main contributor among multiple predictors in explaining soil individual ecological functions, including soil C storage, nutrient cycling, and matter decomposition (Supplementary Fig. 6).

Fig. 1. Ecosystem multifunctionality and soil multidiversity in native forests, croplands, and planted forests along a 4000 km south-north transect.

a Map showing sampling sites. The free continental data of the world map (on the left) was sourced from Natural Earth, supported by the North American Cartographic Information Society (https://www.naturalearthdata.com/), and the free data of the Vegetation Continuous Fields map (on the right) was from https://www.earthdata.nasa.gov/data/catalog/lpcloud-mod44b-061⁷⁶. ArcGIS Desktop 10.8 (Esri, West Redlands, CA, USA) was employed for mapping the distribution of the study sites. The blue triangle represents that the same sampling site contains three types of paired ecosystems, i.e., 3 natural forests, 3 croplands, and 3 planted forests. The red circle represents that the same sampling site contains two types of paired ecosystems, i.e., 3 croplands and 3 planted forests. b, c Respectively show ecosystem multifunctionality and soil multidiversity. Soil multidiversity is calculated as the average richness of all soil biota groups after min-max normalization (i.e., bacteria, fungi, archaea, viruses, protists, and invertebrates) (Hereafter). Statistical analysis was performed using Linear mixed-effects models (see “Method”), and the exact P value for the overall comparison is shown in each panel. Different letters in panels indicate significant differences at P < 0.05 (two-sided Tukey’s HSD test with multiple-comparison adjustment). Native forest: sample size n = 69; Cropland: n = 168; Planted forest: n = 168. Box (in violin plot) limits are the upper and lower quartiles. Whiskers are 1.5 times the interquartile range. Horizontal lines inside the box represent the median. Source data are provided as a Source Data File.

Shift in BEF_m relationships following agricultural deforestation and afforestation

Significant positive relationships between soil multidiversity and ecosystem multifunctionality (BEF_m) were observed in the native forests across all four climate zones (P < 0.05; Fig. 2a–d). On the contrary, no significant BEF_m relationships were found in croplands (P > 0.05; Fig. 2e–h), suggesting that agricultural deforestation has disrupted the positive BEF_m relationships. In the planted forests, BEF_m relationships were positive and significant in tropical and subtropical zones (warmer zones hereafter), but not in temperate and boreal zones (colder zones hereafter) (Fig. 2i–l), indicating that afforestation has restored the positive BEF_m relationship in warmer zones but not in colder zones. These patterns were consistent regardless of the EMF calculating methods (weighted vs. multi-threshold approaches) (Supplementary Fig. 7).

Fig. 2. The relationships between ecosystem multifunctionality and soil multidiversity.

a–d Show the data of native forests in the tropics, subtropics, temperate, and boreal zones, respectively. e–h Show the data of croplands in the tropics, subtropics, temperate, and boreal zones, respectively. i–l Show the data of planted forests in the tropics, subtropics, temperate, and boreal zones, respectively. Relationships were assessed using simple linear regression analyses with two-sided tests, and shaded areas indicate 95% confidence intervals. P values are reported for all regressions, whereas R² values are provided only for significant relationships. “n” represents the sample size for the linear regression. Soil multidiversity is the average richness of all soil biota groups after min-max normalization, i.e., bacteria, fungi, archaea, viruses, protists, and invertebrates. Source data are provided as a Source Data File.

The effects of afforestation on the BEF_m relationship strength were further confirmed by a moving-window analysis combined with random forest modeling along the south-north latitude gradient (Fig. 3). Mean annual temperature emerged as the strongest predictor of both the natural logarithm of response ratio (LRR planted forest/cropland) and the change in ΔZ (ΔZ = Z_{planted forest} − Z_cropland) of BEF_m relationship strength, followed by the soil water content, elevation, soil pH, soil multitrophic network average variation degree (AVD), etc. (Fig. 3a, c). The effect of afforestation on the BEF_m relationship strength increased significantly with mean annual temperature, regardless of whether EMF was considered by the weighted or multi-threshold approach (Fig. 3b, d). For the individual category functions, the natural logarithm of response ratio (LRR planted forest/cropland) of the BEF relationship strength increased with increasing mean annual temperature, including plant productivity, soil C storage, water regulation, matter decomposition, and nutrient cycling (Supplementary Fig. 8a–e).

Fig. 3. Factors influencing the effect of afforestation on the strength of the relationship between soil multidiversity and ecosystem multifunctionality (BEF_m).

a, c Show the mean importance of influencing factors (estimated using random forest based on the increase in mean squared error (%IncMSE)) for BEF_m relationships based on LRR (log response ratio, ln((R_f + 1)/(R_c + 1))) and ΔZ (difference between Z-transformed correlations in planted forest and cropland), respectively. Sample size n = 324 and 309 for a, c, respectively. R_f and R_c is the Spearman’s correlation coefficient in planted forests and croplands, respectively. The BEF_m relationship strength is quantified as the Spearman’s correlation coefficient between soil multidiversity and ecosystem multifunctionality using a moving window approach. The effects of afforestation on the BEF_m relationships are quantified as the LRR and the ΔZ. Significance in a, c were assessed using rfPermute permutation-derived P values for %IncMSE (500 permutations). * and ** indicate P < 0.05 and P < 0.01, exact P values are provided in the corresponding table in the Source Data. R² suggests the proportion of variance explained by the model. b, d Show the simple linear relationships (two-sided tests) between MAT and the change of BEF_m relationships quantified by LRR and ΔZ, respectively. Weighted, T20, T40, T60, and T80 denote ecosystem multifunctionality quantified using the weighting method and multiple-threshold method at thresholds of 20%, 40%, 60%, and 80%, respectively. Sample sizes (n) for weighted, T20, T40, T60, and T80 were 324, 316, 307, 294, and 196 in (b), and 309, 316, 307, 294, and 196 in (d). MAT mean annual temperature, S genes sulfur genes abundances, SWC soil water content, N genes nitrogen genes abundances, P genes phosphorus genes abundances, Network AVD soil multitrophic network average variation degree, Network complexity, soil multitrophic network complexity. Source data are provided as a Source Data File.

How soil biodiversity affects EMF after afforestation in colder and warmer zones was further examined using structural equation models (Fig. 4). In the colder zone, afforestation directly increased EMF, explaining most of the variance in EMF (Fig. 4a, b, and Supplementary Table 2). Moreover, increased fungal richness induced by afforestation was positively associated with EMF (Fig. 4a and Supplementary Fig. 9d). In the warmer zone, the increased richness of all soil biotic groups except archaea, induced by afforestation, was consistently positively related to EMF (Fig. 4c, Supplementary Fig. 9g–l, and Supplementary Table 3), explaining the majority of variance in EMF in the warmer zone (Fig. 4d). These results underscore a tighter coupling between soil biodiversity and EMF in the warmer zone. Moreover, bacterial diversity indirectly influenced EMF by cascading its impacts on protistan diversity up to invertebrate diversity (Fig. 4c). The increased fungal diversity also affected invertebrate diversity, which in turn facilitated EMF (Fig. 4c). These bottom-up multitrophic interactions was not observed in the colder zone (Fig. 4a) and indicated that afforestation increased EMF by complex multitrophic interactions in the warmer zone.

Fig. 4. Factors influencing ecosystem multifunctionality following afforestation.

a, c Show the piecewise SEM describing the direct and indirect effects of afforestation, climate, soil properties, and soil biodiversity on ecosystem multifunctionality in the colder and warmer zones, respectively. Numbers next to the arrows indicate standardized path coefficients derived from partial regression analyses (two-sided tests), reflecting effect sizes. Red arrows represent negative paths, and blue arrows represent positive paths. b, d Show the relative contribution of multiple predictors in explaining soil ecosystem multifunctionality in the colder (n = 180) and warmer (n = 156) zones, respectively, as determined by multi-model selection. The center of the error bars is the estimated mean standardized regression coefficient of each predictor, with error bars indicating the 95% confidence intervals, calculated as the estimate ± standard error (SE) of the regression coefficients. The percentage of variance explained by each factor quantifies its relative contribution. Statistical significance is denoted by asterisks (*P < 0.05, ** P < 0.01, and ***P < 0.001, based on a two-sided test), exact P values are provided in the corresponding table in the Source Data. Marginal R² (R²m) reflects the variance explained by fixed effects alone, whereas conditional R² (R²c) accounts for both fixed effects and the random effect of sampling site. MAT mean annual temperature, MAP mean annual precipitation, SWC soil water content, BD bulk density. Source data are provided as a Source Data File.

The soil multitrophic co-occurrence networks were established to further explore how multitrophic interactions of soil organisms predict EMF after afforestation in colder and warmer zones (Supplementary Fig. 10 and Supplementary Table 4). Results showed that afforestation significantly enhanced soil multitrophic network complexity in the warmer zone (Supplementary Fig. 10h). The significant positive relationships between network complexity and EMF were observed in warmer zones (P < 0.05) but not in colder zones (Supplementary Fig. 10i, k).

Changes in soil biogeochemical genes and their associations with EMF

Overall, agricultural deforestation decreased the abundance of genes related to C decomposition and methane oxidation, and increased the abundance of genes involved in C fixation and methanogenesis (Fig. 5a, d). Conversely, afforestation generally reversed these trends (Fig. 5b, e). Agricultural deforestation decreased the abundance of the organic N mineralization gene (ureC) and the key genes involved in organic phosphorus (P) mineralization (phnH and phnN), while afforestation reversed this trend (Fig. 5c, f, and Supplementary Fig. 11). Agricultural deforestation increased the abundance of denitrification, nitrification, N fixation, and assimilatory nitrite reduction genes (narG, nirS/K, hao, and nifH, and nirB), while afforestation decreased ammonia-oxidizing and organic N metabolism genes (amoA/B and gdh) (Fig. 5c, f). For sulfur (S) cycling, agricultural deforestation boosted the S reduction genes (dsrA, dsrB, asrA, and asrB). Conversely, afforestation increased S retention by increasing the S oxidation and assimilation genes (soxA, soxC, cysI, and cysJ) (Supplementary Fig. 12).

Fig. 5. The effects of agricultural deforestation and afforestation on the abundance of soil functional genes.

a, b Respectively present heatmaps of differential abundance (log2-fold change, LogFC) of carbon-cycling genes in response to agricultural deforestation (native forests vs. croplands) and afforestation (croplands vs. planted forests) in colder zones, based on metagenomic shotgun sequencing. d, e Respectively present heatmaps of differential abundance (log2-fold change, LogFC) of carbon-cycling genes in response to agricultural deforestation (native forests vs. croplands) and afforestation (croplands vs. planted forests) in warmer zones. c, f Respectively show heatmaps of differential abundance (log2-fold change, LogFC) of nitrogen-cycling genes in colder and warmer zones. The significance of changes in gene abundance was assessed using a two-sided Wald test in DESeq2, with adjusted P < 0.05. Asterisks indicate significance levels (*P < 0.05, **P < 0.01, and ***P < 0.001). LogFC: log2-fold change. Native forest → cropland indicates conversion of native forest to cropland. Cropland → planted forest indicates conversion of cropland to planted forest. Only the top 30 carbon-cycling genes in terms of abundance are displayed. Source data are provided as a Source Data File.

In the colder zone, afforestation significantly increased the abundance of C fixation gene cbbl, C decomposition genes bcsz and CHS1 (Fig. 5b), while this effect was not observed in the warmer zone. Afforestation significantly increased the abundance of the N fixation gene nifH and assimilatory nitrate reduction gene nasA, but decreased the abundance of the nitrification gene hao in the warmer zone (Fig. 5f). However, afforestation significantly reduced the abundance of the nifH gene and did not affect the abundance of the nasA gene and hao gene in the colder zone. Afforestation significantly increased the abundance of P transportation gene pit, polyphosphate degradation genes ppgk and ppk2, but reduced the abundance of organic P mineralization gene aphA in the colder zone (Supplementary Fig. 11b). Conversely, afforestation significantly reduced the abundance of pit gene and did not affect the ppgk, ppk2, and aphA in the warmer zone in the warmer zone (Supplementary Fig. 11d).

In the warmer zone, afforestation drove EMF via regulating the composition of functional genes involved in biogeochemical cycling processes (i.e., C, N, P, and S; Fig. 6b). However, this pattern was not observed in the colder zone, and afforestation directly drove EMF or indirectly affected EMF by altering soil properties in the colder zone (Fig. 6a). These results indicated that the changes in biogeochemical-cycling functional genes induced by afforestation exerted more significant driving effects on EMF in the warmer zone than the colder zone.

Fig. 6. Direct and indirect effects of afforestation, climate, soil properties, and functional gene composition on ecosystem multifunctionality.

a, b Show piecewise structural equation models for colder (n = 180) and warmer (n = 156) regions, respectively. The functional gene compositions were represented by the first axis of NMDS of C, N, P, and S cycling genes. Numbers next to the arrows indicate standardized path coefficients derived from partial regression analyses, reflecting effect sizes. Red arrows represent negative paths, and blue arrows represent positive paths. Statistical significance is denoted by asterisks (*P < 0.05, **P < 0.01, and ***P < 0.001, based on a two-sided test). Marginal R² (R²m) reflects the variance explained by fixed effects alone, whereas conditional R² (R²c) accounts for both fixed effects and the random effect of sampling site. MAT mean annual temperature, MAP mean annual precipitation, SWC soil water content, BD bulk density. Source data are provided as a Source Data File.

Discussion

Afforestation restores soil multidiversity and EMF, yet not to the level of the native forest

Our large-scale paired sites study provides solid evidence that agricultural deforestation reduced soil multidiversity (i.e., the overall diversity of bacteria, fungi, archaea, viruses, protists, and invertebrates) and EMF, while afforestation has partially restored them, but pristine levels have not been reached (Fig. 1b, c). The obvious gap between planted forests and native forests indicates the ecological complexity of native forests and the challenges of full restoration. In our study, native forests exhibit high structural and compositional complexity, including multi-layered canopies and diverse tree assemblages^4,25,26, which may create heterogeneous niches, support diverse soil biota, and enhance key ecosystem processes (e.g., matter decomposition and nutrient cycling). In contrast, the planted forests in our study are predominantly even-aged stands with simplified structure and limited tree-species diversity, and all were established on former croplands with a history of agricultural use, potentially imposing lasting legacy effects that constrain the recovery of soil biodiversity and functions.

Importantly, our study provides metagenomic evidence into the ecosystem function dynamics driven by agricultural deforestation and afforestation, demonstrating that the functional genes involved in biogeochemical cycling processes meditated the response of EMF (Fig. 6b and Supplementary Fig. 3). The altered profiles of biogeochemical-cycling (C, N, P, and S) functional genes after deforestation may suggest a potentially diminished decomposition and mineralization, which is consistent with impaired nutrient turnover and weakened energy flow, both of which are critical for sustaining ecosystem multifunctionality¹⁸. In particular, we observed that agricultural deforestation decreased the abundance of C decomposition, organic N and P mineralization, and methane oxidation genes, and increased the abundance of methanogenesis and denitrification genes, while afforestation generally reversed these trends (Fig. 5), indicating that biogeochemical-cycling functional genes can sensitively respond to land-use changes.

Overall, our findings support afforestation as a viable strategy for enhancing biodiversity and ecosystem functions, but also highlight that afforestation cannot fully replace the native forests in restoration programs. Protecting the remaining native forests plays an irreplaceable role in biodiversity and ecological ecosystem function conservation. Afforestation should complement but not substitute forest conservation. It should be pointed out that belowground communities may exhibit pronounced time lags in recovery, leading to only partial recovery toward native forest conditions due to short timescales. Meanwhile, we admit the potential influence of the stand development stage on biodiversity and functions. Moreover, our measurements focus on the topsoil (0–10 cm), which captures the most biologically active layer but does not reflect deeper soil horizons. Thus, further research should focus on longer timescales and deeper soil layers. Although a paired-site design was employed, residual confounding (e.g., microclimate) cannot be fully excluded due to the observational design of this study. Future controlled experiments are encouraged to explore these factors and test their causal effects on soil biodiversity and EMF.

Agricultural deforestation decouples the positive BEF_m in four climatic zones, while BEF_m recovery is climate-dependent

Our results revealed that agricultural deforestation decoupled the positive soil multidiversity and ecosystem multifunctionality (BEF_m) relationships across four climatic zones (Fig. 2e–h), indicating that agricultural deforestation not only diminishes soil biodiversity and EMF individually but also erodes the capacity of biodiversity to support EMF. The absence of positive BEF_m relationships in croplands can be mainly attributed to two reasons: (1) environmental homogenization. Positive BEF relationships tend to be stronger in heterogeneous and complex environments that facilitate niche differentiation^27,28. The BEF relationships may become non-significant or even negative in homogeneous resource environments^27,28. In contrast to native forests, agricultural deforestation often reduces environmental heterogeneity through practices, such as monoculture planting, tillage and fertilization^3,29, which may weaken or disrupt the BEF_m relationships in croplands. (2) Changes in soil biotic community. In agricultural soils, fertilization (e.g., N and P) and other management regimes often cause resource homogeneity, which favors the competitive dominance of some copiotrophic organisms^3,30. The observed increase in Bacteroidetes, Ascomycota, and Actinobacteria in croplands in our study confirms this (Supplementary Fig. 13c, d). Previous studies have revealed that biotic homogenization can decrease EMF, thereby diminishing the positive associations between biodiversity and multiple ecological functions^31,32. Thus, ecosystem functioning may be decoupled with soil multidiversity as soil biotic communities are dominated by only a few abundant species³⁰, which diminishes facilitative interactions and complementary resource use among soil microorganisms³³, ultimately weakening soil BEF_m linkages in croplands.

Interestingly, our study reveals clear temperature-dependent patterns in BEF_m relationships following afforestation (Fig. 3), and afforestation restores the positive BEF_m relationship in warmer zones (tropics and subtropics) but not in colder zones (temperate and boreal zones). These results indicate that planted forests in warmer zones were more dependent on biodiversity to support ecosystem function. In contrast, planted forests in colder zones did not exhibit significant BEF_m relationships, suggesting that the recovery of BEF linkages is more fragile or constrained in these environments. These results highlight the importance of incorporating the content-dependent effects into the BEF relationship study^24,34–36. Metabolic theory may provide a potential explanation for these patterns^37,38. Metabolic activity and population growth rate increase exponentially with temperature^37–39, leading to higher microbial species turnover, activity, and growth rate at higher temperatures^38,40,41. The enhanced microbial dynamics with temperatures may be associated with changes in key ecological processes and functions⁴¹, which may potentially enhance biodiversity-function coupling and temperature-dependent patterns. The climate dependency of BEF_m recovery indicates that restoration strategies must take into account regional climate differences, and longer expected recovery times is required for ecosystem restoration projects in colder zones to restore the BEF_m.

Soil multitrophic cascades and biogeochemical genes contribute to the recovery of BEF_m relationships in warmer zones

Our structural equation modeling further indicates the reasons for the observed climate-dependent BEF_m patterns (Figs. 4 and 6). In the warmer zone, we found that the increased richness of all soil biotic groups (except archaea) was positively associated with EMF, and the bottom-up multitrophic interactions enhanced EMF (Fig. 4c), suggesting that afforestation is associated with potential multitrophic cascades that may enhance EMF and contribute to BEF_m relationships^42–44. One possible explanation is that complex multitrophic interactions may contribute to promote trophic complementarity by differential resource utilization and predation by distinct predators⁴⁵, then facilitating a division of labour that enhances the functions of complex ecosystems⁴⁶. The significant positive effects of the multiple trophic groups on EMF in the warmer zone partly indicate that species within trophic groups may be complementary⁴². The observed bottom-up recovery of soil micro-food web in the warmer zone may strengthen the linkage between restored soil biodiversity and EMF, given that many ecological functions emerge from coordinated interactions among multiple trophic levels^24,43,47,48. For example, the predator-prey interactions between soil invertebrates and lower trophic groups may promote trophic complementarity, reduce the functional dominance of some individual groups, and maintain trophic balance across soil biota¹¹, which may mediated ecological functions and amplify the contribution of soil biodiversity to EMF. Our multitrophic network analysis provides further support for our inference that multitrophic interactions play a key role in driving EMF and mediating BEF_m relationships. As we found, network complexity was positively correlated with EMF only in the warmer zone, and the network AVD was identified as an important factor influencing the effect of afforestation on the BEF_m relationship strength (Fig. 3). The importance of multitrophic interactions in restoring BEF_m relationships indicates that soil multitrophic organisms should be considered as an essential goal when making restoration policies in the future.

Meanwhile, our results further indicate that the soil biogeochemical cycling gene composition may be linked to the recovery of BEF_m relationships following afforestation in the warmer zone (Fig. 6b). In warmer zones, elevated temperature may enhance microbial metabolism and enzymatic activity, which may enhance coupling among biogeochemical processes^49,50. Meanwhile, higher metabolic rates are often accompanied by greater diversity and expression of functional genes involved in nutrient acquisition and transformation, which may potentially promote microbial metabolic complementarity⁵¹. These reasons may explain the tighter linkage between soil biodiversity and ecosystem multifunctionality in the warmer zone. However, it should be pointed out that metagenomic data only reflect potential metabolic capacity rather than the in situ functions. Additionally, unmeasured confounding factors (e.g., cropland management and fertilization history) may influence the recovery of BEF_m relationships.

However, in the colder zone, the multitrophic interactions have not yet been found. Instead, the increase in fungal diversity caused by afforestation promotes the enhancement of EMF (Fig. 4a), indicating that soil fungi play a vital role in maintaining EMF in the colder zone. In general, the higher trophic-level organisms (e.g., protists and invertebrates) are more susceptible and sensitive to environmental changes, especially temperature⁵², than lower trophic-level organisms due to their longer generation period and lower fecundity^48,52,53. Meanwhile, low temperatures may constrain microbial metabolism and nutrient turnover, thereby weakening the contribution of microbial biogeochemical processes to EMF (Fig. 6a). Collectively, these constraints may explain why BEF_m relationships in planted forests remained weak in the colder zone. However, from a long-term recovery timescales perspective, especially in the context of climate warming, we assume that the BEF_m relationships in colder zones may be expected to gradually restore. We admit that, although our SEM provides insights into the associations between trophic groups and EMF and is widely applied in this field^23,30,33,35, the statistical relationships identified here remain correlational rather than causal relationships. To clarify the causality, future studies should use manipulated experiments to directly test the effects of biodiversity, multitrophic cascades, and the biogeochemical genes.

Methods

Study sites and experiment design

We sampled paired native forests, croplands converted from native forests, planted forests converted from croplands from an ~4000 km south-north transect in China, spanning latitudes from 18.69° N to 52.97° N and longitudes from 100.9° E to 128.99° E (Fig. 1a). This broad geographic transect spans across the tropical, subtropical, temperate, and boreal zones, with significant spatial variability in climate. The mean annual temperature ranges from −5 to 25 °C, and mean annual precipitation varies between 380 and 1905 mm across the study sites.

To explore the effects of agricultural deforestation and afforestation on soil biodiversity, metagenomic functional genes, individual ecological functions, EMF, and their relationships, we investigated 23 pairs (i.e., 23 study sites) of paired native forests-croplands-planted forests and 33 pairs of paired planted forests-croplands (Fig. 1a). At each pair, 3 plots were sampled for each type of ecosystem. Thus, a total of 405 plots were sampled, i.e., 23 pairs × 3 types of ecosystems × 3 plots + 33 pairs × 2 types of ecosystems × 3 plots = 405 plots. There are 90, 102, 108, and 105 plots in the tropical, subtropical, temperate, and boreal zones, respectively. Planted forests are even-aged monocultures converted from dry croplands, primarily located within forestry farms. The forest age information was obtained from the landowners. The age of planted forests ranged from 14 to 27 years, with a mean of 22 years, and 80% of plots are between 17 and 25 years. There is no significant difference in the age of planted forests in different climate zones (P = 0.22). The paired adjacent croplands are dry croplands historically converted from native forests and cultivated continuously with stable management regimes for at least 30 years, predominantly planted with maize, and thus represent the control sites pre-afforestation. It should be pointed out that unmeasured agricultural management practices, such as fertilization and pesticide application, may still influence soil communities and functions⁵⁴. However, this is an unavoidable limitation in large-scale research. The paired-sites approach can diminish these impacts to the greatest extent possible. Each pair (native forests-croplands-planted forests or planted forests-croplands) was carefully selected to ensure comparability. The pairs were located in close distance and shared the same topography, soil type and original vegetation, minimizing the confounding effects from climate, soil types, and agricultural management practices. Thus, within-site pairs differences effectively reflect the impacts of agricultural deforestation and afforestation. The paired native forests, mainly located within national nature reserves or national forest parks, are primary or old secondary forests without a history of land use change, and are situated within 5 km of the paired planted forests-cropland systems. Since native forests do not exist at every study site, 23 pairs of paired native forests-croplands-planted forests were investigated (Fig. 1a). The completeness of native forest coverage is a limitation in our study. However, the native forest sites cover ~4000 km south-north transect, with at least five sites in each climate zone (Figs. 1 and 2).

Within a pair, three plots (10 × 10 m²) were randomly selected within each ecosystem type (i.e., native forest, cropland, planted forest) during the vegetation growing season (July–August 2023). Soil sampling within each pair was conducted on the same day to minimize temporal variability. The distance between any two plots was greater than 100 m but less than 2 km to avoid potential confounding effects from local environmental factors. Soil samples were collected by five cores (0–10 cm depth) from each plot, mixed thoroughly, and sieved through a 2.0-mm mesh. Meanwhile, cutting ring samples (100 cm³) were collected for measuring bulk density and soil water content. Soil parent material is extracted from the Chinese Soil Scientific Database (http://vdb3.soil.csdb.cn). Climate data were obtained from the WorldClim Global Climate Database. The soil samples were then divided into three portions: one part was immediately preserved in liquid N for DNA extraction, a second portion was stored at 4 °C for subsequent analyses of soil enzyme activities, nitrate (NO₃⁻-N) and ammonium (NH₄⁺-N) content, while the third portion was air-dried, ball-milled, and sieved for the measurement of other physicochemical properties.

Measurement of soil physiochemical properties and extracellular enzymes

Soil pH was determined in a 1:2.5 (w/v) soil-to-water suspension with a pH electrode. Soil organic C (SOC) and total N contents were determined by an elemental analyzer. Soil water content (SWC) was measured by oven-drying soils at 105 °C to constant weight. Soil bulk density (BD) and field water-holding capacity (FWC) were measured using the cutting ring method (volume of g cm⁻³). The soil NO₃⁻-N, NH₄⁺-N, and available P content were measured by the colorimetric methods⁵⁵. Soil extracellular enzymes involved in C (α-1,4-glucosidase, β-1,4-glucosidase, xylanase, peroxidase, cellobiohydrolase, and polyphenol oxidase), N (N-acetyl-β-D-glucosaminidase and leucine aminopeptidase), and P (acid phosphatase) cycling were measured by fluorimetric methods^56,57. The SOC stock was calculated as follows⁵⁸:

$$\mathrm{Soil}\mathrm{C}\mathrm{stocks}=\left(\mathrm{Soil}\mathrm{C}\mathrm{content}\right)\times \mathrm{BD}\times \left(\mathrm{soil}\mathrm{depth}\right)$$

Amplicon sequencing and soil biodiversity

Soil DNA was extracted using the TGuide S96 Magnetic Soil/Stool DNA Kit following the instructions. DNA concentration in the soil samples was determined using the Qubit dsDNA HS Assay Kit and Qubit 4.0 Fluorometer. Soil biota community composition was analyzed by targeting the V4-V5 region of 16S rRNA for bacteria, the ITS1 region for fungi, and the V4 region of 18S rRNA for protists and invertebrates, using the primers 515F/926R (5′-GTGYCAGCMGCCGCGGTAA-3′, 5′-CCGYCAATTYMTTTRAGTTT-3′), ITS1F/ITS2 (5′-CTTGGTCATTTAGAGGAAGTAA-3′, 5′-GCTGCGTTCTTCATCGATGC-3′), and TAReuk454FWD1/TAReukREV3 (5′-CCAGCASCYGCGGTAATTCC-3′, 5′-ACTTTCGTTCTTGATYRA-3′), respectively. Sequencing was performed on the Illumina NovaSeq 6000 platform.

The raw sequencing data were processed according to the standard operating procedure outlined in ref. ⁵⁹ using QIIME2 (2023.5 core distribution). Briefly, sequences were denoised and classified into amplicon sequence variants (ASVs) using the DADA2 plugin. Taxonomy assignment to each ASV was performed using the SLIVA v138 database for bacteria, protists, and invertebrates, and the UNITE database for fungi. Protists were defined as all eukaryotic taxa, excluding fungi, invertebrates (Metazoa), and vascular plants (Streptophyta)¹⁷. The fungal functional groups were identified using FungalTraits⁶⁰ from the ASV tables.

Species richness, defined as the total number of soil phylotypes (i.e., ASVs), was used as the biodiversity metric^16,17,61. Richness was derived from ASV tables for the bacteria, fungi, protists, invertebrates, archaea, and viruses. The richness of soil archaea and viruses was obtained from shotgun metagenome data. Given that soil viruses represent an important component of the soil microbiome and contribute to ecosystem functioning across biomes^62,63, we included viruses together with that of bacteria, fungi, protists, invertebrates, and archaea to capture overall soil biodiversity (i.e., multidiversity). Multidiversity was then calculated as the mean of min–max normalized richness scores (scaled between 0 and 1) across the six biota groups using a widely used approach^16,22,39:

$$\mathrm{Soil\; multidiversity}=\frac{1}{6}{\sum }_{i=1}^6\left(\frac{{\mathrm{rawDiversity}}_i-\min \left({\mathrm{rawDiversity}}_i\right)}{\mathrm{ma}\mathrm{x}\left({\mathrm{rawDiversity}}_i\right)-\min \left({\mathrm{rawDiversity}}_i\right)}\right)$$

Meanwhile, we tested the sensitivity of multidiversity to the contribution of each group by performing a leave-one-out analysis, recalculating multidiversity by excluding one biotic group at a time. We also calculated multidiversity using alternative standardization methods (Z-score standardization) to evaluate the robustness of our results. The consistently high correlations indicate that multidiversity is robust to the exclusion of any single group (Supplementary Fig. 14a). Especially, for different standardization methods (max-min vs. Z-score standardization), the two metrics are nearly identical (R² = 0.99, P < 0.001) (Supplementary Fig. 14b), demonstrating that multidiversity is insensitive to the choice of standardization method. In most cases, the core patterns remained unchanged (Supplementary Table 5), with the exception that excluding fungi weakened the BEF_m relationship in native forests of the temperate zone, demonstrating that our central patterns are robust to methodological choices and the specific taxa included.

It should be pointed out that although the sequencing-based approach is widely used to assess soil biodiversity, it may introduce biases arising from differences in taxonomic resolution and marker coverage. This method may also under-represent larger invertebrates, though they are increasingly applied to quantify soil invertebrate diversity⁶¹. Additionally, although rarefaction is widely used to normalize sequencing depth, it may lead to data loss and biased diversity estimates. Nevertheless, sensitivity analyses indicated that our main conclusions were robust to the choice of rarefaction (Supplementary Table 5).

Shotgun metagenome sequencing

Metagenomic sequencing was further conducted to analyze the soil biota functional potential. A paired-end library was constructed using NEXTFLEX®Rapid DNA-Seq. The library was sequenced on an Illumina Xplus at Majorbio Bio-Pharm Technology Co., Ltd. according to the manufacturer’s instructions. In total, 2.96 × 10¹⁰ clear reads were sequenced across samples, yielding 4444.5 Gbp of total sequence information with an average data volume of about 11 Gbp per sample. Metagenomic data were processed using a standardized pipeline^64–66. Briefly, Fastp (v 0.23) was used to perform quality control, and contigs were assembled using MEGAHIT (v1.1.2)⁶⁴. The open reading frames (ORFs) of contigs were predicted using Prodigal (v2.6.3)⁶⁶. Representative sequences of the non-redundant gene catalog were aligned against the NR, KEGG, and CARD databases using Diamond (v 2.0.13)⁶⁷ with an e-value cutoff of 1e⁻⁵, to perform taxonomic, KEGG, and antibiotic resistance annotation.

Assessment of individual ecological functions and EMF

We measured 19 surrogates of ecological functions (six categories) that are affected by soil organisms and widely used in EMF research^16,17 (Supplementary Table 6). Collectively, these variables serve as reliable proxies for ecosystem functions and key processes/properties associated with soil biodiversity, nutrient cycling, biological productivity, and plant health^16,17. The reasons for the selection of these variables are shown in Supplementary Table 6. Briefly, the mean normalized difference vegetation index (NDVI) was employed as a proxy for plant productivity^17,67, derived from cloud-free Sentinel-2 imagery (10 m spatial resolution) and calculated as the mean value of NDVI in the sampling year (i.e., 2023), including all phenological periods of ecosystems. Although using NDVI as a proxy indicator has some limitations, our further analyses excluding NDVI from the EMF calculation yielded consistent patterns (Supplementary Figs. 15d and 16), indicating that our results are robust and that the inclusion of NDVI does not introduce a systematic bias in EMF estimation. Soil C storage was assessed based on SOC stocks. Soil water regulation was quantified by FWC and soil porosity. Soil FWC reflects multiple dimensions of soil water and regulates plant-water provision⁶⁷, and soil porosity controls infiltration and water storage capacity³⁴. Organic matter decomposition was characterized by the activities of nine extracellular enzymes involved in C, N, and P cycling. For soil nutrient cycling, we measured soil available P, total N, NO₃⁻-N, and NH₄⁺-N^34,68. Plant-soil mutualism was evaluated by the relative abundance of ectomycorrhizal fungi (ECM) and arbuscular mycorrhizal fungi (AMF)^16,69.

The EMF of each plot was quantified using three complementary approaches: the averaging method, the multiple-threshold method, and a weighted multifunctionality approach. Following Byrnes et al.⁷⁰, the averaging methods were based on standardizing all 19 ecological functional surrogates to a 0–1 range and calculating their mean value. We then calculated the weighted EMF, using agglomerative cluster analysis to down-weight highly correlated functions (r = 0.67; Supplementary Fig. 15a)⁷¹. The three EMF calculation approaches were highly correlated (Supplementary Fig. 15b, c), suggesting that our results are robust to the choice of EMF calculation approach. Accordingly, we report results based on weighted multifunctionality, which explicitly accounts for redundancy among correlated ecosystem functions. For the multiple-threshold approach, we used the thresholds of 20, 40, 60, and 80% to analyze how soil biodiversity affects EMF at different levels of thresholds of the observed maximum functioning⁷⁰.

Statistical analysis

Linear mixed-effects models (LMMs) were used to evaluate the effects of agricultural deforestation and afforestation on soil biodiversity, multidiversity, individual functional categories, and EMF. Ecosystem type (native forest, cropland, and planted forest) was included as a fixed factor. To account for site-level heterogeneity and potential confounding influences, plots nested within sites was included as random effects, and parent material as well as plantation species were included as fixed covariates. All LMMs were implemented using the lme4 (1.1.31) package in R (4.2.2). Post hoc pairwise comparisons among ecosystem types were performed using Tukey’s HSD in the emmeans (2.0.3) package with adjustments for multiple testing. Given that the forest age was not fully consistent for all the planted forests, we used linear mixed effects models to analyze the effects of forest age on the EMF and multidiversity. Forest age was taken as a fixed factor, and climate zones were considered a random factor. The results indicated that variations in forest age had no significant impact on either multidiversity or EMF (Supplementary Table 7 and Supplementary Fig. 5b, d).

Ordinary least squares linear regression was applied to examine the relationships between soil multidiversity and EMF for different ecosystem types in four climate zones, respectively. Results showed that, in tropical and subtropical planted forests, EMF is positively correlated with multidiversity, but not in temperate and cold zones (Fig. 2i–l and Supplementary Fig. 7i–l). Thus, we further explored the drivers of afforestation’s effects on the BEF_m relationship using moving-window analysis combined with a random forest model. We first performed a moving-window analysis. To maintain sufficient samples, we used window sizes of 12 consecutive samples across sites, generating 325 (e.g., 1–12, 2–13, …, 325–336) data points. For each sub-dataset (windows), we estimated the BEF_m relationship for cropland and planted forest, which was characterized by the Spearman’s correlation coefficients between multidiversity and EMF (both weighted approach and multiple-threshold approach)⁶⁹. The P-values were adjusted for multiple comparisons using the “fdr” method to control for the chance of false positives. Stronger positive correlations indicate a higher capacity of soil biodiversity to promote EMF.

To quantify the magnitude of the afforestation’s effects on the BEF_m relationship strength, we calculated the natural logarithm of the response ratio (LRR) between planted forests and croplands following Eq. 1. Given that Spearman’s correlation coefficients can have negative values, we adjust Spearman’s correlation coefficient by adding 1 before calculating the LRR. We also calculated the LRR for EMF and soil multidiversity using the following Eq. 2:

$$\mathrm{LRR}=\ln \left(R_{\mathrm{f}}+1\right)/\left(R_{\mathrm{c}}+1\right)$$ 1

$$\mathrm{LRR}=\ln \left({\mathrm{X}}_{\mathrm{f}}/{\mathrm{X}}_{\mathrm{c}}\right)$$ 2

Where R_f and R_c is the Spearman’s correlation coefficient in planted forests and croplands, respectively. The X_f and X_c are the mean value of EMF (or soil multidiversity) in planted forests and croplands, respectively.

Given that the magnitude of Spearman’s correlation coefficient approaches ±1, its distribution becomes increasingly skewed. We thus applied Fisher’s Z-transformation to the correlation coefficients in each window for cropland and planted forest following Eq. 3:

$$Z=1/2\ln \left(1+r\right)/\left(1-r\right)$$ 3

Then, the afforestation’s effects on the BEF_m relationship strength were assessed as the change of transformed correlation coefficients (ΔZ = Z_{planted forest} − Z_cropland) in planted forest and cropland. The ΔZ serves as an effect size, representing the change in the BEF_m relationship strength due to afforestation. Positive values indicate that afforestation strengthens the BEF_m relationship, whereas negative values suggest a weakening effect. Random forest analysis was subsequently used to determine the main factors influencing how afforestation modifies the BEF_m relationship. All predictors used in the Random Forest and regression analyses are shown in Supplementary Table 8.

Meanwhile, to further evaluate whether our conclusions depend on the degree of overlap among windows, we performed a systematic sensitivity analysis in which we fixed the window size at 12 but varied the step size (i.e., 1, 2, 4, 6, and 12), spanning highly overlapping to fully nonoverlapping windows. Across all step sizes, mean annual temperature consistently emerged as the dominant driver of the effect of afforestation on the BEF_m relationship strength, indicating that the main conclusions are robust to the choice of window overlap.

Given that afforestation restored the positive soil multidiversity and ecosystem multifunctionality (BEF_m) relationship in tropics and subtropics, but not in temperate and boreal zones. Mean annual temperature (MAT) as the strongest predictor of this climate-dependent BEF_m recovery pattern (Fig. 3). Accordingly, all study sites were grouped into warmer and colder zones, and separate piecewise structural equation models were further independently constructed for the colder and warmer zones to disentangle the direct and indirect contributions of climatic factors, soil properties (pH, SWC, and BD), soil biodiversity, and afforestation on EMF, using the psem function of the piecewiseSEM (2.3.0) package. To ensure comparability between the colder and warmer zones, we maintained a consistent prior model including all reasonable pathways based on prior knowledge (Supplementary Fig. 17 and Supplementary Table 9). Structural equation modeling was conducted using linear mixed-effects models, with plots nested within sites included as random effects. The model was refined by stepwise removal of non-significant paths (P > 0.05), and its adequacy was evaluated using Fisher’s C test (P > 0.05). The residuals of all sub-models in the SEM did not exhibit spatial autocorrelation higher than expected by chance (Supplementary Table 10). This indicates that spatial dependence among nearby plots was not an issue. Spatial autocorrelation was evaluated using Moran’s I implemented in the spdep (1.3.3) packages in R. Similarly, a piecewise structural equation model was used to investigate the direct and indirect effects of climate, soil properties, and functional gene composition (represented by the first axis of NMDS of C, N, P, and S cycling genes) on EMF in the colder and warmer zones following afforestation.

The relative contribution of multiple factors on EMF and individual category functions was determined by multi-model selection, and the data underwent standardization before analyses. To minimize the potential effects of high multicollinearity among variables on model performance, the variables with variable inflation factors exceeding 5 were removed. The best model predicting EMF and individual category function was selected based on the AIC ($△$AIC < 3) using the MuMIn (1.48.4) package. Finally, we selected the resulting models and performed model averaging to mitigate the potential instability of any single model, ensuring more robust and reliable inference.

Soil biota networks were constructed separately in the warmer and colder climate zones to assess potential intratrophic interactions among major biotic groups (i.e., archaea, bacteria, fungi, viruses, protists, and invertebrates) using SparCC⁷², an algorithm that is compositionally robust for inferring correlations from high-dimensional microbial community data. Prior to network inference, only dominant and ubiquitous phylotypes were retained, defined as the 10% most abundant phylotypes for all taxa, present in >50% of all samples for prokaryotes and viruses, and present in >10% of all samples for eukaryotes^16,61. Zeros were retained as true absences following this filtering, with no additional zero replacement applied. Then, networks were constructed separately for croplands and planted forests within the warmer and colder climate zones, resulting in four distinct networks. SparCC correlations were estimated using five inference iterations, and statistical significance was assessed using two-sided pseudo P-values derived from 100 bootstrap permutations generated with MakeBootstraps.py and PseudoPvals.py. Only robust associations (|ρ| ≥ 0.65) with false discovery rate (FDR)-adjusted pseudo P-values < 0.001 were retained for network construction^16,61. Gephi (v.01) was used for visualization. Then, the network topological features of each sample were extracted by preserving the phylotypes in each sample using the “subgraph” function in the igraph (2.0.3) package, including nodes, edge, average degree, average path length, network diameter, graph density, and clustering coefficient^61,73. Then, we analyzed an index reflecting the network complexity based on the mean of standardized topological feature scores⁶¹. The average path length and network diameter denote the network sparsity, which were calculated as the inverse of the variables (× −1) before calculating the index⁶¹. To further compare the stability of networks among ecosystem types, we calculated the AVD to reflect the biota network stability⁷⁴. Higher AVD values suggest greater variability and fluctuations in the community, whereas lower AVD values indicate higher community stability. The AVD index reflects relative variability within the inferred networks, rather than dynamical or temporal stability in an ecological sense.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

K.Z.: conceptualization, funding acquisition, and writing-review & editing. Z.Y., J.S. and L.C.: investigation, data curation, formal analysis, visualization, and writing-original draft. X.Z., Y.Z., F.L., X.C., X.W. and Q.Z.: methodology and writing-review & editing.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This research was supported by the National Natural Science Foundation of China (32471650, 32130069).

Data availability

The raw data generated in this study have been deposited in the Figshare repository (10.6084/m9.figshare.32657256). All display items presented in the main manuscript, supplementary information, and Supplementary Data 1 can be reproduced from raw data and code that have already been shared in a public repository. The raw sequence data of bacteria, fungi, protists, invertebrates, and shotgun metagenome were deposited to National Center for Biotechnology Information under PRJNA1250129, PRJNA1250134, PRJNA1250136, and PRJNA1250137, respectively. The source data generated in this study are provided in the Source Data file. Source data are provided with this paper.

Code availability

R code is available via Figshare at 10.6084/m9.figshare.32657256⁷⁵.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains Supplementary material available at 10.1038/s41467-026-75740-z.