Abstract
Forest ecosystems contribute substantially to global terrestrial primary productivity and climate regulation, but, in contrast to grasslands, experimental evidence for a positive biodiversity-productivity relationship in highly diverse forests is still lacking1. Here, we provide such evidence from a large forest biodiversity experiment with a novel design2 in subtropical China. Productivity (stand-level tree basal area, aboveground volume and carbon and their annual increment) increased linearly with the logarithm of tree species richness. Additive partitioning3 showed that increasing positive complementarity effects combined with weakening negative selection effects caused a strengthening of the relationship over time. In 2-species mixed stands, complementary effects increased with functional distance and selection effects with vertical crown dissimilarity between species. Understorey shrubs reduced stand-level tree productivity, but this effect of competition was attenuated by shrub species richness, indicating that a diverse understorey may facilitate overall ecosystem functioning. Identical biodiversity-productivity relationships were found in plots of different size, suggesting that extrapolation to larger scales is possible. Our results highlight the potential of multi-species afforestation strategies to simultaneously contribute to mitigation of climate change and biodiversity restoration.
Forest ecosystems harbor around two thirds of all terrestrial plant species, but currently lose biodiversity at high rates which may threaten the production of timber, fiber, fuel and other services beneficial to humans4. Observational studies suggest that species-rich forests exceed the productivity of less diverse forests5,6, but co-varying factors (e.g. spatial heterogeneity in abiotic environment, species composition and successional stages; interventions by forest management) make assigning causation difficult. Systematic experimental manipulations of plant species composition in grassland communities7–9 have demonstrated that plant diversity promotes community productivity. This effect has been attributed to positive effects of niche partitioning between species, specifically to complementarity in the use of abiotic resources10 or interactions with enemies11, or to an increasing contribution of highly productive species in more diverse communities12. These two types of mechanisms have been related to statistical complementarity and selection effects obtained by additive partitioning3. However, these mechanisms may differ in species-rich forests in which neutral processes may be important13,14 and where “diffuse” coevolution may result in niche convergence toward generalist strategies15. Furthermore, trees have large and persistent vertical structures that support the long-term accumulation of biomass. Several forest experiments have recently been initiated16,17, but these are mainly in the temperate zone or implemented in small plots with a limited species richness gradients18–23. To close these critical gaps in knowledge1, controlled experiments in which the diversity of tree species is systematically manipulated are needed. The largest such study concerning numbers of treatments and plots has been established in 2009/2010 in subtropical south-east China and is referred to as the BEF-China experiment2.
Here, we report how stand-level productivity in the BEF-China experiment 3-7 years after planting was related to species richness and how variation within species-richness levels was related to trait differences among species. Experimental forest communities were constructed systematically from a pool of 40 tree (Extended Data Table 1) and 20 shrub species, and were established in plots at two hilly sites (in 2009 at site A and in 2010 at site B). By the time of our later measurements the tree communities were well established with some canopies exceeding 12 m in height in 2016. The design of previous biodiversity experiments had been criticised because not all species were found at all diversity levels, and because the compositions of the experimental communities that were realized were not nested as would be expected with sequential extinction24. We adopted a novel design that avoided these caveats2 (see Methods, Extended Data Fig. 1, Extended Data Table 2). In brief, we first created three pools of 16 species per site. These were then repeatedly split into halves, resulting in nested, non-overlapping subsets of 8, 4, 2 and 1 species. We used these sets, and in addition also the full sets of 24 species per site, to plant tree communities comprising 1 to 24 species. We further established plots with two sizes: 0.067 ha (equivalent to the Chinese area unit of 1 mu; 400 individual trees) and 0.267 ha (4 mu; 1600 individuals). The larger plots were established for one of the three 16-species pools at each site and included a split-plot treatment that consisted of understorey shrubs planted in the centre of the quadrats formed by four neighbouring trees. Shrubs were planted at a species richness of 0 (no shrubs), 2, 4 or 8, in factorial combination with the tree species-richness treatment. We assessed stand-level tree productivity in all 1-mu plots (including all 1-mu subplots of the larger plots) non-destructively by measuring stem basal area and height of the 16 central trees every year from 2013-2016 in September/October. We used these data, together with data from separately harvested trees to obtain conversion factors, to calculate tree volume and aggregated the individual volume data of live trees to the stand level. To characterize annual stand growth, we further derived yearly increments of stand volume from successive inventories. Using the same method, we determined the same metrics at the population-level (stand-level data separated into species).
Notes:
Fixed effects were fitted sequentially (type-I sum of squares) as indicated in the table (random terms were community composition, plot, subplot and the interaction of these with year, with site-specific variance components for species composition and plot). Abbreviations: n = numbers of plots in analysis; df = nominator degree of freedom; ddf = denominator degree of freedom; logSR = log2(tree species richness). F and P indicate F-ratios and the P-value of the significance test.
We found significantly positive effects of the logarithm of tree species richness on both stand volume and annual stand volume increment of trees (F1,89 = 5.26, P = 0.024 and F1,94 = 9.34, P = 0.003, respectively; Fig. 1 and Extended Data Fig. 2, Table 1). The size of these effects increased over time (F1,95 = 10.83, P = 0.001 and F1,95 = 12.01, P < 0.001, respectively, for interaction species richness × year). Similar results were obtained for stand basal area and its increment (Extended Data Fig. 3, Table 1). In 2016, at the end of our measuring period, stand basal area increased on average by 1.65 m2 ha−1 and stand volume by 5.09 m3 ha−1 with each doubling of tree species richness. After seven years of growth, the average 16-species mixture stored 22.0 ± 4.5 Mg C ha−1 above ground, which is double the amount found in monocultures (9.4 ± 1.1 Mg ha−1, Extended Data Fig. 4) and similar to the productivity of monocultures of commercial plantation species Cunninghamia lanceolata (22.4 ± 10.7 Mg C ha−1) and Pinus massoniana (21.0 ± 3.0 Mg C ha−1) that we had planted for reference at the same site (Extended Data Fig. 4, Extended Data Table 4). System-level C sequestration likely is higher, given that additional C will have been allocated to belowground tree organs25 and in part transferred to persistent soil pools important for long-term carbon sequestration. These strong positive effects of tree species richness were driven by faster growth of live trees in more diverse stands, and were unrelated to tree survival rate, which was independent of species richness; if anything, there was a trend towards lower survival at higher richness (Extended Data Fig. 5).
The net biodiversity effect26 on productivity increased through time for mixtures of all species-richness levels (Fig. 2, F1,48 = 23.61, P < 0.001). The positive effects of tree species richness on productivity were also reflected in a higher frequency of mixtures that overyielded relative to the ones that underyielded and in many cases of transgressive overyielding26 (Extended Data Table 5). Additive partitioning revealed that the increases of net biodiversity effects were primarily driven by increases in complementarity effects (Extended Data Table 6, F1,31 = 9.61, P = 0.004) and weakening negative selection effects (Extended Data Table 6, F1,37 = 4.61, P = 0.038). In the last year of measurements, selection effects were no longer significantly different from zero (Fig. 2, F1,31 = 3.40, P = 0.075).
We observed considerable variation in overyielding among communities of the same species-richness level. Some of this variation was explained by functional diversity but phylogenetic diversity had low explanatory power. For the 48 different 2-species mixtures, complementarity effects were positively correlated with the functional distance and selection effects with vertical crown dissimilarity, also referred to as crown complementarity between species (Fig. 3, Extended Data Table 7). That vertical crown complementarity22 contributed to overyielding via selection rather than complementarity effects indicated that it was due to asymmetric light competition27 and is consistent with the “competition-trait hierarchy hypothesis”28.
Species with high monoculture productivity (Fig. 4a) explained large amounts of variation in stand-level productivity (Fig. 4b), but their contribution was not always positive, as demonstrated by several negative species-level selection effects (Fig. 4c). Despite the positive effect of species richness on community productivity, the population-level responses of each species to species richness varied from positive to neutral to negative (Fig. 4d). These responses did not differ between evergreen and deciduous species (Fig. 4d, F1,159 = 0.89, P = 0.347). A similar decoupling between community- and population-level responses has previously been reported from grassland biodiversity experiments8 and indicates that a few positive population-level responses can overcompensate a larger number of negative population-level responses. Nevertheless, the number of species with positive responses to community diversity and the magnitude of their responses increased with time (Fig. 4d).
Competition by understorey shrubs planted in the gaps between the trees reduced stand-level tree volume (Extended Data Table 8, F1,234 = 4.80, P = 0.029), but this effect decreased with shrub species richness (Extended Data Table 8, F1,499 = 5.40, P = 0.022) and was negligible when mixtures of 8 shrub species were planted (Extended Data Fig. 6), indicating reduced competition between shrubs and trees at higher shrub diversity levels. The diversity-productivity relationships we found were scale-independent, i.e. they did not differ between 1- and 4-mu plots (Extended Data Table 8, F1,114 = 0.20, P = 0.694 for interaction species richness × plot size).
Our results provide strong evidence for a positive effect of tree species richness on tree productivity at stand-level in establishing subtropical forest ecosystems, and support the idea that highly diverse subtropical forest ecosystems are niche-structured22,27. Seven-year old mixed-species stands can produce an estimated additional aboveground wood volume of 25 m3 ha−1 relative to the average monoculture, which translates to the sequestration of approximately an extra 10 Mg C ha−1 (Fig. 1, Extended Data Fig.4). We expect this effect to grow further, given that we did not observed any signs of a deceleration over the present measurement period.
The size of the biodiversity effects we found for these forests is similar to biodiversity effects reported from grassland studies8,9. Given that plant biomass is higher in forests, and that the largest fraction of tree carbon is bound in relatively persistent woody biomass, these effects translate into significant diversity-mediated rates of carbon sequestration. Substantial forest areas are managed world-wide, with large afforestation programs underway in many countries. In China, huge economic efforts are made for afforestation, with a net growth of total forested area by 1.5×106 ha yr−1 achieved from 2010 to 2015 29. However, the overwhelming fraction of newly established forests are monoculture plantations of species with highest productivity in the short term30. Our analysis suggests that a similar productivity could be achieved with mixed plantations of native species, which would result in co-benefits in the form of biodiversity management and a likely higher level and stability of ecosystem services in the longer term.
Online content
Methods, along with additional Extended Data display items and Source Data, are available in the online version of the paper; references unique to these sections appear only in the online paper.
Extended Data are available in the online version of the paper.
METHODS
Study site and experimental design
The BEF-China experimental platform was established in Jiangxi Province, China (29°08′-29°11′N, 117°90′-117°93′E). Climate at the site is subtropical, with mean annual temperature and precipitation of 16.7°C and 1800 mm, respectively (averaged from 1971-2000)31. A large-scale tree biodiversity experiment was established in 2009-2010 on two sites (A and B) of approximately 20 ha each, with a total of 226’400 individual trees planted. Here, we use all plots in which random species-loss scenarios were simulated. The species pool contains 40 tree species (Extended Data Table 1), 24 for each site (of which eight are shared between sites). The 24 species at each site were divided into three 8-species sets. By combing these 8-species sets in all possible ways, three pools of 16 species were created. The species in each 16-species pool were put in random sequence and then repeatedly divided in halves until monocultures were obtained. This procedure resulted in 70 unique species compositions per site (Extended Data Table 2) and ensured that each tree species occurred in equal overall proportion at each diversity level. We further included monoculture plots with two commercially important tree species, Pinus massoniana and Cunninghamia lanceolata, as reference, with 5 replicate plots per species and site. Each plot was 25.8 × 25.8 m in size and planted with 400 tree individuals arranged on a rectangular 20 × 20 grid with 1.29 m spacing between rows and columns. To minimize edge effects, plots were established adjacent to each other, with trees thus forming a continuous cover across the entire site. Site A was planted in 2009, site B in 2010.
Plots of one species pool per site (pools A1 and B1 at sites A and B, respectively, Extended Table 2) were additionally replicated in plots that were four times larger and thus contained 1600 trees. These large plots were subdivided into four quadrants in which a factorial understorey shrub-diversity treatment was established. These four subplots either had no shrub understorey (0 species), or shrubs planted in all the centers between 4 adjacent trees, at a diversity of 2, 4 or 8 shrub species (Fig. 1a).
The design we use here consisted of 140 small plots (1 mu) and 64 large plots (4 mu). Out of this total of 396 1-mu sized (sub)plots, nine had to be excluded because these were not established due to a lack of sapling material or high initial mortality. All plots were weeded annually to remove emerging herbs and woody species that were not part of the planting design.
Tree measurements
We assessed stand-level and population-level tree growth by measuring the height of trees and maximum and minimum stem diameter at 5 cm above ground to calculate basal area. We focused on the central 4 × 4 =16 trees of each 1-mu (sub)plot to avoid edge effects. These measurements were repeated annually in September/October from 2013 to 2016. We aggregated these tree-level data at the species (i.e. population) and stand level.
We further calculated a cylindrical tree volume as the product of basal area and height. The true volume was then obtained by multiplying this proxy with a form factor determined by a complete harvest of 154 trees in natural forest near the experimental sites. The total volume of each harvested trees was calculated as ratio of total aboveground dry biomass and average wood density. Similarly, tree biomass was determined by multiplying the cylindrical volume of each experimental tree with a biomass conversion factor determined based on the harvested trees (Extended Data). Biomass was converted to carbon content32 by multiplying with 0.474 g C g−1.
Complementarity effect and selection effect
We used the additive partitioning method of Loreau & Hector3 to decompose net biodiversity effects (NEs) of productivity measures into complementarity (CEs) and selection effects (SEs), separately for each year and diversity level. CEs and SEs depend on relative yields of species, which we calculated using monoculture biomass as denominator. If a species failed to establish in monoculture (which was the case for Meliosma flexuosa, Castanopsis eyrei and Machilus grijsii), or had a mortality exceeding 80% (Quercus phillyreoides, Phoebe bournei), it was excluded from the set of target species in the corresponding mixtures33. Formally, CEs and SEs are related to (co)variances and therefore were square-root transformed with sign reconstruction prior to analysis, which improved the normality of residuals3.
Overyielding and underyielding
Overyielding describes the case where the productivity of a mixture exceeds the average productivity of monocultures of component trees26. Conversely, underyielding identifies a lower yield of the mixture relative to monocultures. Transgressive overyielding indicates that the productivity of a mixture exceeds the productivity of the monoculture of the most productive component species. Transgressive underyielding is defined similarly. We determined overyielding and underyielding of all mixtures relative to monocultures. Capitalizing on the nested nature of our design, we further determined the same metrics using the two mixtures with half the set of species as reference, instead of monocultures, i.e. we tested whether combining communities with two sets of species resulted in a community that produced more or less biomass than expected on the assumption of no interactions among the sets (overyielding) or that community productivity would be determined by the more productive set of species alone (transgressive overyielding).
Vertical crown complementarity
We quantified the interspecific complementarity in vertical crown extent of trees in 2016. The crown extent was determined as interval between the lowest side-branch and the top of a tree in monocultures. These data were averaged across all surviving trees of the 16 central individuals planted in a plot. We then calculated vertical crown complementarity in 2-species mixtures as proportional dissimilarity of the crown extents between the two species: where xA\B indicates the vertical extent (in meters) that is occupied by A but not by B (vice versa for xB\A), and xA⋃B indicates the extent occupied by at least one of the species. This index is equivalent to one minus the proportional similarity index proposed by Colwell and Futuyma34.
Statistical analysis
We used analysis of variance based on type-I sum of squares linear mixed-effects models to assess the effects of tree species richness (and additional design variables) on productivity35. All analyses were done in R 3.3.2 and ASReml-R36. The models included the fixed effects site, tree species richness (log2-transformed), year (continuous variable, centered over our observation period), the interaction log2(tree species richness) × year, and the interaction site × year. Random effects were species composition (with a separate variance component for each site), plot (with a separate variance component for each site), subplot, and the interactions of all these random terms with year. Model residuals were checked for normality and homogeneity of variances.
For the analyses of shrub diversity effects, the model contained the additional fixed effects shrub presence (a two-level factor: 0 vs. 2, 4 or 8 shrub species), plot size (a two-level factor: 1 vs. 4 mu), log2 of shrub species richness (for shrub-species richness >0), and the interactions of all these terms with log2(tree species richness) and with year. Random effects were species composition (with a separate variance component for each site), plot (with a separate variance component for each site), subplot, and the interactions of all these random terms with year (Extended Data Table 6). The interaction of year and site and the site-specific variance terms estimated for some random terms accounted for the fact that site B was established one year after site A and that trees at site B were therefore smaller.
Data availability statement
The data supporting the findings of this study will be deposited in Pangaea with the accession code https://doi.pangaea.de/xxxxxxxxxxxxxx.
Acknowledgements
We thank the farmers and Chen Lin for help in the field. This study was supported by the German Research Foundation (DFG FOR 891), the National Natural Science Foundation of China (NSFC No. 31270496 and No. 31300353), the Swiss National Science Foundation (SNSF No. 130720, 147092) and the European Union (EC 7th Framework Program No. 608422).
Author contributions
HB, KM and BS conceived the project with help from all co-authors; YH carried out the measurements; YH, YC, KM, PAN and BS led the data analysis and interpretation. All authors contributed to the writing of the manuscript.
Author information
The authors declare no competing financial interests. Correspondence and requests for materials should be addressed to Y.H. (yuanyuan.huang{at}ieu.uzh.ch), H.B. (helge.bruelheide{at}botanik.uni-halle.de), K.M. (kpma{at}ibcas.ac.cn), P.A.N. (pascal.niklaus{at}ieu.uzh.ch) or B.S. (bernhard.schmid{at}ieu.uzh.ch).
Footnotes
Emails of other authors:
Yuanyuan Huang <yuansjob{at}gmail.com>
Yuxin Chen <yuxin.chen{at}ieu.uzh.ch>
Nadia Castro-Izaguirre <nadia.cci{at}gmail.com>
Martin Baruffol <martin.baruffol{at}gmail.com>
Matteo Brezzi <Matteo.brezzi{at}gmail.com>
Anne Lang <anne.lang{at}gutaltenoythe.de>
Ying Li <yingli8441{at}hotmail.com>
Werner Härdtle <haerdtle{at}uni-lueneburg.de>
Goddert von Oheimb <Goddert_v_Oheimb{at}tu-dresden.de>
Xuefei Yang <xuefei{at}mail.kib.ac.cn>
Kequan Pei <peikequan{at}ibcas.ac.cn>
Sabine Both <s.both{at}abdn.ac.uk>
Xiaojuan Liu <liuxiaojuan06{at}ibcas.ac.cn>
Bo Yang <yangbomvp{at}aliyun.com>
David Eichenberg <david.eichenberg{at}idiv.de>
Thorsten Assmann <thorsten.assmann{at}leuphana.de>
Jürgen Bauhus <juergen.bauhus{at}waldbau.uni-freiburg.de>
Thorsten Behrens <thorsten.behrens{at}uni-tuebingen.de>
Francois Buscot <francois.buscot{at}ufz.de>
Xiao-Yong Chen <xychen{at}des.ecnu.edu.cn>
Douglas Chesters <dchesters{at}ioz.ac.cn>
Bing-Yang Ding <dingby2005{at}126.com>
Walter Durka <walter.durka{at}ufz.de>
Alexandra Erfmeier <aerfmeier{at}ecology.uni-kiel.de>
Jingyun Fang <jyfang{at}urban.pku.edu.cn>
Markus Fischer <markus.fischer{at}ips.unibe.ch>
Liang-Dong Guo <guold{at}im.ac.cn>
Dali Guo <guodl{at}igsnrr.ac.cn>
Jessica L. M. Gutknecht <jgut{at}umn.edu>
Jin-Sheng He <jshe{at}pku.edu.cn>
Chun-Ling He <hechunling68{at}126.com>
Andy Hector <andrew.hector{at}plants.ox.ac.uk>
Lydia Hönig <lydia.hoenig{at}botanik.uni-halle.de>
Ren-Yong Hu <wzryhu{at}163.com>
Alexandra-Maria Klein <alexandra.klein{at}nature.uni-freiburg.de>
Peter Kuehn <peter.kuehn{at}uni-tuebingen.de>
Yu Liang <coolrain{at}ibcas.ac.cn>
Stefan Michalski <stefan.michalski{at}ufz.de>
Michael Scherer-Lorenzen <michael.scherer{at}biologie.uni-freiburg.de>
Karsten Schmidt <karsten.schmidt{at}uni-tuebingen.de>
Thomas Scholten <thomas.scholten{at}uni-tuebingen.de>
Andreas Schuldt <andreas.schuldt{at}idiv.de>
Xuezheng Shi <xzshi{at}issas.ac.cn>
Man-Zhi Tan <mzhtan{at}issas.ac.cn>
Zhiyao Tang <zytang{at}urban.pku.edu.cn>
Stefan Trogisch <stefan.trogisch{at}botanik.uni-halle.de>
Zhengwen Wang <wangzw{at}iae.ac.cn>
Erik Welk <erik.welk{at}botanik.uni-halle.de>
Christian Wirth <cwirth{at}uni-leipzig.de>
Tesfaye Wubet <tesfaye.wubet{at}ufz.de>
Wenhua Xiang <xiangwh2005{at}163.com>
Jiye Yan <yanjiye{at}baafs.net.cn>
Mingjian Yu <fishmj{at}zju.edu.cn>
Xiao-Dong Yu <yuxd{at}ioz.ac.cn>
Jiayong Zhang <zhangjiayong{at}zjnu.cn>
Shouren Zhang <zsr{at}ibcas.ac.cn>
Naili Zhang <zhangnl{at}ibcas.ac.cn>
Hong-Zhang Zhou <zhouhz{at}ioz.ac.cn>
Chao-Dong Zhu <zhucd{at}ioz.ac.cn>
Li Zhu <julie{at}ibcas.ac.cn>