Tree species richness modulates water supply in the local tree neighbourhood: evidence from wood δ¹³C signatures in a large-scale forest experiment

Abstract

Biodiversity is considered to mitigate the adverse effects of changing precipitation patterns. However, our understanding of how tree diversity at the local neighbourhood scale modulates the water use and leaf physiology of individual trees remains unclear. We made use of a large-scale tree diversity experiment in subtropical China to study eight tree species along an experimentally manipulated gradient of local neighbourhood tree species richness. Twig wood carbon isotope composition (δ¹³C_wood) was used as an indicator for immediate leaf-level responses to water availability in relation to local neighbourhood conditions and a target tree's functional traits. Across species, a target tree's δ¹³C_wood signatures decreased progressively with increasing neighbourhood species richness, with effects being strongest at high neighbourhood shading intensity. Moreover, the δ¹³C_wood-shading relationship shifted from positive (thin-leaved species) or neutral (thick-leaved species) in conspecific to negative in heterospecific neighbourhoods, most likely owing to a lower interspecific competition for water and microclimate amelioration. This suggests that promoting tree species richness at the local neighbourhood scale may improve a tree's local water supply with potential effects for an optimized water-use efficiency of tree communities during drought. This assumption, however, requires validation by further studies that focus on mechanisms that regulate the water availability in mixtures.

1. Introduction

Trees in most forest biomes are living at their hydraulic limits [1]. Given that competition for water is likely to increase among trees in the face of climate change, promoting tree species richness is considered a promising strategy to improve the water availability of forest ecosystems. Although tree diversity has been shown to enhance the transpiration and decrease water-use efficiency (WUE) of individual trees and communities [2–4], results remain controversial [5–7].

Exploring species interactions at the local neighbourhood level is an important further step to understand how tree diversity may modulate the water availability and drought susceptibility of forest ecosystems, as previous studies have experimentally demonstrated that the response of tree communities to species mixing is largely the result of biodiversity-mediated tree–tree interactions at the local neighbourhood level [8,9]. For example, neighbourhood tree species richness (NSR) has been shown to mitigate adverse effects of limited soil water resources during drought on the growth of young subtropical trees, with effects being strongest for drought-sensitive species [10]. Moreover, a study on tropical tree seedlings using physiological variables, such as stomatal conductance, found that drought resistance was higher in heterospecific than in conspecific neighbourhoods [11]. However, the vast majority of research on leaf physiological responses to tree diversity has focused on tree diversity at the community level [4,6,7,12–14], whereas only few studies dealt with local neighbourhood conditions [11,15,16]. Importantly, neighbourhood diversity effects were addressed in these studies by neighbourhood treatments (e.g. monocultures versus two- or three-species mixtures) rather than systematically varying the number of heterospecific neighbours. According to our knowledge, research on leaf physiological responses to species-rich neighbourhoods is still missing. Thus, we focus on leaf physiological responses of young individual trees to varying levels of local neighbourhood conditions by experimentally manipulating a long gradient of NSR with a maximum of seven heterospecific neighbours.

Stable carbon isotope signatures in plant material (δ¹³C) are an integrative marker for leaf gas exchange and δ¹³C in tree rings are useful proxies of long-term changes in intrinsic WUE [17]. Wood δ¹³C data have been often applied to assess the effects of tree diversity on stomatal conductance [4,6,7,12,15], elucidating leaf-level processes behind growth responses. Leaf physiological processes such as stomatal conductance (g_s) and carbon assimilation rates (A) are imprinted into the δ¹³C signal in plant material [18]. As a consequence, both water and light availability can affect plant tissue δ¹³C signatures. Firstly, plants respond to soil water shortage by closing their stomata, thereby decreasing g_s (as a protection against water loss). Under these conditions, the intercellular partial pressure of carbon dioxide (c_i) is lowered, and the enzyme that is involved in the first major step of carbon fixation, Rubisco, shows an enhanced affinity to ¹³C compared to carbon fixation under high c_i. As a result, the ratio of ¹³C/¹²C in the assimilates produced during the period of water deficit is enhanced [19]. Secondly, shading decreases a plant's δ¹³C signature [20] owing to the effects on higher air humidity and lower light availability. On the one hand, air humidity, one of the main drivers of stomatal responses [21], is enhanced within tree crowns because of shading [22], thereby increasing g_s and c_i (decreasing plant δ¹³C). On the other hand, shading decreases light availability and A [23], which again increases c_i and decreases plant δ¹³C.

In this study, we focus on plant δ¹³C signals as an indicator for leaf physiological responses to water availability. Variation in plant δ¹³C signals is associated with two opposing pathways: changes in g_s owing to reduced soil water availability [19], and changes in A and g_s owing to shading effects [20]. To account for shading effects, we used an index of neighbourhood shading based on the average tree height of all closest neighbours relative to the height of the target tree (neighbourhood height index, NHI). For high levels of NHI (i.e. neighbouring trees are on average taller than the target tree) we assume that decreasing plant δ¹³C values are associated with shading effects, whereas for low levels of NHI (i.e. neighbouring trees are on average smaller than the target tree) we expect higher plant δ¹³C values because of conditions with less shading. The relationship between plant δ¹³C and NHI can either be negative, owing to increasing shading effects, such as reduced assimilation rates or improved microclimatic conditions, or positive, owing to other effects than shading that regulate c_i and plant δ¹³C, such as reduced water availability or drought stress [18,19]. Consequently, a positive plant δ¹³C–NHI relationship is most likely the result of increasing competition for water among local neighbours. This allows us to disentangle water availability and shading effects on δ¹³C, in order to identify stomatal responses towards water supply.

In subtropical forests, diversity in plant functional traits was shown to explain biodiversity effects better than species richness per se [24,25]. Stomatal responses to water availability are interlinked with functional traits associated with leaf morphology (such as leaf thickness, LT, specific leaf area, SLA and leaf habit, LH) and wood properties (such as wood density, WD [26]). Moreover, SLA and WD strongly regulate a trees' competition response, both in terms of its competitive effect on neighbours and its ability to tolerate competition [27]. Trees with high SLA and low WD often show high growth rates and a low competition tolerance [27]. Such a trait combination can be associated with the ‘fast-slow’ plant economics spectrum [28] where species capable of rapid water transport have low tissue density and high rates of resource acquisition (acquisitive versus conservative species). We thus expect differences in water acquisition and use to affect the depletion of soil water resources and the stomatal responses to water availability in species with acquisitive versus conservative traits. In this context, the present study examines stomatal responses to water availability (as indicated by plant δ¹³C signals) in relation to changing neighbourhood conditions (NSR and NHI) and (potential) interaction effects of species-specific functional traits associated with resource acquisition, water uptake and use (such as WD, SLA, LT and LH). We used four evergreen and four deciduous subtropical tree species planted in a large-scale biodiversity-ecosystem functioning experiment (BEF-China [29]) to elucidate how NSR and a target tree's functional traits affect its twig wood δ¹³C signal. We hypothesize (i) that NSR would decrease δ¹³C signals of individual trees because of reduced competition for water (e.g. because of different water-use strategies of component species) or owing to facilitative species interactions that improve the local water supply and microclimate. We furthermore hypothesize that (ii) δ¹³C signals of individual trees decrease with increasing shading by neighbours (NHI), and (iii) that functional traits of the target tree mediate the δ¹³C–NSR relationship, with acquisitive species (i.e. those associated with high SLA and low LT and WD) benefitting most from growing in species-rich neighbourhoods.

2. Material and methods

(a). Study site and experimental design

The study was conducted in one of the two experimental sites (site A) which belongs to the Biodiversity-Ecosystem Functioning Experiment China Platform (BEF-China, www.bef-china.com) at Xingangshan, Dexing, Jiangxi (29° 08′–29° 11′ N, 117° 90′–117°93′ E; [29]). The climate of the study area is characterized by warm and humid summers and dry winters with occasional frost events (subtropical summer monsoon) with a mean annual precipitation of 1821 mm and a mean annual temperature of 16.7°C (measured 1971–2000 in the nearest city Wuyuan County, [30]). The annual water balance (characterized by the standardized precipitation-evapotranspiration index; SPEI) in the year preceding twig sampling (SPEI-12_Sept) was 1.33, indicating a year with ample water supply on average [31]. However, intermonthly variation of water balances was high, ranging from (severely) dry periods during winter to (severely) wet periods in summer (electronic supplementary material, figure S1). For site A, altitude ranges between 105 and 275 m with a slope between 0° and 45°. The predominant vegetation type in the region is a species-rich subtropical broadleaf forest with dominating evergreen tree species [32], containing species with a wide range in growth rates and shade tolerance [33].

The BEF-China experiment comprises a pool of 40 native broadleaved tree species, both evergreen and deciduous. Total tree number in the experiment was 219 200 on a total area of 38.4 ha, with 400 tree individuals planted per plot (for further information on the design and establishment of the BEF-China experiment, see [29]). Site A of the BEF-China experiment (electronic supplementary material, figure S2) was planted in 2009 with 1- or 2-year old saplings (described in detail in [30]). Thus, all trees were 7–8 years old during our sampling campaign in September 2015. Trees were planted in a regular quadratic grid with an equal distance of 1.29 m (see the electronic supplementary material, figure S3). In total, we selected 326 target trees that were surrounded by a maximum of eight direct neighbours. In 2015, the mean number of living neighbours amounted to 6.5, and 76% of the target trees were surrounded by 6–8 direct neighbours (see the electronic supplementary material, table S1).

(b). Sampling of plant material

We sampled a total number of 43 plots covering eight a long gradient of tree species richness (monocultures, two-, four-, eight- and 16-species mixtures). All study species were represented in all levels of plot tree species richness. To avoid edge effects, sampling was restricted to the core zones of the plots, which included the central 36 trees (monocultures and two-species mixtures) or 144 trees (four-, eight-, 16-species mixtures [32]).

The 326 individuals belong to eight broadleaved tree species including four evergreen species (Lithocarpus glaber, Schima superba, Cyclobalanopsis glauca, Castanopsis sclerophylla) and four deciduous species (Liquidambar formosana, Choerospondias axillaris, Sapindus saponaria, Nyssa sinensis; nomenclature follows ‘The Flora of China’; http://flora.huh.harvard.edu/china; electronic supplementary material, table S2). We took twig samples within four weeks in September 2015. To account for the effects of sun exposure on leaf traits [34] and leaf gas exchange [35], samples were taken from branches in the top crown with an expandable pruner. Twigs were checked for age by looking at internodes and number of rings (microscopically). We discarded any summer shoots as well as twigs that contained more than 1 year ring (see the electronic supplementary material, figure S4) to restrict our analyses to twig wood that had been formed largely in the growing season of 2015.

(c). Isotope analysis

All samples were oven-dried for 24 h at 105°C. The twig samples were separated into bark and wood, and the latter stored for isotope analysis. The samples were ground in a mill (Retsch GmbH, mixer mill MM 400, 1–2 × 2.5 min at 30–1 s) and analysed in an isotope ratio mass spectrometer (Isoprime Ltd., Cheadle Hulme, UK) coupled to an elemental analyser (Vario el cube, Elementar, Hanau, Germany). Carbon isotope values were expressed in δ notation relative to Vienna Pee Dee Belemnite (VPDB). The precision of δ¹³C per run, as determined by repeated measurements, was better than 0.10‰.

(d). Tree productivity

We used the trunk volume increment between 2014 and 2015 as a measure for tree productivity in the year prior to twig sampling. Wood volume was calculated based on annual tree height and ground diameter measurements according to [9]. Negative volume increment values can result from measurement errors or mechanical tree damage such as damage due to falling large-sized branches [36]. We, therefore, excluded trees with negative trunk volume increment (9.8% of our samples) in the subsequent analyses (resulting in n = 294 trees).

Tree productivity is closely related to assimilation rates, and thus facilitates the interpretation of NHI effects on δ¹³C_wood. Potential causes for ¹³C enrichment are increases in assimilation rates (A) and decreases in stomatal conductance (g_s). For example, concomitant increases of productivity and δ¹³C_wood with increasing shading (increasing values of NHI) indicate that an increase in A is the most likely explanation for the increase in δ¹³C_wood. By contrast, a positive δ¹³C_wood–NHI relationship with a concomitant reduction in productivity with increasing NHI indicate that a decrease in g_s is the most likely explanation for the increase in δ¹³C_wood.

(e). Local neighbourhood model

We used linear mixed-effects models to evaluate the role of neighbourhood conditions and functional traits (related to wood hydraulic properties as well as LH, morphology and physiology, see the electronic supplementary material, table S3) of the target tree on δ¹³C_wood. Data on species-specific functional traits were obtained from [25]. Local neighbourhood conditions were characterized by NSR, NHI, neighbour density (ND) and conspecific neighbour density (CND). NSR was calculated as the total number of closest heterospecific neighbours (i.e. trees with different species identity compared with the target tree) with NSR equal to 0 indicating a target tree growing in conspecific neighbourhoods and NSR equal to or greater than 1 indicating a target tree growing in heterospecific neighbourhoods. Note that the maximum of eight heterospecific neighbours was not realized for the target trees in our study. NHI as an index of neighbour shading was computed for each target tree as the difference in tree height between a target tree and the mean height of all closest neighbours. Thus, negative NHI-values indicate a target tree surrounded by smaller neighbours on average (i.e. low levels of neighbour shading), while positive values indicate a target tree surrounded by taller neighbouring trees on average (i.e. high levels of neighbour shading). Given that individuals surrounded by taller neighbours might be subjected to lower light levels (indicated by lower δ¹³C), and/or to enhanced competition for soil water (indicated by higher δ¹³C), NHI can be used to disentangle δ¹³C responses to low light conditions (negative δ¹³C–NHI relationship) from δ¹³C responses to water scarcity (positive δ¹³C–NHI relationship). ND was computed as the total number of living neighbours to account for neighbour mortality. CND was calculated as the total number of direct conspecific neighbours to separate the effects of CND and NSR on target tree δ¹³C responses (see [9]). We used the values of NSR, NHI, ND and CND obtained in the study year 2015, because they reflect the neighbourhood conditions that are imprinted into the carbon isotope signal and in the twig wood (containing 1 year ring) that has been formed in the last 12 months.

First, we fitted several candidate models that included all predictors, three-way and two-way interactions between NSR, NHI and a specific functional trait (using SLA, LT, LH or WD). The model with the lowest Akaike information criterion (corrected for small sample sizes) and the highest Akaike weight (i.e. the relative likelihood of each model being the best-fitting model, given the complete set of candidate models [37]; electronic supplementary material, table S3) was used in subsequent analyses. The candidate models had the basic form:

$$\begin{array}{c}{\delta }^{13}{\mathrm{C}}_{\mathrm{wood}}=\mathrm{ND}\hspace{0.17em}+\mathrm{CND}\hspace{0.17em}+{(\mathrm{NSR}\hspace{0.17em}+\mathrm{NHI}\hspace{0.17em}+\mathrm{trait})}^3+(1|\mathrm{species})\\ +(1|\mathrm{NSC})+(1|\mathrm{plot}).\end{array}$$

The syntax (1|X) denotes random intercepts for a target tree's species identity (species), tree species composition of the local neighbours (NSC) and plot identity (plot, to account for variation in abiotic growing conditions within the study site).

Second, we identified significant predictors (i.e. optimal fixed structure) using likelihood ratio tests based on maximum likelihood estimation. Finally, we refitted the best-fitting model with the restricted maximum likelihood method [38].

The importance of NSR for δ¹³C_wood was quantified as the absolute difference in δ¹³C_wood-values between heterospecific (NSR ≥ 1) and conspecific (NSR = 0) neighbourhoods by varying the number of heterospecific neighbours (with 1–7 species) and NHI (using the 20%, 50% and 80% quantiles) based on predictions from our best-fitting mixed-effects model (net effects). Similarly, we predicted changes in δ¹³C_wood-values along a gradient of NHI (using the 10–90% quantile) for low (using the 20% quantile) and high (using the 80% quantile) values of NHI and different levels of NSR. All predictors were standardized (mean = 0, s.d. = 1) prior to analysis. Model residuals were checked for normality, independence and homoscedasticity, and the correlation between covariates did not indicate serious collinearity (variance inflation factors less than 5). Statistical analyses were performed with R-3.5.0 [39] and the packages MuMIn [40], lme4 [41] and ggplot2 [42].

3. Results

The best-fitting model included the effects of NSR, NHI, LT and the interactions between NSR and NHI, as well as between NHI and LT (table 1). The model explained 17% (fixed effects) and 78% (fixed and random effects) of the total variance in δ¹³C_wood signatures, respectively. Although the three-way interaction between NSR, NHI and LT was marginally significant (χ: 3.11, p = 0.078), we did not consider this term in the best-fitting model. Total neighbour density (ND, χ: 0.86, p = 0.353) and conspecific neighbour density (CND; χ: 0.11, p = 0.740) did not significantly affect δ¹³C_wood signatures.

Table 1.

Results of the best-fitting linear mixed-effects model for the effects of neighbourhood tree species richness (NSR), neighbourhood height index (NHI) and leaf thickness (LT) on wood carbon isotope composition (δ¹³C_wood) of individual trees. (The target tree's species identity (species), neighbourhood tree species composition (NSC) and plot identity (plot) were used as crossed random effects. The variance explained by the fixed effects alone (marginal R²) and by both the fixed and random effects (conditional R²) was calculated according to [43]. The percentage of variance explained by a specific random effect is given by partial R²-values [44]. s.e., standard error; s.d., standard deviation; d.f., degrees of freedom. Regression coefficients are standardized.)

fixed effects	estimate	s.e.	d.f.	t-value	p-value
intercept	−29.619	0.404	6.3	−73.24	<0.001
NSR	−0.209	0.078	67.2	−2.67	0.009
NHI	−0.103	0.072	282.3	−1.43	0.155
LT	0.612	0.414	6.0	1.48	0.189
NSR × NHI	−0.212	0.054	270.8	−3.90	<0.001
NHI × LT	−0.128	0.065	259.7	−1.97	0.049

random effects	s.d.	partial R2
plot	0.317	0.04
NSC	0.304	0.04
species	1.116	0.53
residual	0.717	0.22
marginal R2	0.17
conditional R2	0.78
n (trees)	294

Overall, δ¹³C_wood signatures significantly (t: −2.67, p = 0.009) decreased with increasing NSR, although absolute changes were relatively small (figure 1a). Across species, δ¹³C_wood signatures of trees growing in species-rich neighbourhoods (NSR = 7) that experienced an average level of NHI were 0.85‰ (0.27‰ and 1.27‰ at low and high levels of NHI, respectively) lower than those growing in conspecific neighbourhoods (NSR = 0; figure 1b). The negative δ¹³C_wood–NSR relationship, however, depended on NHI, with mean net effects (all significantly different from zero) becoming stronger with increasing NHI (figure 1b,c). For example, the mean net effect at high NHI was almost four-times stronger compared with low NHI (figure 1c). Moreover, the effect of NHI on δ¹³C_wood depended on LT of the target tree. On average, target trees with thicker leaves showed higher δ¹³C_wood-values than thin-leaved species (figure 2). Interestingly, we observed two contrasting response patterns: (i) for conspecific neighbourhoods, δ¹³C_wood increased with increasing NHI for thin-leaved-species (figure 2a), while δ¹³C_wood was hardly affected by changes in NHI (with a trend towards a positive δ¹³C_wood–NHI relationship) for thick-leaved-species (figure 2b); and (ii) for heterospecific neighbourhoods, δ¹³C_wood declined with increasing NHI, and this effect became stronger with NSR and for species associated with thick leaves compared with thin-leaved species (figure 2). As expected, individual tree productivity (log-trunk volume increment) significantly (t: −12.72, p < 0.001) decreased with increasing NHI (electronic supplementary material, figure S5).

Figure 1.

(a) Relationship between wood carbon isotope composition (δ¹³C_wood) and neighbourhood tree species richness (NSR) across neighbourhood height index (NHI; fixed at its median) and leaf thickness (LT; fixed at its median). The solid dark green line corresponds to a linear mixed-effects model fit (table 1) with shaded area indicating the 95% confidence interval. Points represent the observed values and are slightly jittered to facilitate visibility. (b) Changes in net biodiversity effects (absolute difference in δ¹³C_wood between heterospecific and conspecific neighbourhoods) with varying levels of NHI. Lines represent the predicted response at low, average and high levels of NHI. LT as fixed at its median. (c) Mean net biodiversity effects across heterospecific neighbourhoods (NSR 1–7) for varying levels of NHI. Error bars denote the 95% confidence intervals. Note that higher values of NHI are associated with a higher shading intensity by neighbours. (Online version in colour.)

Figure 2.

Effects of neighbourhood height index (NHI) on wood carbon isotope composition (δ¹³C_wood) for varying levels of neighbourhood tree species richness (NSR) and leaf thickness. (a) Thin-leaved species and (b) thick-leaved species. NSR ranges between conspecific (NSR = 0), moderately species-rich (NSR = 3) and species-rich (NSR = 5) neighbourhoods. Regression lines correspond to linear mixed-effects model fits (table 1). Note that higher values of NHI are associated with a higher shading intensity by neighbours. (Online version in colour.)

4. Discussion

Our results show that increasing neighbourhood species richness (NSR) leads to lower δ¹³C signals of individual trees, confirming our first hypothesis. This response was consistent across species and various levels of neighbourhood shading intensity (NHI), but effects were strongest for trees experiencing a high level of NHI (i.e. a high level of neighbour shading). We are aware that our effect sizes (i.e. the difference in difference in δ¹³C_wood-values between heterospecific and conspecific neighbourhoods) were small. This might result from the relatively wet annual water balance (SPEI-12: 1.33) in the year of sampling. However, within-year variation of monthly water balances was high, ranging from severely dry (−1.26) to severely wet (1.81) months, thus indicating that trees have experienced periods of (strong) water deficits in our study. Although this is a typical feature for a subtropical summer monsoon climate, within-year variation largely increased within the last decade (electronic supplementary material, figure S1). It is, therefore, conceivable that the positive effects of NSR on stomatal responses of individual trees will become stronger during longer and more intense drought events compared to those investigated in this study. This is further supported by the fact that our effect sizes at the local neighbourhood level are (at least) similar to those that were reported for stand- and individual tree-level responses to changing community tree diversity between wet and dry years [5,6]. As expected, species identity accounted for a substantial variation in δ¹³C_wood signatures, whereas the importance of small-scale variation in site conditions (plot effects) and differences in neighbourhood composition was rather low. This indicates that stomatal responses of trees growing in heterospecific neighbourhoods are mainly driven by species identity rather than neighbourhood conditions in our study.

Two main mechanisms may explain shifts in δ¹³C_wood patterns in relation to NSR and NHI: reduced competition for water among heterospecific neighbours and/or facilitative species interactions in heterospecific neighbourhoods [45,46]. The competitive reduction can occur owing to partitioning of water resources via differences in rooting depth or in timing of water uptake [47]. Facilitative mechanisms include processes such as microclimate amelioration [48,49] or hydraulic lift [50,51]. All these mechanisms have the potential to attenuate competition for water, and thereby improve the water supply of individual trees (i.e. affecting plant δ¹³C signatures via stomatal conductance). Several studies reported decreasing δ¹³C signatures of trees growing in mixtures and often attributed these changes to improved water supply in mixed-species communities [4,6,12,15,52]. We found that the direction and magnitude of the δ¹³C_wood–NHI relationship depended on leaf traits (i.e. LT) of a target tree, which partially supports our second hypothesis. Particularly for thin-leaved species growing in conspecific neighbourhoods, δ¹³C_wood increased with increasing NHI. This finding is remarkable, as a reduction in light availability because of neighbour shading generally results in decreasing assimilation rates [23], and hence ¹³C depletion. In addition, amelioration of microclimate (e.g. higher relative air humidity and reduced vapour pressure deficit at the leaf surface and reduced evaporative demand of whole trees; [49]) owing to shading can also lead to increased stomatal conductance and ¹³C depletion. Consequently, the observed positive δ¹³C_wood–NHI relationship for thin-leaved species in conspecific neighbourhoods should not result from lower assimilation rates or higher stomatal conductance associated with shading. Instead, changes in stomatal responses are most likely the result of more intense competition for water in conspecific neighbourhoods, which in turn results in lower stomatal conductance and increasing δ¹³C_wood signatures, respectively. The same might hold for species associated with thick leaves, although this effect was less evident (note that δ¹³C_wood–NHI relationship was neutral or tended to increase, but did not become negative). This assumption is further supported by the significant decline in productivity with NHI (electronic supplementary material, figure S5), suggesting that an increase in assimilation rates (as a theoretical cause for ¹³C enrichment) can mainly be neglected as an explanation for the observed increase in δ¹³C_wood, whereas a decrease in stomatal conductance is most plausible. We, therefore, conclude that δ¹³C_wood signatures of trees growing in conspecific neighbourhoods, particularly those of thin-leaved species, were predominantly affected by stomatal closure (overriding the general shade effect on c_i) owing to decreasing local water supply as a result of higher intraspecific competition for water. This also coincides with results from our study species, where the positive biodiversity–growth relationship of acquisitive species (those with e.g. thin leaves) was mainly driven by competitive reduction [53].

By contrast, the observed δ¹³C_wood–NHI relationship for both thin- and thick-leaved species growing in heterospecific neighbourhoods was negative and became stronger with increasing NSR. This response can either be attributed to decreasing assimilation rates under reduced light availability (which is in line with the finding that both δ¹³C_wood and tree productivity declined with increasing NHI) and/or improved soil water supply or microclimate amelioration in heterospecific neighbourhoods. Although we were not able to disentangle the processes underlying shading effects in this study, we assume that increasing NSR and NHI jointly improve—at least to a certain extend—the microclimate in the canopy of species-rich neighbourhoods, and thereby reduces δ¹³C_wood signals, particularly of thick-leaved species. Importantly, δ¹³C_wood signatures decreased with both increasing NSR and NHI. Consequently, δ¹³C_wood was lower in species-rich compared with moderately species-rich neighbourhoods at a given level of shading. These diversity-mediated differences indicate that factors other than shading should additionally regulate δ¹³C_wood signals within our study. Moreover, facilitative neighbourhood interactions such as microclimate amelioration has been identified as a key mechanism that allows conservative species to be more productive when growing in species-rich neighbourhoods [53]. This would explain the stronger decline in δ¹³C_wood of thick-leaved species at high levels of shading compared with thin-leaved species. In summary, the observed diversity-mediated (further) depletion in δ¹³C_wood signals are not solely attributable to shading effects, but most likely also to diversity-mediated increases in stomatal conductance. This in turn suggests an improved water supply and/or improved microclimatic conditions in species-rich neighbourhoods.

In this study, we have provided experimental support that δ¹³C_wood signals decline with increasing NSR. The observed trait-mediated responses to NHI along the NSR gradient might result from species-specific water-use strategies. The investigated thin-leaved species are primarily deciduous and are characterized by high nitrogen (N) inputs in leaf structure and high hydraulic conductivity [54]. These traits suggest high water consumption rates (‘water-demanding species' [55]), which supports our assumption of higher intra- than interspecific competition for soil water (leading to soil water depletion). By contrast, evergreen species were primarily associated with high LT in our study. These species are characterized by traits such as N-conservative leaves with low SLA, low turnover rates, low hydraulic conductivity and high WD [54–57]. These traits are associated with a rather conservative use of water resources (‘water-conservative species’ [55]), and may explain why those species eventually benefit more from microclimate amelioration in species-rich neighbourhoods. However, it should be noted that we were not able to directly quantify processes underlying changes in δ¹³C_wood–NSR relationships. Future studies should therefore focus on analysing physiological responses to neighbourhood diversity and changing climatic conditions along with direct measurements of possible underlying processes. This would advance our understanding of biodiversity–ecosystem functioning relationships in forests in the prospect of substantial changes in precipitation patterns and water availability.

Data accessibility

All data are available on the BEF-China project database: link https://data.botanik.uni-halle.de/bef-china/datasets/630 [58].

Authors' contributions

K.J., G.v.O., H.B., W.H. and A.F. developed the project idea. K.J., W.H. and G.v.O. developed the sampling design. K.J. conducted the twig sampling and laboratory analyses. K.J. and A.F. analysed the data. K.J. wrote the first draft of the manuscript and all other authors contributed substantially to revisions.

Competing interests

We declare we have no competing interests.

Funding

We received no funding for this study.