Figure 3: The contribution of SR, plot ID, season and year to total covariance between the 116 pairs of EFs, separated for EF pairs showing positive relationships (synergies, 78 pairs) and negative relationships (trade-offs, 38 EF pairs) according to their mean correlation. The violin plots show for each driver the mean (solid line), the standard error, and the distribution of contributions to covariance of EF pairs. Positive contributions indicate that the driver causes positive covariances between pairs of EFs, synergistically driving the two individual EFs. Negative contributions indicate that the driver causes negative covariances between pairs of EFs, driving the two individual EFs antagonistically. Results are derived by partitioning overall covariances into contributions of the different drivers; see method section for explanation. In this graph, only effects are shown, which on average explain >5% of covariance. In Supporting information F), the same graph with all variables (including FGR, and the presence/ absence of grasses, legumes and herbs) is shown.

DISCUSSION

We investigated the variation in the correlations between different EFs and the drivers of these relationships. We found that correlations were variable, and correlations between two particular EFs could range from weak to strong or from negative to positive among the repeated measurements. Overall, EF pairs generally showed an increasing variation in their correlations with a higher number of times the EF-pair was measured. The correlations among pairwise EFs were differently affected by the identity of time points (years and seasons). That means that some EF-pairs showed more stable correlations throughout time, whereas other EF-pairs were more affected by differences in years and seasons and therefore showed a higher temporal variation. Species richness and plot identity (including the presence of legumes, grasses, herbs) explained the largest fraction of covariance among EFs, while the effects of time (year, season, and their interaction) explained little covariance. We found that most of the covariance for synergies was explained by species richness (~26.5%), whereas for trade-offs, most covariance was explained by plot identity (-29.5%). Time explained 13.4% of covariance for trade-offs but little for synergies (3.3%). Correlations among EFs and the drivers of these correlations varied over time. These results indicate the importance of repeated measurements of ecosystem functions (EFs) over time to avoid spurious conclusions, and suggests that land management practices that promote biodiversity and reduce negative identity effects can enhance multifunctionality in grasslands.
We found that even under the controlled conditions of our experiment, correlations among EFs were variable. High temporal variation of individual EFs had been documented before (Carpenter, Mooney et al. 2009, Cardinale, Duffy et al. 2012, Gaglio, Aschonitis et al. 2020, Qiu, Carpenter et al. 2020, van der Plas, Schröder-Georgi et al. 2020). However, until now, inconsistent correlations between EFs or classes of EFs have only been found when different studies were compared (Lee and Lautenbach 2016). Although all functions were measured with a consistent methodology at a single field site. Trade-offs were as variable as synergies (Supporting information E, Fig. S5), and relationships for many pairs of functions could range from synergy to trade-off when correlations were calculated for different time points, which confirms the previous study of Lee and Lautenbach (2016). Lee and Lautenbach (2016) found that the agreement on the type of relationship for a particular pair of EFs, i.e. synergy, trade-off, or no-effect relationship, decreased the more often the relationships were measured. Similarly, we found that the variability of EF relationships increased with the number of measurements (Supplementary information E, Fig. S6), which indicates, that single measurements can be misleading when EF relationships are identified. Furthermore we showed, that not the identity of time points (years and seasons), but the identity of EF pairs were associated with a high variation in EF relationships (Supporting information, Fig S7). That means, that it depends on the particular EF pair, whether their correlation was highly variable because of differences between years or seasons. One explanation could be that ecosystem processes vary caused by a change or adjustment of biotic assemblages as a response to their environmental conditions (Turner and Chapin 2005), leading to changes in EF relationships or multifunctionality with changing environmental conditions (Zirbel, Grman et al. 2019). In our study, the variation in relationships between EFs originated from the temporal variation in EF drivers (possible reasons could be inter-annual variation in rainfall, temperature or other cyclic patterns such as boom and bust cycles of herbivory), while in Lee and Lautenbach (2016), the variation in the relationships among classes of EFs was introduced by different studies, and therefore additional site-dependent contexts.
Regarding the identified EF relationships (mean correlations among all the different EFs), we found both, synergies and trade-offs that can be explained by biological processes and therefore confirm other studies investigating the individual EFs (Jarrell and Beverly 1981, Allan, Weisser et al. 2013). For example, EFs of the classes plant nutrients and plant productivity showed often a trade-off, indicating a dilution effect. i.e. when plant growth improved, plant nutrient concentrations decreased in the plant tissue (Jarrell and Beverly 1981). However, the carbon concentration of plants (PlantC) showed mainly synergies with EFs of the class plant productivity. One reason could be that a high biomass reflects a high nutrient-efficiency and thus comparatively low nutrient concentrations and correspondingly high C concentrations (Allan, Weisser et al. 2013). Furthermore, organic carbon in the soil (SoilDOC, SoilCorg) was positively related to plant productivity (Fig. 2). This is consistent with studies, showing that a high biomass production leads to an accumulation of dead plant material in the soil (Post and Kwon 2000) or root exodation of plants (Raich and Tufekciogul 2000). As the Jena experiment was established on depleted arable soil, a higher carbon concentration in the soil occured faster with higher biomass production, but in the end the carbon concentration might be the same on all plots due to accelerated litter decomposition (Weisser, Roscher et al. 2017). EFs of the class Invasion resistance showed synergies with EFs of the class plant productivity. This is consistent with former studies, showing that a high biomass of the native species suppressed invasive species (Yannelli, MacLaren et al. 2020, Rojas-Botero, Kollmann et al. 2022), often due to a more complete use of available resources (Hector, Dobson et al. 2001, Roscher, Beßler et al. 2009). Summarizing these examples, the relationships identified here for the classes of EFs are consistent with the underlying biological processes.
We found that synergies and trade-offs have different drivers. Species richness is a known driver of many EFs (Gamfeldt and Roger 2017, Weisser, Roscher et al. 2017, Craven, Eisenhauer et al. 2018) and ecosystem services (van der Plas 2019). When two EFs improve with higher SR, a positive covariance is introduced, strengthening their relationship. The fact that we showed that SR affected the majority of investigated EFs positively explains that the large majority of SR effects on the covariance between EFs were positive and that EFs in synergies were stronger effected by SR than EFs in trade-offs (Fig. 2). However, the relationship between two EFs weakens when SR has contrasting effects on the two EFs, as indicated by a few pairs of EFs for which we showed SR to cause negative covariance. In our study, plot identity represents all other differences among plant communities within a diversity level, such as FGR, the presence of different functional groups (Supporting information F, Fig S8), species identity, and the presence of other groups of organisms (e.g. microbes, insects) associated with particular plant communities. These identity effects were mostly positive for EFs in synergies and mostly negative for EFs in trade-offs. This can be explained by selection effects, which have been documented repeatedly by comparing the performance, such as biomass production, of plant communities (Marquard, Weigelt et al. 2009). A high performance of a plant community may be associated with a high abundance of certain species (Loreau and Hector 2001) and therefore with many simultaneously occurring EFs (many synergies). A low performance is associated with a negative selection effect (Loreau and Hector 2001) and could be related to the occurrence of just a few EFs, as they are restricted by trade-offs. Outside an experimental setup, the equivalent of SR and plotID would be the different communities associated with landscape patches. As a consequence of different biotic communities, different levels of individual EFs would occur in these patches and different correlations among EFs could be identified. While the effect of different aspects of SR and plotID (FGR, presence/absence of functional groups or individual species) on individual EFs is frequently investigated, further research is needed to identify how different SR/ plotID impacts the relationships between EF.
Time (year, season, and their interaction) explained some, albeit little, covariance among EFs and affected synergies and trade-offs differently. Time can become a driver of EF relationships when environmental conditions vary over time, e.g., temperature and extreme events, affecting the biological activity of organisms. Trade-offs were more affected by temporal effects than synergies. Season was among the main drivers of trade-offs, reflecting the pronounced change in abiotic conditions among seasons in the temperate zone. As an example, soil microbial activity strongly responds to climatic conditions, affecting carbon and nutrient cycling (Frey, Lee et al. 2013). Further, competition strongly affects trade-offs, which can be changed by altering the environmental conditions, such as the availability of water or light that fluctuate with time. Consequently, temporal variability in these drivers can induce variability in EF correlations underlying the importance of repeated measurements to identify the true relationships between EF.
When investigating the drivers of EF correlations, we found that some covariance among EFs could not be explained by any of the drivers tested in our study. We interpret this unexplained covariance as EF pairs affected by the ecological-constraints-mechanism. Plants have access to a limited pool of resources they can invest in, e.g. in growth or defence against natural enemies, resultingin a growth-defense trade-off (Karasov, Chae et al. 2017). Because providing unlimited resources within a local patch is impossible, ecological trade-offs resulting from resource limitation are inevitable. Further, the simultaneous provision of EFs can be limited by competition. For example, in our study, improved plant productivity was associated with higher invasion resistance (considered good), likely due to intensified competition for space and light between the resident plant community and potential invading plant individuals in our plots. Furthermore, the higher the root biomass was, the lower were the soil nutrient concentrations, implying competition among plant species for available nutrients. Resource limitations and competition may limit biological activities, leading to trade-offs between EFs. These trade-offs can be weakened when competition in diverse communities is reduced by complementarity between species (Weisser, Roscher et al. 2017). Understanding how ecological constraints affect relationships between EFs is an important topic for further investigation.
Our results have implications for land management aiming at promoting multifunctionality. Relationships among EFs affect multifunctionality since they can either promote (synergies) or limit (trade-offs) multifunctionality. Analysing the drivers of relationships between EFs, we showed that SR can promote synergies among EFs, resulting in increased multifunctionality. Consequently, promoting diversity is a mean to foster multifunctionality, confirming previous empirical biodiversity multifunctionality relationships (Isbell, Calcagno et al. 2011, Lefcheck, Byrnes et al. 2015, Meyer, Ptacnik et al. 2018). Further, we showed that plot identity effects, including functional group richness and the presence/absence of individual functional groups, were important drivers for trade-offs between EFs. While we tested for identity effects of plots, there are likely individual plant species that cause these trade-offs by maximising some EFs at the expense of other EFs. When future research can identify such plant species with strong effects on trade-offs, land management can target low densities of these disadvantageous species to reduce trade-offs between EF and promote multifunctionality. Nevertheless, competition for resources and the resulting ecological trade-offs between EFs are challenging to resolve.

CONCLUSIONS

Our study showed that even under the controlled conditions of a single experimental field site, correlations among EFs were variable over time. Consequently, repeated measurements of EF are needed to avoid spurious and non-generalisable conclusions about relationships among EFs.
Moreover, our results show the potential for land management to promote multifunctionality in both the establishment and the management phase by incorporating two principles. First, maintaining or increasing biodiversity of grasslands, which we showed increases synergies among EFs, promoting multifunctionality. Second, reducing negative identity effects by reducing the proportion of disadvantageous species or communities as indicated by the strong effect of plot identity strengthening trade-offs.
Future studies should continue to investigate the drivers of EF relationships to identify common drivers causing trade-offs and separate common drivers from potential ecological constraints. Importantly, these studies should also address how environmental conditions can change these relationships and identify influential species enabling recommendations on how to adapt management for maximising multifunctionality.