Figure 2: Variation in the correlations between pairs of ecosystem functions (EFs) (lower triangle) and average correlation between these EFs (upper triangle). The different EFs (see list in table 1) were grouped into classes. Mean correlations were calculated using Fisher’s Z transformation of EF correlations per season and year that were averaged over time. The standard deviation of the EF correlations per year and season was calculated to estimate the variation of EF correlations. When no average correlation is shown, the respective EF was not measured in the same season and year. A missing standard deviation for an EF pair shown to have a correlation coefficient represents cases where a correlation coefficient could only be calculated for a single time point.

Drivers of the covariance of EF pairs

To test if the drivers year or season, species richness, and plot identity affect relationships among EFs, we quantified the covariance between all pairs of EFs and the contribution of each driver to these covariances in percentage. These percentages were signed because the drivers can contribute to the EF covariances by affecting the underlying EFs synergistically (signed positive) or antagonistically (signed negative). Importantly, the contribution of individual drivers can have antagonistic (more negative covariance) or synergistic (more positive covariance) effects irrespective of the overall relationship between the respective EF being a synergy or a trade-off.
All tested drivers (year, season, SR, and plot ID) affected the covariances between EFs. The largest fraction of covariance among EFs was explained by SR and plotID. However, effects differed between EF pairs in synergies and trade-offs (defined by the sign of the mean correlation, Fig. 2, upper triangle). For synergies, most of the covariance was explained by SR (26.5%). In contrast, for trade-offs, most of the covariance was explained by plot identity (–29.5%, Fig. 3), with the negative value indicating that the individual EFs were driven antagonistically, causing a trade-off. When further investigating plotID, the presence of herbs and legumes already explained half of the effect of plot ID (Supporting information F, Table S4). For synergies, plot ID had intermediate positive effects (18%), mainly due to the presence of grasses and herbs. Year and season caused both positive and negative covariances, so that the average percentages explained by year and season were low (2.8% and -0.6%). For trade-offs, the average percentage of explained covariance by SR was low (–4.2%), contributing positively and negatively to covariance. Season contributed an additional –12.1% to covariance, while year explained very little (-1.3%). Interactions between drivers explained very little covariance (Supporting Information E, Table S3). Unexplained residual covariance was, on average, |4.5%| of the covariance (for synergies 5.5% and for trade-offs –3.6%), suggesting a low amount of random covariation between EFs.