Figure 1: Percentage of variation in individual EFs that was explained by year, season, species richness (SR), plot identity (plotID), and the interaction among these variables. The influence of the explanatory terms is plotted in % of total sum of squares, corresponding to increments in multiple R2 * 100. Explanatory terms are plotted for individual effects > 5%. All effects less <5% are summarized as ”other”, e.g. various interaction-effects. Hatched barplots represent a simpler model, including only year, SR, plotID and their interactions for EF, which were measured in only one season. Non-hatched barplots represent full models including all terms. The graph corresponds to a hierarchical partitioning of type one (Groemping 2006), but because explanatory terms were not correlated, there was no need to average across different fitting sequences.

Variation in EF correlations

Positive correlations (indicating synergies) and negative correlations (indicating trade-offs) were observed across all measures (Fig. 2, upper triangle). For instance, plant height and shoot biomass showed a synergy, while soil dissolved carbon and plant height showed a trade-off in their mean correlations. The strength of these correlations differed among pairs of EFs, with some EF pairs showed no correlation, while others showed weak, moderate, or strong correlations. We observed no strong negative correlations. All EFs showed positive correlations to some and negative correlations to other EFs (according to their mean correlation) (Fig. 2, upper triangle), with EFs in some classes showing predominantly positive correlations (plant productivity and invasion resistance) and others mostly negative correlations (plant productivity and plant nutrients). The EF correlations were robust against the method of calculating correlations, i.e. whether we used mean correlations (correlation coefficient averaged across time points), grand-total correlations (one correlation coefficient using all data from all time points), or between-group correlations (one correlation coefficient calculated with datapoints averaged across time points) (Supporting information D, Fig. S4). As expected, EF correlations tended to be stronger when the variation of individual measurements across time points was removed, i.e. for the between-group correlations (Supporting information D, Fig. S4).
The variation per EF pair was quantified by the standard deviation of correlation coefficients, which were calculated for every time point when the two EFs were measured in the same year and season. Overall, there was considerable variation in EF correlations (mean standard deviation = 0.16, mean correlation coefficient = 0.14) (Fig. 2, lower triangle). We tested if the variation in the EF correlations depended on the correlation’s average strength. EF pairs generally showed a higher variation in their correlation when they show a stronger correlation irrespective of this correlation being positive or negative (F5,281=9.4, p<0.001, Supporting information E, Fig. S5). Additionally, we tested if the variation in the EF correlations depended on the number of times the EF-pair was measured (number of years * number of seasons). EF pairs generally showed an increasing variation in their correlations with a higher number of times the EF-pair was measured (F1,572=120.91, p<0.001) (Supporting information E, Fig. S6). Lastly, we checked whether the variation in correlations among EFs depend on the identity of time points they were measured. These ranges of temporal variation were rather small, on average showing a standard deviation ± 0.08 (Supporting information D, Fig S7). Furthermore, the range of temporal variation of correlations, are different for the individual EF-pairs, some EF-pairs show a strong identity effect of time points (e.g. SoilNH4__SoilNmin, PlantCover__WeedCover) and some a weak identity effect of time points (e.g. ShootBM__SoilN, PlantC__SoilCorg) (Supporting information E, Fig S7).