Results

Plant-Soil Feedback

At the species level (i.e., across soil types) there were six negative, one positive, and two neutral PSFs in 2017 (Fig. 1). The mean absolute value of species-level PSFs was 0.37 (the 95% confidence interval CI95 was 0.16 to 0.58). In other words, plants created soils that changed subsequent plant growth by 37 %. The arithmetic mean value of species-level PSF was -0.15 (CI95 -0.50 to 0.19). In 2018, five species demonstrated a species-level (Fig. 1), though neither absolute (0.35) nor arithmetic (-0.17) PSF values differed between 2017 and 2018 (tabs= 0.25; tart = 0.17; P > 0.05, DF = 8; paired t-test; Fig. 1).
At the soil-level, 27 of 72 PSFs were negative and eight were positive in 2017 (Fig. 2). The mean of absolute soil-level PSFs was 0.40 (CI95 = 0.34 to 0.45), while the arithmetic mean value was -0.14 (CI95 = -0.24 to -0.03). In 2018, absolute values (0.36, CI95 = 0.31 to 0.42), and the arithmetic mean (-0.15, CI95 = -0.25 to -0.05) were similar to and did not differ from 2017 values (tabs = 1.50; tart = 0.44; P > 0.05, df=71;paired t-test ) though only 13 of 72 soil-level PSFs differed from zero in 2018 (Fig. 2).
Competitive species demonstrated small PSF values and poor competitors demonstrated large positive or negative PSF values (Fig. 3). More specifically, RCI values from both the current and pre-existing experiments were correlated with absolute PSF values from 2017 (current: F1,134 = 28.7, P < 0.05, R = 0.18; pre-existing: F1,142 = 29.0, P < 0.001, R = 0.17) and 2018 (current: F1,134 = 29.4P < 0.001, R = 0.261; pre-existing: F1,142 = 50.06, P < 0.05, R = 0.18; Fig. 3).

Observed and Predicted Biodiversity Effects

Polycultures produced 55 % (current) and 40 % (pre-existing) more biomass than monocultures, respectively (Fig. 4; Supplementary Table 1). In both experiments, selection effects were greater than complementarity effects (Fig. 4; Supplementary Table 1). Selection effects increased with species richness in the pre-existing experiments, but complementarity effects were unrelated to species richness in either experiment (Fig. 4; Supplementary Table 1). Between experiments, community biomass, net biodiversity and selection effects were greater in the current than pre-existing experiments. Complementarity effects were smaller in the current experiment than the pre-existing experiment (Fig. 4; Supplementary Table 1). Predictions of biodiversity effects never differed between PSF and Null models (Supplementary Table 2). Null and PSF model predictions did not differ from observed biomass or net biodiversity, but Null and PSF model predictions were smaller than selection effects and larger than complementarity effects in the current experiment (Fig. 4, Supplementary Table 2).
In regression analysis of community biomass, Null and PSF models explained 28% and 30%, respectively of the variation in the current experiment and both explained 8% of variation in the pre-existing experiment (Supplementary Table 3). Similarly, for species biomass, Null and PSF models explained 42% and 40%, respectively of the variation in the current experiment and 38% and 36%, respectively of the variation in the pre-existing experiment.