Extended Data Fig. 4: Different modelling approaches yield similar results of temporal biomass change in the Jena Experiment.

All panels are based on replicate-level (n = 160) median point estimates and predictions from 1,000 linear mixed-effects models drawing from shuffled data subsets, avoiding the reuse of sampling events in multiple pairwise comparisons. Each panel compares predicted regression lines for temporal biomass change according to the different modelling approaches: Fixed baseline comparison (2010 as the baseline; red), moving average comparison (2010-2019 as baseline; blue) and restricted moving average comparison (2010-2014 as baseline; black; used in the main manuscript). Temporal biomass change was associated with a) species richness loss, b) species richness gain, c) species identity loss, d) species identity gain, and e) abundance change of persisting species (see Fig. 2 for detailed explanations). f) Total biomass change without partitioning. Solid lines indicate p < 0.05 (significant), dotted lines indicate p < 0.1 (marginally significant), dashed lines indicate non-significant relationships (p ≥ 0.1); exact two-sided p-values are provided in the panels. Shaded areas around the regression line represent 95% confidence intervals.