Results

The global seed bank database31 contains the results of studies carried out in 94 countries across the world, including observations of more than 35 million seeds using a large range in sampling effort and extent (for a more in-depth summary of the data, see ‘Methods’). In such a large and heterogeneous dataset, there was generally a large variation in seed bank richness and density within and between ecosystems. Here, we describe noteworthy results in terms of overlap in 50% and 90% credible intervals (hereafter CIs) and consistency in the relative values of model estimates that are rounded to whole numbers (i.e. whole species or whole seeds m−2). All model results and further supplementary information can be found in Tables S1–S4, and Fig. S3, and full details of statistical models, including the definitions, interactions, inclusion and transformation of and between predictor variables are described in the ‘Methods’. Sample sizes can be found in Table 1.

Seed bank richness and density across ecosystems and biomes

Using models analysing species richness to predict average soil seed bank diversity for each ecosystem and at multiple spatial scales, we found that predicted species richness generally increased from the 0.01 m2 alpha scale to the 15 m2 gamma scale. Estimated richness was always higher at the gamma scale, with 50% credible intervals not overlapping in all undisturbed and 80% of degraded (including arable) systems. Relative values of predicted richness varied according to scale, with one example being that boreal and temperate grasslands had a higher predicted richness than tropical forests at the alpha scale (but with overlapping CIs), while tropical forests had higher richness at the gamma scale (no overlap of 50% CIs). Generally, differences in estimated richness across ecosystems were more clear at the gamma scale, so we place most focus there. In undisturbed systems, Mediterranean forests and shrublands were predicted to have the highest richness compared to all eleven other undisturbed ecosystems (no overlap of 50% CIs, overlap with six undisturbed ecosystems at 90% CIs; Fig. 2, Table S3). Tundra and aquatic systems generally had the lowest species richness, each having a lower estimate than the other ten undisturbed ecosystems, and lower than eight of the other ecosystems (50% CIs). Undisturbed tropical ecosystems were estimated to be broadly more species-rich than their equivalent temperate and boreal ecosystems. Estimated richness was higher for tropical forests than both boreal and temperate forests, for tropical grasslands compared to temperate and boreal grasslands and tropical arable compared to temperate and boreal arable, while tropical wetlands had similar estimated richness as temperate and boreal wetlands. Despite this consistency, estimates were only clearly different in forest ecosystems (50% CIs), although tropical wetlands were more species-rich than Mediterranean and desert wetlands.

Fig. 2: Scale-dependent richness in soil seed banks across ecosystems.Panels a–d represent realms, with the terrestrial realm separated into arable and non-arable systems. Each ecosystem is represented by four density curves and point-and-whiskers, showing species richness at the alpha (α; 0.01 m2; lower curves, smaller points) and gamma (γ; 15 m2; upper curves, larger points) scales, and in undisturbed (coloured curves and circular points) and degraded (grey curves and triangular points) ecosystems. Note that arable systems are not separated into disturbed and undisturbed and have only two curves per ecosystem and a square point. Points represent average predicted species richness, surrounded by 50% (thick) and 90% (thin) credible intervals. Density curves represent 24,000 predictive draws of the posterior distributions of overall richness at each scale, within each ecosystem and degradation group. Sample sizes can be found in Table 1. Values of modelled slope estimates and credible intervals are shown in Table S3.

The average estimated number of seeds m−2 in the soil seed bank was 4214 m−2 (median 3590) in the (non-arable) terrestrial realm, 9434 m−2 (median 11,001) for arable systems, 10,318 m−2 (median 10,802 m−2) in the transitional realm, and 8124 m−2 (median 9562) in the aquatic realm (Fig. 3, Table S4). Wetland ecosystems exhibited high seed densities, often higher than other ecosystems within their equivalent ecoregion (90% CIs). For example, tropical wetlands had higher seed densities than tropical forests, and temperate and boreal wetlands had higher seed densities than forests and grasslands in the same regions. Within the terrestrial realm, undisturbed temperate and boreal grasslands had the highest predicted seed density, higher than boreal forests and temperate forests (90% CIs).

Fig. 3: Density of seeds m−2 in the soil seed bank across ecosystems.Panels a–d represent realms, with the terrestrial realm separated into arable and non-arable systems. Each ecosystem is represented by two density curves and point-and-whiskers, showing soil seed bank density in undisturbed (coloured curves and circular points) and degraded (grey curves and triangular points) ecosystems. Note that arable systems are not separated into disturbed and undisturbed and have only two curves per ecosystem and a square point. Large points show the mean densities, surrounded by 50% (thick) and 90% (thin) credible intervals. Density curves represent 24,000 predictive draws of fitted (population-level) seed density, within each ecosystem and degradation group. Background jittered points represent empirical estimates of seeds density within sampled soil banks. Thick vertical lines in each panel represent the mean fixed effect value across a Terrestrial (non-arable records), b Terrestrial (arable records), c Transitional and d Aquatic realms. Sample sizes can be found in Table 1. Values of modelled slope estimates and credible intervals are shown in Table S5.

The effect of habitat degradation

In all ecosystems, habitat degradation always resulted in a lower estimated species richness at the 15 m2 scale in terms of predicted values, but again, despite this consistency, the difference was only clear in Mediterranean forests, woodlands and scrub according to the 50% credible intervals. At the 0.01 m2 scale, the predicted values were often the same in undisturbed and degraded ecosystems. Compared to species richness, habitat degradation had much more variable effects on seed density m−2 (Figs. 3 and 4). Broadly speaking, degradation was more likely to result in lower predicted seed densities compared to undisturbed habitats in open ecosystems, with clear differences (50% CIs) for tropical grasslands, Mediterranean and desert wetlands and aquatic ecosystems. On the other hand, seed density estimates were generally higher with degradation in forest ecosystems, with clear differences in both temperate and tropical forests (50% CIs). No ecosystems exhibited clear differences in richness or density with degradation according to the 90% CIs.

Fig. 4: The joint relationship between predicted species richness m−2 and seed density m−2 in the soil seed bank.Panels a–g represent biomes. Each ecosystem is represented by two sets of and point-and-whiskers. Points show the mean estimated species richness and seed density m−2, surrounded by 50% (thick) and 90% (thin) credible intervals. Circular points denote undisturbed and triangles denote degraded ecosystems. Arable systems are not separated into disturbed and undisturbed and have one (square) point with whiskers per ecosystem. Sample sizes can be found in Table 1. Values of modelled slope estimates and credible intervals are shown in Table S6.

Divergent patterns of richness and density

In many cases, relative patterns of richness and density in the soil seed bank differed both within and among ecosystems, and in response to degradation. Despite relatively high species richness, seed bank density was generally low in tropical systems. Tropical forests had the lowest density across undisturbed ecosystems, being lower than all eleven other undisturbed ecosystems (no overlap of 50% CIs, of which eight had no overlap of 90% CIs). Undisturbed tropical wetlands, forests and grasslands had a lower density than their equivalent ecosystems in temperate and boreal regions (50% CI). Disturbed tropical grasslands were similarly low in seed density, while arable ecosystems in tropical regions were lower than the equivalent systems in other regions (90% CI). Otherwise, arable systems combined relatively low richness with high seed density. Mediterranean and desert, and temperate and boreal arable ecosystems were predicted to have more seeds in the soil than all other ecosystems in the terrestrial realm (50% CIs), while arable ecosystems often had lower richness at the gamma scale than other degraded ecosystems within similar ecoregions (for example, temperate and boreal arable ecosystems had lower richness than temperate forests and temperate and boreal wetlands; 50% CIs). On the other hand, Mediterranean forests, woodlands and scrub combined high species richness with relatively high seed density (higher than all other forest ecosystems, 90% CI), while tundra combined its low species richness with low density, especially when degraded, where density was lower than all other degraded ecosystems (50% CI).