Discussion

Although there was a high uncertainty in our results, we found that across the world’s biomes and ecosystems, patterns in species richness in the seed bank often diverged from patterns of seed density. Strikingly, we also found that the richness of soil seed banks appeared higher in undisturbed compared to degraded habitats, suggesting how across the world’s ecosystems, soil seed banks also reflect the ongoing depletion of biodiversity as a result of anthropogenic activities23,34. This indicates that the potential for seed banks to contribute to the recovery and restoration of degraded ecosystems might be limited. Seed density, on the other hand, was not always lower in disturbed habitats, showing how different aspects of biodiversity can show diverging patterns in response to global change pressures35. Our estimated global average of more than 5000 seeds m−2 also means that in the broadest sense, if you put your finger to the earth, you are likely to be pointing at a seed. The high density of seeds in the soil is a testament to the formation of soil seed banks as an important component of plant population and community dynamics, and a valuable strategy for many plant species to facilitate establishment following disturbance3,6,36.

Our analyses revealed substantial variation in the diversity of the soil seed bank among the world’s ecosystems, identifying divergent patterns between species richness and seed density. Although the broad relationship between total abundance and species richness is well-established35,37, our understanding of plant biogeography and trait-environment relationships helps us to interpret these results and put them into context. Overall, species richness in the soil seed bank reflected the latitudinal gradient in biodiversity29,30,38. However, while tropical and Mediterranean ecosystems exhibited relatively high richness compared to temperate, boreal and tundra regions, the former areas differed strongly in terms of seed density. Mediterranean forests and shrublands exhibited both high richness and density in the seed bank (Figs. 2–4, Tables S3–S6). In regions with warm climates and frequent extreme climatic events, the soil seed bank can play an important role in vegetation recovery and stability of biodiversity39,40. Therefore, the high plant biodiversity in these regions41, together with the large proportion of species producing persistent seeds42, results in the relatively large and rich seed banks in these regions. On the other hand, tropical ecosystems host relatively few species that produce seeds that are able to survive desiccation13,14, which is a vital aspect of longevity in the soil. Moreover, the warm and often moist conditions combined with high diversity of other taxa mean that seeds of species that are able to form seed banks are subject to degradation, predation and attacks from pathogens43–45. Together, these mechanisms result in low soil seed bank density, but the high overall diversity in the vegetation means that species richness is still high relative to other ecosystems.

The relatively low richness in the soil seed bank in temperate and boreal regions matched expectations relating to broad gradients in biodiversity, and previous work suggests that this effect may be compounded by lower proportions of species in these regions producing persistent seeds compared to lower-latitude species12. In these regions, the divergence in patterns of richness and density was more apparent when comparing ecosystems, with density clearly higher in grasslands than forests. The vegetation in the temperate and boreal zones can be considered more stable compared to other regions, and this is especially true in forests. Because seed banking is expected to provide a competitive advantage in more dynamic systems46,47, it follows that these relatively stable systems are instead dominated by long-lived perennial species, for which regeneration via the soil seed bank is a less important strategy than for example clonality48. In alpine and tundra environments, low seed bank density and a high proportion of clonal species could also be expected, in this case due to a low diversity and density of adult individuals, combined with the risks of seed production during short growing seasons49,50.

Our results reveal an imprint of anthropogenic activity on soil seed banks worldwide, and further variation in the relationship between richness and density. While richness was consistently lower in degraded systems, reflecting patterns in degraded habitats in the above-ground vegetation51–53, the effect on density was more variable. Unsurprisingly, arable soils contained among the highest number of seeds per square metre (Figs. 3 and 4). The high magnitude and frequency of disturbance in arable fields supports relatively species-poor communities of opportunistic ruderal “weed” species whose traits include both high seed longevity in the soil and high seed production20,21, and that can be problematic for agricultural production54–56. In forest ecosystems, lower richness in the seed bank was often accompanied by higher seed densities. Again, this can be related to plant species’ adaptations to disturbance. While undisturbed forests are relatively stable systems, degraded forests often retain the imprint of the more dynamic management regimes that they are experiencing or recovering from, and therefore their seed banks reflect plant communities that are adapted to such highly-disturbed conditions57–59. However, in grasslands and other habitats, degradation generally appeared to result in both reduced richness and density in the soil seed bank. Our findings are concerning, matching general trends of reduced biodiversity in the established vegetation over time60,61 (but see ref. 62), and have consequences for the potential of passive restoration to reverse biodiversity losses63,64.

In addition to detecting differences in species richness in the seed bank among ecosystems, we also found differences in the accumulation of richness in space. Our findings support those of Vandvik et al.25, who observed that species accumulate more rapidly in alpine and subalpine grasslands than in boreal zones (empirical), and in forests compared to grassland and heathland (literature), and we extend the results to include other global biomes. We found that richness estimates across scales (in our case 0.01 m2 to 15 m2) appeared to show relatively large differences in Mediterranean and tropical ecosystems compared to temperate ecosystems, demonstrating how species’ spatial distributions affect how we interpret relative differences in biodiversity across regions (Table S3). These findings have implications for how we understand biodiversity in soil seed banks at both local and regional-global scales. At local scales, we found that differences in richness across ecosystems, but critically also between undisturbed and degraded ecosystems, were marginal to non-existent at the 0.01 m2 scale. This shows that making comparisons at small spatial grains could inhibit our ability to detect important differences in biodiversity that might be relevant for conservation management. At larger spatial scales, comparisons among regions can depend on the spatial grain of predicted richness. This has previously been shown in the aboveground vegetation, where Sabatini et al.30 identified a clearer gradient in species richness at large spatial grains, while Keil and Chase29 found that areas of relatively high beta diversity in trees—where richness was relatively low at smaller scales and relatively high at larger scales—to be concentrated in low latitudes. Our modelled patterns of soil seed bank richness at the 0.01 m2 scale were broadly similar to those found by Yang et al.26, although their findings of low richness in the tropics and (predicted) higher richness in northern regions were not supported when considering larger spatial scales.

The large number of soil seed bank studies that have been carried out across the world has allowed our synthesis to cover a greater global extent than many large reviews of biodiversity, which often suffer from biases relating to the absence or restricted availability of data from some regions65–67. However, our study may still suffer from other biases, which we discuss here. First, seed banks have not been sampled randomly across the world, but in places where scientists have been interested in finding and identifying seeds in the seed bank68. This means that our results rest on the assumption that seed banks were not consistently sampled in areas that are relatively rich or poor in seeds or species, compared to areas that have not been sampled. Second, the database is made up of studies using different sampling efforts to sample the soil seed bank. To address variation in sampled area, we used an approach (ref. 29, ‘Methods’), whereby the putative drivers of variation in biodiversity are allowed to vary with sample area in our statistical models, meaning that not only the direct effect of area is controlled for, but also the fact that species-area relationships can also vary across ecosystems and in response to degradation28,69. While we also controlled for sampling extent (number of sites) in our models, around half of all observations in the database were represented by a single site. It is well-known that sampling multiple sites per treatment gives improved confidence in ecological findings (sample replication70), and we echo previous calls for seed bank sampling to improve representativeness71,72. Third, there was a lack of balance in the number of observations across ecosystems, with some better represented than others. Although we hierarchically modelled each biome separately, the distribution of observations among ecoregions and between undisturbed and degraded observations were often uneven. This will have affected our statistical power, increasing uncertainty in our model estimates, particularly where sample sizes were low.

There is another aspect of the study that will have contributed to the uncertainty in our results that is not related to the initial data collection. Our binary categorisation of ecosystem degradation represents a major simplification of the effects of degradation on biodiversity. While we are convinced that such simple classifications can be useful and informative in large-scale studies (see e.g. refs. 23,63), our general lack of clear differences between undisturbed and degraded ecosystems are likely related to the fact that systems undergoing environmental change are biased towards increasing richness over time73. That is, recently degraded or restored ecosystems are likely to contain both colonising species representing the new regime, and species representing the previous regime, which have not yet disappeared. This is especially relevant in the context of seed banks, which exist to buffer short-term environmental heterogeneity, and whose communities often reflect previous habitat conditions74,75. In more long-term degraded ecosystems, we may expect the negative effect on seed bank richness to be more clear. It is also important to note that the seed bank dynamics of—for example—a young post-agricultural forest are likely to be quite different from those of a mature plantation, both of which would be classed as degraded forests here. Therefore, while our results may be broadly representative of average effects of degradation at the global scale, the understanding of specific systems undergoing degradation or restoration will require more detailed investigation.

What can we learn from studying the soil seed bank? In addition to the continual publication of studies investigating soil seed banks at local scales, there has been an upswing in the synthesis of large datasets to understand broad spatial or ecological patterns relating to soil seed banks. This includes the macro-environmental determinants of small-scale richness and density26, and the evolutionary history of seed bank persistence76. Here, we complement such findings with an understanding of the broad biogeographical patterns in soil seed bank richness and density, as well as the spatial scaling of species richness in the seed bank. Other large-scale work considers the important role that soil seed banks may play in species and community responses to global changes, related to the trade-off related to seeds being able to persist or disperse in space77, or the role of soil seed banks in facilitating biological invasions78. We believe that the future of soil seed bank synthesis lies beyond simple measures of community richness and density, or categorisation of species as having or not having persistent seeds. Collating a large number of datasets including community data from the soil seed bank and established vegetation—not forgetting to standardise according to spatial scale—across biomes and systems (e.g. ref. 24), would allow us to delve deeper into a mechanistic understanding of the role of soil seed banks, biodiversity, restoration and responses to global change.