Global patterns of potential future plant diversity hidden in soil seed banks
Our global database was derived from studies measuring soil seed bank diversity and density of natural plant communities across all continents, albeit with a strong data availability bias towards North America, Europe, eastern Asia and Oceania as compared to elsewhere (Fig. 1). The database contains 15,698 records for soil seed banks worldwide, including 6,480 for diversity (represented here by species richness) and 9,218 for density (number of seeds per soil surface area). The database represents more than a century of research with the oldest publication dating back to 191815. This most exhaustive and comprehensive set of research data on soil seed bank to date allowed us to identify the determinants and patterns of soil seed bank at the global scale.Fig. 1: Locations of the soil seed bank studies included in our database.a Diversity; b Density.Full size imageTo make data among studies comparable, we standardized them using a three-step process. First, we identified soil seed banks that showed seasonal patterns in both diversity and density, all of which peaked slightly in winter (Supplementary Fig. 1a, b). Thus, we standardized all data (from non-(sub-)tropical regions) for other seasons to winter. Second, sampling area for soil seed bank diversity varied among studies, with 0.01 m2 being the most commonly reported (Supplementary Fig. 2), to which we standardized all data using a species-area curve (Supplementary Table 1). Third, sampling depth also varied among studies, with 0–5 cm being the most frequently reported soil depth (Supplementary Fig. 3). Therefore, 0–5 cm was chosen as the soil depth for standardization of data in various soil depths. Such standardization is needed to find the relationships of seed bank data between different soil depths. We used the upper and lower limits of soil depths (e.g., for 0–5 cm, the upper limit was 0 cm and the lower 5 cm). The log-scale regressions showed that both soil seed bank diversity and density decreased significantly with lowering upper boundaries of soil depths but increased with lower ones (Supplementary Table 2), and thus we standardized all data to 0–5 cm depth using these relationships. To account for possible variation among biomes, the second and third standardization procedures were conducted for each biome separately. The analyses during standardization confirmed the need to standardize empirical findings when comparing seed bank patterns across studies, as previously stressed in a study on grassland soil seed banks10. Our standardization procedures made all data comparable in terms of season, sampling area and soil depth.Non-parametric Kruskal–Wallis tests showed that soil seed banks differed significantly among ecosystem types. Mangroves, tundra and tropical & subtropical dry broadleaf forests had a lower diversity of soil seed banks, whereas Mediterranean forests, woodlands & scrub, tropical & subtropical moist broadleaf forests and tropical & subtropical coniferous forests had a higher diversity (Supplementary Fig. 4a). For density, mangroves and flooded grasslands & savanna had the lowest value, while temperate broadleaf & mixed forests and temperate conifer forests had the highest value (Supplementary Fig. 4b).Prior to spatial analyses, we computed semivariograms to determine whether spatial autocorrelation could affect our models. We found that there was no obvious spatial autocorrelation in the data of soil seed bank diversity or density (Supplementary Fig. 5), indicating no spatial dependence in our data. We then used the random-forest algorithm (see Methods for details) to determine the importance (as increase in node purity) of the influence of 31 variables related to climate, soil, human disturbance and spatial coordinates (Supplementary Table 3) on diversity and density of soil seed banks. These variables previously were reported to affect plant performance at the global scale16,17,18, and thus they could affect soil seed banks via their effects on seed production. Moreover, we expected that potentially these variables could affect seed longevity in the soil. Full models using all 31 predictors showed that climate and soil were important in predicting soil seed banks (Fig. 2a and Supplementary Fig. 6). Moreover, spatial coordinates (absolute latitude) were the most important predictor for diversity, i.e., diversity of soil seed banks exhibit clear spatial patterns at the global scale. Net primary productivity (NPP) and soil characteristics were important in predicting the density of soil seed banks (Fig. 2b and Supplementary Fig. 6).Fig. 2: Variable importance (increase in node purity) of random forests run with all 31 predictors.a Soil seed bank diversity; b Density. abs.latit, absolute latitude; AMT, annual mean temperature; AP, annual precipitation; ATR, annual temperature range; AVWAC, available water capacity (%); BULK, bulk density; CEC, cation exchange capacity; CLAYC, clay (mass %); diversity, plant diversity; HFP, human footprint; Isoth, isothermality; npp, plant productivity (net primary production); ORGNC, organic carbon content; PCQ, precipitation of coldest quarter; PDM, precipitation of driest month; PDQ, precipitation of driest quarter; pH, pH measured in water; Pseason, precipitation seasonality (coefficient of variation); PWeQ, precipitation of wettest quarter; PWM, precipitation of wettest month; PWQ, precipitation of warmest quarter; SANDC, sand (mass %); SILTC, silt (mass %); TCM, min temperature of coldest month; TCQ, mean temperature of coldest quarter; TDQ, mean temperature of driest quarter; TDR, mean diurnal range (mean of monthly (max temp–min temp)); Tseason, temperature seasonality (standard deviation *100); TWeQ, mean temperature of wettest quarter; TWM, max temperature of warmest month; TWQ, mean temperature of warmest quarter.Full size imageWe then built final random-forest models using the most important predictors of seed banks selected from full models: nine variables for diversity and five for density (Supplementary Fig. 7). Final models explained more of the total variance than did full models (Supplementary Table 4), and they were robust to K-fold cross-validation (Supplementary Fig. 8), indicating that a small number of variables predicted soil seed bank diversity and density. Absolute latitude (abs.latit) was the most important predictor for diversity, which varied between 0–55° and then decreased beyond this range (Fig. 3a). Five climatic variables were important for diversity. Diversity peaked at intermediate annual temperature ranges (ATR), while it was the lowest at intermediate mean temperature of driest quarter of the year (TDQ), precipitation of the coldest quarter (PCQ) and precipitation of the driest quarter (PDQ). Diversity increased with increasing annual precipitation (AP). In addition, three soil variables were important for diversity. Diversity showed a humped relationship with soil pH, with pH 6–7 having the highest diversity. Diversity increased with soil cation exchange capacity (CEC) and soil silt content (SILT). These results indicate that diversity exhibits strong spatial patterns at the global scale. However, our spatial patterns differ from those found for a specific ecosystem worldwide (e.g., grasslands), where there were only weak latitudinal gradients in seed bank diversity10. In addition, climate emerged as an important predictor for seed bank diversity, which is consistent with the report that climate acts as environmental filters affecting soil seed bank of grasslands around the world13. Our results agree with a continental study in Europe, where ATR was more important than mean annual temperature for determining seed bank richness and warmer temperatures were associated with lower seed bank richness11. Possible mechanisms by which temperature affects soil seed banks are that it (1) influences seed bank inputs via its effects on seed production; (2) cues dormancy-breaking and germination1, thus determining germinable seed output from seed banks; and (3) affects seed metabolic activity and soil fungal activity19, thereby determining seed viability and persistence in the soil. Finally, our findings of a significant effect of soil pH are supported by some regional and local studies. For instance, seed bank composition is significantly associated with soil pH at high elevations on the Tibetan Plateau20. A negative effect of low pH also has been reported in a large-scale study of acidic and calcareous grasslands in England21. Two possible mechanisms for the effects of soil pH are that (1) low pH may cause loss of seed viability due to the toxicity from aluminum or other metals that become more readily available in soils with low pH22; and (2) high pH may accelerate decomposition and promote growth of pathogens that negatively affect seed persistence23. In our study, the two mechanism may operate synchronously, thereby resulting in the highest diversity of soil seed banks at intermediate pH at the global scale. Further, our results show that soil CEC and SILT affect seed bank diversity, which agrees with a study on the Tibetan Plateau20. The physical and chemical properties of soils can affect seed bank directly by affecting seed germination and aging via regulating soil water-holding capacity24, or indirectly by affecting seed viability via controlling the activity of soil pathogens21,22,25.Fig. 3: Partial feature contributions (the marginal effect of a variable on response) of the most important variables for soil seed banks.a diversity; b density. Variable importance (inc. node) is the decrease in the residual sum of squares that results from splitting regression trees using the variable. The percentage increase in mean squared error (% inc. MSE) is the increase in model error as a result of randomly shuffling the order of values in the vector. abs.latit, absolute latitude; AP, annual precipitation; ATR, annual temperature range; BULK, bulk density; CEC, cation exchange capacity; npp, plant productivity (net primary production); PCQ, precipitation of coldest quarter; PDM, precipitation of driest month; PDQ, precipitation of driest quarter; pH, pH measured in water; SILTC, silt (mass %); TDQ, mean temperature of driest quarter; TWM, max temperature of warmest month.Full size imageFor soil seed bank density, soil bulk density (BULK) was the most important predictor; density increased below 750 g/cm3 BULK but remained stable when BULK was higher than 800 g/cm3 (Fig. 3b). Density peaked when temperature of the warmest month (TWM) was 34 °C. Density showed similar variation with NPP, precipitation of the driest quarter of the year (PDQ) and of the driest month (PDM), i.e., it peaked at intermediate values of these variables. Precipitation influences the success of sexual reproduction of plants and the size of the seed bank through seed input26, and it also affects soil pathogenic fungi, which cause seed mortality27. Therefore, precipitation has a strong effect on seed bank density, as reported for 27 alpine meadows on the Tibetan Plateau28. Our results further illustrate that PDQ and PDM are the key factors determining seed bank density worldwide, suggesting that moisture fluctuation in soils triggered by precipitation of the driest time of the year can affect seed bank density. If soil moisture fluctuations are high, seed germination will be primed by increasing moisture24.At the global scale, we mapped soil seed bank diversity and density using the final random-forest models. Mapping soil seed bank values onto global maps revealed considerable geospatial variation, the pattern of which varied between diversity and density (Fig. 4). For diversity, western North America, central South America, central Africa, central Europe, southern and eastern Asia and eastern Oceania had high values. In contrast, eastern and central North America, northern Africa and central Asia had low values (Fig. 4a). For density, northern North America, northern Europe and northern Asia had higher values than elsewhere (Fig. 4a). Our results are consistent with the reports that larger seed banks are more common in cooler temperate climates19,29. The latitudinal pattern of higher density in colder regions in the Northern Hemisphere may be driven by lower seed mortality in colder soils6, resulting in stable seed bank densities of long-lived seeds that counteract low seed production in some years at cold northern latitudes, as shown in a study of temperate forests along a 1900 km latitudinal gradient in northwestern Europe29. The latitudinal pattern highlights that particularly species rich low-latitude biomes such as tropical rainforests generally have very low seed bank densities, while their seed bank diversity does not exceed that in higher latitudes biomes. However, our global assessment should be interpreted with caution since some studies in azonal vegetation or in rare habitats in our database did not fully reflect soil seed banks in that region, and thus these data shortcomings may have induced bias in our global predictions. Moreover, data gaps in our database are also likely to have had an effect on the global predictions, i.e., fewer data available from some continents (e.g., northern Asia and Africa) could lead to less confidence for prediction in these regions. For example, Russia has very few soil seed bank data, which may have led to an inaccurate prediction for this country. Nevertheless, based on our global patterns of soil seed bank diversity and density, the latitudinal pattern strongly suggests that the biodiversity of (sub-)tropical forests is particularly vulnerable to large-scale climatic or land-use disturbances. However, in-depth investigation is needed to quantify the extent to which temporal integration of seed bank effects for long-lived trees and seed masting events may buffer the effects of low seed bank diversity and density at any given time of sampling. In contrast, the higher-latitude plant diversity, while currently low compared to that in tropical rainforest, may rely on high soil seed bank densities to boost its resilience to large-scale climate- or land-use induced disturbances. Further, our analyses suggest that the least vulnerable ecosystems in terms of hidden diversity should be those that combine high seed-bank diversity with high density; and therefore the relationships between the two variables across the global map certainly would be an interesting topic worthy of further study.Fig. 4: Extrapolated global maps of soil seed banks.a diversity in terms of number of species per 0.01 m2; b density as number of seeds per m2. In b, values are log10-transformed to facilitate viewing. The spatial resolution of grid cells is 5 arcmin-by-5 arcmin.Full size imageOur global assessment reveals that both diversity and density exhibit clear spatial patterns of soil seed banks but differ in their environmental determinants. These findings alone do not necessarily mean that this biodiversity reservoir has strong buffering capacity under climate change, because both climate and soil conditions influence seed bank diversity and density. Based on a large number and long history of studies globally, we provide quantitative evidence of how environmental conditions shape soil seed bank distributions and spatially explicit maps of this biodiversity reservoir in plant communities worldwide. Our quantification of environmental determinants and global mapping can be readily applied to dynamic global vegetation and plant diversity models to enable a more complete and accurate prediction of the impact of ongoing environmental changes on plant diversity (both above- and belowground) at the global scale. The next research challenge will be to plot current (visible) aboveground plant diversity (ideally using the available data in the studies themselves) against soil seed bank diversity under global change scenarios in order to pinpoint even more accurately which plant communities, ecosystems and biomes (and their turn-over) are most at risk of losing their diversity due to global changes. More
