1. We investigated the vertical distribution of seeds in the soil, using data from nine studies in five European countries. We discovered significant correlations between seed shape and distribution in the soil.
2. The classification of the longevity of seeds of plant species has been improved by the introduction of a ‘longevity index’, expressing on a continuous scale the most recent information on seed longevity represented as the proportion of non-transient seed bank records in the database of Thompson et al. (1997). Remarkably, no difference in seed longevity was found if the index was based on direct observations only when compared with the index based on the complete data set where indirect, ‘depth-derived’ observations were included.
3. Seed longevity was best estimated using a multiple regression model with an integrated measure of seed size and shape and depth distribution of seeds.
4. The shape of seeds, known to be a consistent character of species, was shown to be constant within species, whereas depth distribution of seeds was highly variable among sites. This is consistent with the variability of seed longevity found in published seed bank data.
It is often assumed that viable seeds deep in the soil are older than those nearer the surface (e.g. Harper 1977; Brown & Oosterhuis 1981; Bakker 1989; Leck 1989; Bakker et al. 1991). Because seed movement in soil is slow, short-lived seeds are supposed never to reach greater depths in any quantity. This assumption was one of the basic ingredients of the key designed by Thompson, Bakker & Bekker (1997) to assign seed data derived from a single sampling occasion to one of three categories of persistence. This key infers seed persistence from direct estimation of seed longevity through burial experiments, from site historical data and from the depth distribution of seeds in the soil. However, considering all data assembled from different studies, the database of Thompson et al. (1997) has shown that the absolute category of seed longevity of a single species is hard to define. Until now direct experimental evidence for the value of depth distribution in indicating seed longevity has not been available. However, because we now have access to a large data set on the estimation of seed longevity where this principle was applied, the time has come to evaluate this principle.
Thompson et al. (1993) found a relation between shape and mass of seeds and seed longevity. They classified the longevity of species in two groups, transient and persistent, mainly based on their expert judgement and the then incomplete database of Thompson et al. (1997). They concluded that seeds that weigh < 3 mg and have a variance of seed dimensions (see Materials and methods) < 0·14 [0·093, according to Bakker et al. (1996)], i.e. small and nearly spherical seeds, are persistent. Other research points in a similar direction. Transient seed bank species had significantly heavier seeds and higher variance of seed shape than persistent species in a field study of the limestone Alvar grasslands of Öland, Sweden (Bakker et al. 1996). Also a group of species of flooded meadows near Oxford, classified as having a transient seed bank, had a significantly higher variance of seed shape than a group of species classified as persistent (McDonald, Bakker & Vegelin 1996).
The first aim of the present study is to search for correlations between mass, shape and depth distribution of seeds in results of several field studies. The expectation is to find that large, heavy and/or non-spherical seeds are largely confined to surface layers of soil. Second, we aim to identify the attribute [mass (M), shape (Vs) or depth distribution (D) of seeds in the soil] or combination of attributes that best predicts seed longevity.
For this purpose we first present a longevity index (LI) based upon the database of Thompson et al. (1997). In contrast to the classification of seed bank types in three categories (transient, short-term and long-term persistent), this continuous index does justice to the wide range of seed bank types that is found within species included in the database, thus expressing the most accurate information on seed longevity.
Our final aim is to compare the different estimates of seed longevity based upon the attributes of seeds for species found at different sites. The expectation is that species exhibit little variation in seed size and shape but much variation in depth distribution. We make predictions for six species which have been recorded from at least two different sites. The sites from which the data used in this study are derived were located at five European countries.
Materials and methods
Seven field studies on the soil seed bank of species-rich grassland communities and two studies on the seed bank of the understorey vegetation of a woodland and young deciduous forest on lake shores have been selected for this study. The individual trials had to meet the following criteria for adoption.
1. The data of the field study had to be independent of the seed bank database of Thompson et al. (1997). Data from the study on the woodland understorey vegetation were already incorporated in the database but these data were excluded before the longevity index was calculated.
2. The soil seed bank sampling had to be carried out in spring before fresh seed rain had occurred, to detect naturally stratified and possible long-lived seeds. Sampling should have been carried out in at least two soil layers, preferably 0–5 cm and 5–10 cm. The seeds found had to be checked for their viability.
Furthermore some criteria were set for the suitability of individual species at each site.
1. For each site fresh locally collected seeds of the species found in the local seed bank had to be measured in three dimensions; length, width and height. The measurements should have been carried out on the seeds (or fruits) in their simplest form, i.e. without hairs, hooks, pappus, glumes or other detachable appendages.
2. Only abundant species were included in the analyses: ≥ 10 seeds in total were found in the soil seed bank samples of a site or, if seeds were lacking in the soil, the species had to be abundant in the established vegetation (see also paragraph on depth distribution in Materials and methods).
3. Species with a longevity index based on fewer than five different observations in the total database of Thompson et al. (1997) were excluded.
4. If sites were sampled in a successional series or in the same area, the number of seeds found in sites that are considered to have the same soil type and management regime were lumped per soil layer to meet the criterion of at least 10 seeds found.
The first two series are chronosequences of species-rich grassland restoration in the brook valley system of the Drentse A Nature Reserve, the Netherlands (53° 5 ’N, 6° 42 ’E). All four sites of a series vary in ‘age’ after cessation of fertilizer application and therefore in productivity; the sites with the longest period without fertilizer are the most species-rich (Bakker & Olff 1995).
1. Anlo-wet. The wet series is situated on a peaty soil overlaying a sandy subsoil receiving superficial seepage water. The species-poor start (an average of 17 spp. m–2) of the series is dominated by Ranunculus repens and Holcus lanatus. A species-rich community (an average of 30 spp. m–2) with a dominance of Juncus acutiflorus has developed after 26 years of restoration management.
2. Anlo-dry. The dry chronosequence lies on the edge of the brook valley system on a humic soil overlaying sandy subsoil in a dry infiltration area. The vegetation succession in this chronosequence starts with a H. lanatus stand and culminates in a low productivity Agrostis capillaris–Luzula campestris community after 26 years of restoration management.
The results of the seed bank analyses of the four stages of both the wet series and the dry series have been lumped to represent one site on peaty soil (1) and one site on sandy soil (2). Vegetation descriptions of both series can be found in Bakker (1989) and Bakker & Olff (1995). Seed bank data will be presented elsewhere (R. M. Bekker, unpublished data).
3. Alvar. A series of sites in which cattle grazing ceased 0, 20, 50 and 80 years ago in the large alvar area ‘Stora Alvaret’ on the Baltic island Öland in Sweden (56° 35 ’N, 16° 34 ’E). These very species-rich calcareous grasslands (an average of 40 spp. m–2) of Öland are found on relatively deep brown soils overlaying limestone. The most species-rich site of this series is still grazed by cattle and harbours a Veronica spicata–Helictotrichon pratense community. Data on the soil seed bank and descriptions of the established vegetation can be found in Bakker et al. (1996).
4 Sättra. The wooded grassland studied is situated on mineral soil in a nature reserve near Sättra (58° 16 ’N, 14° 49 ’E) in southern Sweden. It has been mown yearly and successively summer grazed for many centuries. Half of the data set originates from plots that had been ungrazed for 18 years. Seed bank data in relation to vegetation data are presented by Milberg (1995).
5. Hjälmaren. A successional vegetation on islands created by a drop in the water table of 1·3 m 100 years ago in Lake Hjälmaren (59° 15 ’N, 15° 45’E) in Sweden. The composition of the soil seed bank of the understorey of the present forest dominated by Populus tremula, Betula pendula and Alnus glutinosa was estimated and compared with the former and present established vegetation (Grandin 1996).
6. Laelatu. A series of wooded meadows (Filipendula vulgaris–Sesleria caerulea community) at Laelatu (58° 35 ’N, 23° 33 ’E) on the western coast of Estonia. The site is situated on old shore ridges of limestone shingle, overlying gleic rendzic leptosol (humus layer c. 20 cm, pH about 7) (Sepp & Rooma 1970). The established vegetation and the soil seed bank were analysed in two open grasslands and two overgrown grasslands (Kalamees & Zobel 1997). Because the grasslands are so close to each other, contain few species in the soil seed bank and have the same substrate, the sites are lumped into one set in this study.
7. Schwäbische Alb. Two calcareous grassland sites situated in the Schwäbische Alb (48° 30 ’N, 9° 45 ’E) in southern Germany, the subject of studies of seasonal dynamics of the seed bank and established vegetation (Poschlod & Jackel 1993). Both sites had been sampled monthly and the results of the sampling of both sites in the spring months (February, March and April) were lumped per soil depth.
8. Woodlands. Seed bank and established vegetation in five different woodlands in south-west England (50° 30 ’N, 4° 30 ’W) (Warr, Kent & Thompson 1994). Different stands of tree species were investigated for their soil seed bank and understorey vegetation. Again in this case we considered all sites as one. For species that occurred in more than one site we calculated an individual mean depth distribution.
9. Zure Dries. The chalk grassland site Zure Dries (50° 47 ’N, 5° 45 ’E) is located in the Savelsbosch Nature Reserve in the south of the Netherlands. Established vegetation and soil seed bank of this site were investigated by Willems (1988, 1995). Soil seed bank data from January, March and April have been lumped to meet the species admission criteria.
Thompson et al. (1997) presented a list of observations per species regardless of the specific region, habitat or soil type from which the observations originate. Each observation was allocated to one of three seed bank categories, transient (seeds surviving less than 1 year), short-term (seeds surviving 1–4 years) or long-term persistent (seeds surviving for > 4 years in the soil), by the strict application of a decision model in the form of a key. For many species these observations are very variable. The conservative approach adopted by the key tends to underestimate longevity. Quite a few of the short-term persistent records would be properly allocated to the long-term persistent category if the study had been performed for longer or if better vegetation data had been available. For the classification of longevity of seeds of a wide range of species, we decided therefore to distinguish only between ‘transient’ and ‘persistent’ (summed short-term persistent and long-term persistent observations) (see also chapter 4 of Thompson et al. 1997). The following index has been developed which enables us to allocate species a single, continuous longevity index:
longevity index (L) = (SP + LP)/(T + SP + LP),
where (SP + LP) represents the total number of short-term + long-term persistent records and (T + SP + LP) the total number of transient + short-term + long-term persistent records in the database of Thompson et al. (1997). The longevity index ranges from 0 (strictly transient) to 1 (strictly persistent). We think this index represents the best ‘Golden standard’ of seed longevity available at present.
The database was so constructed that we could filter out the seed bank type observations that were purely derived from burial experiments and one sampling layer trials, the ‘non-depth-derived’ records. In this paper records of species were only used when they were not derived from the depth distribution of seeds in the soil. However, a comparison between the index calculated from all the available data per species and the index calculated without the ‘depth-derived’ records revealed a highly significant correlation for the set of species involved in the present study (Fig. 1). This, explicitly, indicates already that involving the depth distribution of seeds in the soil in the estimation of seed longevity points is valid.
INTEGRATION OF SEED MASS AND SHAPE INTO ONE SEED ATTRIBUTE
Seed mass (M) is obtained from 1/100th of the mass of 100 propagules in their barest form (expressed in mg). Seed mass in this study ranges from 0·01 mg (Juncus spp.) to 107 mg (Prunus spinosa).
The shape of seeds (Vs) can be captured by measuring length, width and height of a seed and dividing all values through length so that length is unity, and then calculating the variance of these three values by dividing the summed squared deviation from the mean by n = 3: Σ (x – x¯)2/n. In this way the shape becomes dimensionless and can vary between 0, perfectly spherical, and 0·2, shaped like a thin disk or a slim needle. Thompson et al. (1993) proposed the mean of five individually measured seeds or fruits as a standard measurement. Here we measured five individual seeds or fruits in their simplest form, because we still do not know exactly which parts of the seed will fall off or quickly decompose before the seed is incorporated into the soil. In the present study the variance of seed dimensions Vs ranges from 0·003 (Gentianella germanica) to 0·177 (Leontodon hispidus).
It seems likely that if seeds are very small, differences in form (Vs) will not affect incorporation into the soil very much. However, if seeds are large and heavy it seems likely that round seeds will penetrate the soil more easily than awkward-shaped seeds. This reasoning deliberately does not consider other factors like biotic selection mechanisms, e.g. palatability of seeds to animals. To combine seed mass and shape into one parameter we multiply mass by the square root of variance of seed dimensions (M×√Vs), which stresses the primary effect of mass over shape as explained above. A square root transformation was also suitable to reduce the scale factor in at least one of the parameters. We concentrated on M×√Vs as the integrated seed attribute in this paper.
All surveys included in the present study sampled the soil seed bank from at least two depths. In this way it was possible to express the depth distribution of seeds in the first two layers as the percentage of the total found in the upper layer (0–5 cm in most cases). If unequally thick layers had been sampled the figures were recalculated appropriately. This was the case for the Schwäbische Alb where 0–2 cm and 2–6 cm were sampled. These data were proportionally recalculated to the layers 0–2 cm and 2–4 cm by assuming a linear decrease of seed density with increasing depth.
If no seeds were found but the species was abundant in the established vegetation the species was included in this study and considered to have a transient seed bank, according to the key of Thompson et al. (1997). In these species the percentage of seeds in the upper layer was arbitrarily set to 100%.
M, Vs and M×√Vs were log-transformed before analysis. For each site separately a linear regression was conducted on log-transformed data of M×√Vs against depth distribution and the longevity index. All linear regressions employed the package SPSS version 5·01. We also looked for outliers in three linear regressions of the overall data set, i.e. all species × site combinations: 174 points in all. Seven of the 10 worst outliers were from the limited data set of Hjälmaren. As a result of this the Hjälmaren set was excluded from the overall regression analyses. One further outlier, P. spinosa of the Alvar data set, was also excluded from the whole set. The seed mass of this shrub species, 106 mg, is more than 100 times the average seed mass of all other species included in the present study.
With the remaining data set (n = 153) from all sites, except Hjälmaren (excluded after the outlier analysis), linear regression analyses were conducted between depth distribution and log (M), log (Vs) and log (M×√Vs) (three separate analyses), as well as regressions of the longevity index against log (M), log (Vs) and log (M×√Vs) and depth distribution (D), in four separate analyses.
Multiple regression analyses were also carried out with the software package SPSS version 5·01. The equations were obtained using a forward stepwise regression, the ‘probability to enter’ was 0·05. A graphical presentation of the multiple regression analysis in the form of trend surface analysis was performed with the package Vegrow 4·0 (Fresco 1991), where, by means of a posteriori classification, all data were assigned to 5 equally large classes of the longevity index: 1, <0·20; 2, <0·40; 3, <0·60; 4, <0·80; 5, ≥0·80.
For practical purposes only all figures display untransformed values on a log-scale, and include the outcome of the regression analysis which has been calculated with the transformed data.
Linear regression analysis of depth distribution and seed mass and of depth distribution and seed shape over all sites but Hjälmaren gave significant correlations (Table 1). The relative representation of heavy and/or awkward-shaped seeds in the upper soil layer is higher over a wide range of species and field sites. The proportion of variance accounted for by these regressions was, however, very low. The combination of seed mass and shape resulted in an increased but still low R2 of 0·19 (Fig. 2).
Table 1. . Linear regression models plus statistical analysis of log-transformed data of seed attributes mass (M), shape (Vs), their combination (M×√Vs), with depth distribution of seeds in the soil over all sites excluding Hjälmaren: *P < 0·05, **P < 0·01, ***P < 0·001
Multiple regression of log-transformed seed mass and seed shape and their combined attribute against depth distribution shows a significant correlation (P<0·001) with an R2 of 0·22 (Table 2). All other models with less or more factors, i.e. including ‘site’ as a factor, did not improve the level of significance.
Table 2. . Multiple regression models plus statistical analysis for depth distribution and longevity index over all sites excluding Hjälmaren: *P < 0·05, **P < 0·01, ***P < 0·001
Analysis of the whole data set (n = 153) showed that the negative correlation of the longevity index against the depth distribution of seeds is highly significant (P < 0·001) with an R2 of 0·22. This indicates that the longevity of seed of species with equal or more seeds in a deeper layer is higher than species with more seeds in the surface layer. Nevertheless, a wide range of depth distributions can be found at each point of the longevity index (Fig. 3).
Seed mass rather than seed shape (Vs) gives a significant negative correlation with seed longevity (Table 3), with an R2 values of 0·34. The combined seed attribute M×√Vs gives the best fit with an R2 of 0·35 with P < 0·001.
Table 3. . Linear regression models plus statistical analysis of log-transformed data of seed attributes mass (M), shape (Vs), their combination (M×√Vs), with the longevity index over all sites excluding Hjälmaren: *P < 0·05, **P < 0·01, ***P < 0·001
The best model for the longevity index is the multiple regression between the combined seed attribute log (M×√Vs) and depth distribution, shown in Fig. 4. The overall correlation coefficient is 0·64 (P < 0·001). This trend surface plot shows the fit of the multiple regression through the plotting of the observed data points which have been classified a posteriori into the five equally large classes of longevity. The combination of log (M×√Vs), already the best single factor predictor for depth distribution, and observed depth distribution (D) at the site, is a useful tool for the estimation of seed longevity.
PERFORMANCE OF THE MODEL AT SINGLE SITES
After having determined the best single factor model for the longevity index without depth records over a wide range of species and sites, we tested the performance of this model at the local scale. For every site the linear correlations between the longevity index and the seed property log (M×√Vs) were calculated and the regression lines summarized in Fig. 5. The solid lines indicate a significant correlation (P < 0·05). The slopes of the solid lines are closely comparable, the range of intercepts is small. Four of the dotted, non-significant lines differ in slope and/or intercept from the set of significant lines but are closely in range with them. They belong to the sites Sättra, Zure Dries, Laelatu and Schwäbische Alb. The site Hjälmaren is widely out of range compared with the other sites. Over all sites a negative correlation was found, which gives additional support to the validity of the given overall relationship between seed mass and shape and longevity.
PERFORMANCE OF THE MODEL FOR SINGLE SPECIES
We compared the longevity of single species at different sites that results from the multiple regression as presented above, in order to gain an insight into which factor is most responsible for the large variation in seed longevity reported by Thompson et al. (1997). Figure 6 presents the observed field data on depth distribution and seed size property log (M×√Vs) and the outcome of the regression formula for longevity behind Fig. 4. of six species for each site at which they occur (at least two sites per species). Some species (Plantago lanceolata and Hypericum perforatum) appear to fall in different categories of longevity at different sites. The variation in seed size property, illustrated by the coefficient of variation in percentages, is rather small for five species but not for P. lanceolata (Table 4). The depth distribution varies a great deal within five of the six species among the sites. The most transient species, P. lanceolata, shows a very small range of depth distributions, with nearly all seeds in the upper layer.
Table 4. . Coefficients of variation (CV in %) of depth distribution and log (M×√Vs) for six species found at n different sites
Except for P. lanceolata, we can conclude that for all species shown here depth distribution, amongst many other factors, is a site dependent variable, which can mean that within the same species seed longevity varies between sites. However, the estimated longevity never varies more than 0·2, one class of the index. Variation in seed longevity seems unlikely to be caused by variation in the variance of seed dimensions.
This data set does not provide any evidence for the relationship between depth distribution and soil type, partly because only six species occurred in more than one site. A correlation with stand age could not be performed owing to lack of accurate data and the large differences between management of sites included in this study.
With the introduction of the longevity index, we are able to refine the classification of species as either transient or persistent, originally introduced by Thompson & Grime (1979). The index, which is based on the best available published data on seed longevity, is expressed on a continuous scale and allows us to rank species according to their persistence. The index itself is proven well enough to classify individual species provided that at least five observations per species are available, although clearly the accuracy of the index will improve with increasing numbers of observation per species. The conservative classification of species by Thompson et al. (1997) causes problems, particularly concerning species labelled as short-term persistent. Although this group harbours records of species which are correctly classified, some would be classified as long-term persistent with better data. The solution we have adopted was to lump this group with the long-term persistent group in calculating the longevity index. In this way these records were given the benefit of the doubt and the total number of species that could be classified from at least five observations rose significantly.
The fact that the index calculated with and without ‘depth-derived’ records were highly significantly correlated with a close to 1:1 relation through the origin, can be seen as the first proof of the hypothesis that depth ratios of species reflects seed longevity.
The best single predictor of longevity index was M×√Vs, negatively correlated with the index over all sites. This agreement with the study of Thompson et al. (1993) is remarkable when we consider that the index is based upon many kinds of observations, using a wide range of criteria, assembled in the database of Thompson et al. (1997). The present study indicates that the relation can also be applied on real communities.
We found that four individual studies out of nine showed significant correlations between M×√Vs and seed longevity. This shows that the index is also applicable to a small data set with few species. We did not find a common reason for the five sets that show no significant correlation, although two sets (Schwäbische Alb and Sättra) are sampled from 0 to 2/2–6 cm and 0–4/4–8 cm, respectively. Both layers of Schwäbische Alb fit entirely into the first layer of the rest of the studies. Sättra was not recalculated but harbours two species, Campanula persicifolia and Veronica officinalis, which are small seeded but mainly found in the upper layer, and two species that have large seeds but were mainly found in the deeper soil layer, Lotus corniculatus and Leucanthemum vulgare. The sets of Zure Dries and Laelatu might just contain too few data points.
Hjälmaren is widely out of range with the others, most likely because these data originate from a unique system. This site harbours a lot of species with large and/or disk-shape seeds that are apparently well adapted for dispersal by floating on the water surface, e.g. Alisma plantago-aquatica, Cicuta virosa, Solanum dulcamara, Alnus glutinosa (Van der Pijl 1982; Murray 1986; Kleinschmidt & Rosenthal 1995). Water dispersal was the most important factor for the colonization of the islands (Rydin & Borgegård 1991). Grandin (1996, unpublished data) has also found that seed mass is negatively correlated with increasing soil depth at the Hjälmaren site. However, the relation between seed shape and depth distribution was different from the one found in the present study. In his study slightly elongated seeds were associated with deeper soil horizons and spherical seeds were associated with surface soil. An explanation for the deviation of this site from the pattern found in the eight other sites could be that the depth distribution of the species involved was blurred by the impact of incidental flooding during storms (cf. Koutstaal, Markusse & De Munck 1987; Skoglund 1990; Huiskes et al. 1995) or that seeds from early colonists still are over-represented in the seed bank (Grandin 1996, unpublished data).
In Fig. 5 the lines representing a significant correlation diverge along the x-axis with increasing longevity index. This is consistent with the conclusions of Thompson et al. (1997) concerning the classification of single observations in the database (see also Materials and methods). Depending on the number of observations available per species, true transient species are more likely to be correctly classified than true long-term persistent species, because for the latter species false transient or short-term observation are likely. This is reflected in a better fit of the multiple regression at the lower classes of the longevity index than the more scattered observations of the higher classes of longevity (Fig. 4). Also few genuinely persistent species were included in this study. As most data were from grasslands this is in accordance with Bekker et al. (1997), who found very few long-term persistent species in a wide range of grassland communities.
We have shown that the use of seed mass and seed shape combined in one seed attribute is a useful tool for the prediction of the depth distribution of seeds in the soil. The field data over all sites show a positive correlation between these two factors. The best fit was obtained by the multiple regression of seed mass, shape and their combined attribute against depth distribution (Table 2). Considering the fact that these analyses were carried out with data for a diverse range of species, and that we did not consider any possible non-equilibrium between the established vegetation and the soil seed bank at some sites, the correlation is strong. Successional species in the vegetation (increasing or decreasing), because they were not deliberately removed from the data sets, could both contribute to the relatively low R2.
The depth distribution of individual species showed a wide range of variation. Poschlod (1993) also found that a depth index for many species was totally different at different grassland sites. Moreover, depth index of a species can depend on successional age of the site (Werner & Platt 1976; Poschlod 1993).
Evaluating the results of the single factor M×√Vs in individual studies we can state that this factor is a rough tool and gives an indication of seed longevity. The best model for the longevity index, however, was obtained from the multiple regression of the combined seed attribute M×√Vs and depth distribution of seeds. This analysis with independent field data contributes more evidence in support of the generally accepted assumption that seeds deeper in the soil live longer than seeds in the surface layer. This study made clear that seed mass and seed shape are two of the major determinants of the depth distribution of seeds in the soil and therefore also play a significant role in the estimation of seed longevity. Considering the fact that much seed transport in the soil depends on the activity of soil biota (Grant 1983), more experiments are needed to determine the importance of these factors for seed longevity.
The authors thank Geurt Verweij, Marcel Zandvoort, Leonard Bik, Marc Toom, Jörg Böhringer, Silke Jordan and Stuart Band for their help with the measurements of seeds and the soil seed bank analyses. Valuable ideas and comments on earlier manuscripts have been made by Jelte van Andel, Latzi Fresco and Roel Strijkstra. This research was made possible by financial support of the Netherlands Organisation for Scientific Research (NWO) Grant 805–35·855.