How long do tree seeds survive in soil seedbanks? A multi-species comparison in an old-growth deciduous temperate forest

ABSTRACT

We evaluated the performance of tree seeds in the soil seedbank of a temperate deciduous old-growth forest with respect to the total proportion of seeds that emerged as seedlings (P_TOTAL) and the average time seeds were in the soil prior to emergence as seedlings (T_AVERAGE). We analyzed 12 years of data on the seed dispersal and seedling emergence of 16 important tree species collected from more than 100 seed traps and quadrats. Using these estimates, we assessed whether seed banking is an effective regeneration strategy in forests compared to other life-history strategies. Persistence in the soil did not necessarily increase overall seed mortality. Phylogenetic constraints played a limited role in variations in P_TOTAL and T_AVERAGE among species, likely due to pronounced differences between closely related species. Estimated values of T_AVERAGE were mostly less than 1.5 years. Some species had a P_TOTAL greater than 0.20, despite having seeds with no known physical or chemical protection mechanisms. Seed size was not associated with seed performance. By contrast, greater per capita fecundity appeared to compensate for lower performance of banked seeds. Contrary to our expectations, species with longer T_AVERAGE values exhibited higher seedling survivorship, implying an absence of trade-offs between seed banking and seedling banking, and even implying a synergy between these strategies. In old-growth forests, seeds stored in soil seedbanks likely provide a buffer during years with little or no seed production; however, seedbanks may not promote adequate regeneration following natural or artificial large-scale disturbances that remove parent trees.

KEYWORDS:

• deciduous forest
• old-growth forest
• phylogeny
• seedbank
• tree regeneration

Introduction

The soil seedbank is an aggregate of ungerminated seeds in soil that represents a seed source for replacing adults following disturbance (Baker 1989). In mature or old-growth forest, disturbances create temporally unpredictable canopy gaps that function as safe sites for the seedlings of most tree species (S. Abe et al. 1995, M. Abe et al. 2002). Thus, species seedbank strategies are considered important for the regeneration and population-level persistence of tree species in mature or old-growth forests. However, it is necessary to verify whether this strategy is effective because seed banking may have disadvantages in terms of seed and seedling demography.

In contrast to the autotrophic stages (seedling and later stages), seeds frequently succumb to predation or infection by pathogenic organisms, and mortality risk increases with increased residence time in the soil (Notman and Villegas 2005). For example, the acorns of Fagaceae species, which are large and highly nutritious, often experience high mortality due to predation by herbivores in the post-dispersal stages (Greenberg and Zarnoch 2018), and smaller, apparently less nutritious seeds of other species experience heavy predation by wood mice or damage due to pathogenic fungi (Tanaka 1995; Masaki et al. 1998). As another potential disadvantage, in some deciduous forests in the north temperate zone, understory light levels are high in early spring prior to the leaf-out of canopy trees, and seeds banked in the soil experience a gap-like environment that can stimulate germination (Seiwa et al. 2009). This phenomenon can result in the conversion of soil seedbanks into seedling banks, which is disadvantageous to shade-intolerant, gap-dependent species because light supply decreases rapidly upon their emergence.

Given such risks, seedbanks might not necessarily be an effective means of in situ regeneration of trees, particularly in temperate deciduous forests. Alternative strategies, such as high fecundity and longevity of adult trees, might be more effective. However, outstanding questions must be clarified to understand the relative advantages and disadvantages of these strategies. It is imperative to determine how long disseminated seeds remain viable in forest-soil seedbanks, and what proportion of these seeds establish as seedlings. Moreover, to understand the role of seedbanks fully, these factors need to be quantified for a range of species in temperate deciduous forests. From seed germination experiments, Barak et al. (2018) showed that the timing of seed germination was affected by phylogeny as well as seed traits. This implies that assessing constraints on seed banking capacity imposed by phylogeny is essential for understanding the evolutionary context of variation in regeneration strategies among species (Davies et al. 2020).

In this study, we used the total proportion of seeds that emerge as seedlings (P_TOTAL) and the average time seeds remain in soil before emergence as seedlings (T_AVERAGE) at the community level, which can be derived relatively easily from field data as proxies of seed banking capacity. These estimates were based on 12 years of field data on the numbers of disseminated seeds and newly emerged seedlings for 16 major species in an old-growth temperate deciduous forest. Based on these estimates, we assessed the following questions. First, are there specific or phylogenetic differences in T_AVERAGE and P_TOTAL? Second, are T_AVERAGE and P_TOTAL inversely correlated? Third, are T_AVERAGE and P_TOTAL associated with seed morphology? And finally, do life history traits, such as higher per capita fecundity, the capacity to form seedling banks, and higher adult longevity, compensate for lower T_AVERAGE and P_TOTAL? Answering these questions allowed us to determine whether seed banking is an effective regeneration strategy for trees in temperate deciduous forests

Study Site and Field Methods

The study was conducted at the Ogawa Forest Reserve in Japan (36°56′, 140°35′, 610 m above sea level) (Masaki et al. 1992). The site consists of an old-growth deciduous forest comprising trees from the families Fagaceae, Betulaceae, Sapindaceae, Rosaceae and other families. The mean monthly temperature is 10.7°C, with a maximum of 22.6°C in August and a minimum of − 0.9°C in February. Annual precipitation is approximately 1,910 mm; August and September are the wettest months and December and January are the driest.

A 6-ha plot was established at the site in 1987. A 1.0-ha subplot (100 m × 100 m) was established at the center of the 6-ha plot in 1988 to investigate seed rain and seedling emergence. Emergence was defined as the presence of an erect hypocotyl above the soil surface with either cotyledons (epigeal species) or primary leaves (hypogeal species). The subplot was enlarged to 1.2 ha (100 m × 120 m) in 1989 (Tanaka and Nakashizuka 2002). More than 50 tree species with diameters at breast height (DBH) > 5 cm occur within the 6-ha plot. Of these, the dominant species with respect to basal area are Quercus serrata Murray, Fagus japonica Maxim., and Fagus crenata Blume (Masaki et al. 1992). As focal species, we selected 16 species with abundant seed rain and seedling emergence (Table 1): five Betulaceae species (Betula grossa Siebold & Zucc., Carpinus cordata Blume var. cordata, C. japonica Blume, C. tschonoskii Maxim., and C. laxiflora (Siebold & Zucc.) Blume); five Fagaceae species (F. crenata, F. japonica, Quercus crispula Blume var. crispula, Q. serrata, and Castanea crenata Siebold et Zucc.), one Rosaceae species (Cerasus leveilleana (Koehne) H.Ohba); three Sapindaceae species (Acer amoenum Carrière var. amoenum, A. rufinerve Siebold & Zucc., and A. pictum Thunb.); one Cornaceae species (Cornus controversa Hemsl. var. controversa); and one Styracaceae species (Styrax obassia Siebold & Zucc.). The nomenclature follows the latest edition of YList (Yonekura and Kajita 2003–).

Table 1. Focal species, traits, and estimates of soil seedbank performance, including extrapolated values for P_TOTAL and T_AVERAGE. The values in parentheses are the 95% credible intervals.

Family and latin name	Abbreviation	Seed weight (g)	Fecundity (10^3 fruits)	2-year survival of seedlings	Maximum DBH (cm)	PTOTAL			TAVERAGE (year)
						Mean	95% CI	Extrap.	Mean	95% CI	Extrap.
Betulaceae
Betula grossa Siebold & Zucc.	B.g	0.0005	977.4	0.000037	90.2	0.011	(0.003–0.021)	0.011	0.54	(0.500–0.661)	0.54
Carpinus cordata Blume var. cordata	Car.c	0.008	34.4	0.0036	45.0	0.038	(0.005–0.077)	0.038	1.56	(1.415–1.777)	1.56
Carpinus japonica Blume	Car.j	0.005	155.0	0.00056	41.5	0.043	(0.020–0.071)	0.043	0.57	(0.500–0.673)	0.56
Carpinus tschonoskii Maxim.	Car.t	0.015	330.5	0.00038	50.5	0.088	(0.017–0.178)	0.088	0.57	(0.504–0.672)	0.56
Carpinus laxiflora (Siebold & Zucc.) Blume	Car.l	0.003	337.9	0.00049	58.3	0.066	(0.020–0.119)	0.066	0.61	(0.515–0.740)	0.59
Fagaceae
Fagus crenata Blume	F.c	0.109	150.8	0.0081	101.5	0.042	(0.009–0.068)	——	0.50	——	——
Fagus japonica Maxim.	F.j	0.121	364.8	0.00082	91.8	0.009	(0.000–0.024)	——	0.50	——	——
Quercus crispula Blume var. crispula	Q.c	1.6	2.6	0.0054	100.1	0.092	(0.050–0.135)	——	0.50	——	——
Quercus serrata Murray	Q.s	0.81	10.1	0.0016	112.2	0.032	(0.004–0.069)	——	0.50	——	——
Castanea crenata Siebold et Zucc.	Cas.c	1.46	1.5	0.024	81.7	0.161	(0.047–0.316)	——	0.50	——	——
Rosaceae
Cerasus leveilleana (Koehne) H.Ohba	Cel.l	0.011	11.3	0.019	64.4	0.069	(0.000–0.156)	0.070	1.35	(0.500–2.319)	1.39
Sapindaceae
Acer amoenum Carrière var. amoenum	A.a	0.008	82.4	0.0094	61.9	0.055	(0.020–0.095)	0.055	1.36	(1.131–1.525)	1.39
Acer rufinerve Siebold & Zucc.	A.r	0.016	71.8	0.0089	29.9	0.219	(0.014–0.695)	0.219	0.95	(0.500–1.376)	0.89
Acer pictum Thunb.	A.p	0.023	389.8	0.0092	74.8	0.063	(0.025–0.107)	0.063	0.71	(0.528–0.898)	0.70
Cornaceae
Cornus controversa Hemsl. var. controversa	Cor.c	0.032	55.8	0.014	49.4	0.017	(0.007–0.027)	0.017	1.62	(0.884–2.428)	1.66
Styracaceae
Styrax obassia Siebold & Zucc.	S.o	0.12	1.8	0.0094	27.3	0.149	(0.087–0.225)	0.160	2.41	(1.961–3.000)	2.69

Sampling points were established in 10-m grids within the 1-ha/1.2-ha subplot, for a total of 121 points in 1988 and 143 points from 1989 onward. At each point, we installed a conical seed trap supported by three polyvinyl chloride pipes. Seed traps were made of 1-mm mesh polyethylene cloth with a circular opening of 0.5 m² and a depth of 1 m. A 1-m² quadrat was also established adjacent to the seed trap at each point. Seeds were collected from the traps every 2–4 weeks between April and November, starting in 1988, and viable seeds were counted and identified to species. Additionally, newly emerged seedlings in the quadrats were tagged and monitored every 2–4 weeks between April and November. Because a proportion of the germinated seeds died without being observed as emerged seedlings (i.e. germination failure), we referred to the proportion of emerged seedlings to soil seeds as “the total proportion of seeds that emerge as seedlings (P_TOTAL)”, rather than “the germination rate.” Seed rain data for 1988–1998 (11 years) and seedling data for 1989–1999 (11 years) were used for analyses (Supplemental Material 1). There were no extreme climate conditions such as severe drought or drastic environmental changes within the subplot such as large disturbances (M. Shibata, personal observation). The numbers of viable seeds and emerged seedlings were summed over all traps and quadrats, respectively, and used in the analyses.

Analyses

In the study forest, seed dispersal and seedling emergence of 16 species mainly occurred from September to November (exceptionally June to July for Cerasus) and April to May, respectively. For the sake of model simplicity, all seed dispersal was assumed to occur in October and all seedling emergence was assumed to occur in April (0.5 years after October).

A proportion (P₁) of the seeds dispersed in October of year t (where t = 0 corresponds to 1988) emerge as seedlings in May of year t +1 (0.5 year after dispersal). In May of year t +2 (1.5 years after dispersal), P₂ of the seeds dispersed in year t emerges as seedlings, and this process continues for year t +3 and beyond, until the seeds dispersed in year t are extinguished (upper part of Figure 1). Extinguished seeds include seeds that germinate but do not emerge as seedlings (failed germination), seeds that lose viability due to age (physiological death), and seeds that experience predation by animals or infection by pathogenic fungi (Baker 1989; Pickett and McDonnell 1989).

Figure 1. Conceptual diagram of the analyses. Seeds that fall in year t (a) may persist for several years before they die or until they emerge as seedlings. Seedlings in year t (b) represent the emergence of seeds dispersed in preceding years.

The number of seedlings that emerge in year t, SEEDLING_t, represents seedlings originating from seeds dispersed over several preceding years (lower part of Figure 1):

(1) $\begin{array}{c}\mathit{SEEDLIN}{G}_t=\sum _{\mathit{i}=1}^n\left(\mathit{SEE}{D}_{\mathit{t}-\mathit{i}}{P}_i\right)\end{array}$(1)

Here, SEED_t-i represents the number of seeds shed in year t-i, and n represents the number of preceding years during which the dispersed seeds persisted in the seedbank. Using the parameter P_i, P_TOTAL was derived as follows:

(2) $\begin{array}{c}{P}_{\mathrm{TOTAL}}=\sum _{\mathit{i}=1}^n{P}_i\end{array}$(2)

The average time in the soil prior to emergence (T_AVERAGE) was derived as follows:

(3) $\begin{array}{c}{T}_{\mathrm{AVERAGE}}=\sum _{\mathit{i}=1}^n\left(\left(\mathit{i}-0.5\right)P_i\right)/P_{\mathrm{TOTAL}}\end{array}$(3)

We formulated these equations as statistical models and used the data to estimate the parameter P_i and derive the indices P_TOTAL and T_AVERAGE.

We made some modifications to our basic model. First, the acorns produced by some Fagaceae species do not persist in the seedbank for more than 0.5 years, since they either germinate or die in the spring following dispersal. However, preliminary analyses indicated that when the data for Fagaceae species were simply applied to the model, the estimated parameters P_i for i ≥ 2 were slightly above zero. To avoid such unrealistic estimates, a dummy vector (β) of length n was introduced to the model:

(4) $\begin{array}{c}\mathit{SEEDLIN}{G}_t=\sum _{\mathit{i}=1}^n\left(\mathit{SEE}{D}_{\mathit{t}-\mathit{i}}{P}_i{\beta }_i\right)\end{array}$(4)

The first element of β is assigned a value of 1 for all species, whereas subsequent elements are assigned values of 0 for Fagaceae species and 1 for other taxa (β₁ = 1 and β_2–n = 0 for Fagaceae; β_1–n = 1 for others).

Furthermore, seedling emergence may vary among years due to variations in unobserved environmental conditions. To account for this variation, a random effect γ_t was introduced to the model:

(5) $\begin{array}{c}\mathit{SEEDLIN}{G}_t=exp\left({\gamma }_t\right)\sum _{\mathit{i}=1}^n\left(\mathit{SEE}{D}_{\mathit{t}-\mathit{i}}{P}_i{\beta }_i\right)\end{array}$(5)

The parameter γ_t follows a normal distribution with a mean of 0 and a variance of σ_γ².

We assumed that P_i was influenced by temporal effects. Other studies that have been conducted at the study site have reported that for some species, seedling emergence peaks in the first or second spring after seed dispersal and decreases monotonically thereafter in the absence of additional seed rain (Shibata and Nakashizuka 1995; Tanaka 1995). Therefore, we assumed that P_i < P_i-1 for i ≥3, and we introduced the parameter ρ_i, which ranges from 0 to 1; for i < 3, P_i = ρ_i and for i ≥3, P_i = ρ_i × P_i-1.

Because no a priori information was available regarding the value of n, we arbitrarily set the value at n = 5 to avoid overparameterization. To deal with the likelihood that some seeds persist in seedbank for more than 5 years, we introduced a non-negative latent variable α_t to the model except for Fagaceae species:

(6) $\begin{array}{c}\mathit{SEEDLIN}{G}_t={\alpha }_t+exp\left({\gamma }_t\right)\sum _{\mathit{i}=1}^{5-\mathit{k}}\left(\mathit{SEE}{D}_{\mathit{t}-\mathit{i}}{P}_i{\beta }_i\right)\end{array}$(6)

where k (≤4) represents the number of years for which seed dispersal data are missing; 0 for t ≥ 5 and 5–t for t ≤ 4. An alternative model is presented in Supplemental Material 2; however, it was not used due to its arbitrarily more complex assumptions and less likely estimates (see Supplemental Material 2 for details).

Finally, despite the proximity of the quadrats and seed traps, seed dispersal and seedling data were collected from slightly different locations, strictly speaking. To account for this, we added an observation process to the model. The observed numbers of dispersed seeds (SEED_obs) and emerged seedlings (SEEDLING_obs) were assumed to follow Poisson distributions with the expected values SEED_μ and SEEDLING_μ as parameters:

(7) $\begin{array}{c}\mathit{SEE}{D}_{\mathrm{obs},t}\sim \mathrm{Poisson}\left(\mathit{SEE}{D}_{\mathit{\mu },\mathit{t}}\right)\end{array}$(7)

(8) $\begin{array}{c}\mathit{SEEDLIN}{G}_{\mathrm{obs},t}\sim \mathrm{Poisson}\left(\mathit{SEEDLIN}{G}_{\mathit{\mu },\mathit{t}}\right)\end{array}$(8)

The two state variables in these equations, SEED_μ,t and SEEDLING_μ,t, are linked by the following equation:

(9) $\begin{array}{c}\mathit{SEEDLIN}{G}_{\mathit{\mu },\mathit{t}}={\alpha }_t+exp\left({\gamma }_t\right)\sum _{\mathit{i}=1}^{\mathit{n}-\mathit{k}}\left(S_{\mathit{t}-\mathit{i}}\mathit{SEE}{D}_{\mathit{\mu },\mathit{t}-\mathit{i}}{P}_i{\beta }_i\right)\end{array}$(9)

where SEED_μ was subject to estimation, and used to derive SEEDLING_μ,t. The newly added parameter S_t−i adjusts for the difference in sampling procedures between the seed and seedling-emergence datasets. The area of the sampling units differed between the seed-dispersal (0.5 m²) and seedling-emergence (1 m²) data, and the numbers of survey points differed between 1988 (121) and other years (143). Thus, S_t−i has a value of 2 × 143 ∕ 121 = 2.36 when t − i = 0, and a value of 2 in other years.

We sampled posterior distributions using the Markov chain Monte Carlo (MCMC) method in JAGS ver. 4.3.0, implemented in R ver.4.0.0 with the packages rjags ver. 4–10 and runjugs ver. 2.2.0–2. We generated 600,000-step chains, which we thinned every 500 steps after discarding the first 100,000 steps as burn-in, yielding 1,000 samples per chain. Three chains were run for each species, resulting in 3,000 posterior samples per species for each parameter and derived variable. We used vague priors: a uniform distribution from 0 to SEED_obs,t ×2 for SEED_μ,t, a gamma distribution with a shape parameter of 0.01 and a rate parameter of 100 for a reciprocal of σ_γ², a uniform distribution from 0 to SEEDLING_obs,t/2 for α_t, a normal distribution with mean of 0 and variance of 10 for the logit of ρ_t. The initial values were generated as follows: a random number between 0 and SEED_obs,t +2 for SEED_μ,t, random numbers between − 10 and − 9 for the logit of ρ_t, between 0 and 1 for the reciprocal of σ_γ², between 0 and SEEDLING_obs,t/2 for α_t, and a random number between − 1 and 1 for γ_t. Convergence was visually and statistically assessed using $\hat{R}$ (Gelman and Rubin 1992). The MCMC chains converged satisfactorily and $\hat{R}$ did not exceed 1.06. The mean values of the posterior distributions were used as estimates for each species. To check for bias in the derived P_TOTAL and T_AVERAGE based on P_1–5, we calculated them for a 100-year period using extrapolated values of P₆–P₁₀₀, by regressing log(P₃), log(P₄), and log(P₅) by 3–5 for each species.

We assessed phylogenetic effects on P_TOTAL and T_AVERAGE and the correlations among P_TOTAL, T_AVERAGE, and four traits (seed morphology, per capita fecundity, seedling banking capacity, and adult longevity). The phylogenetic tree obtained from Masaki et al. (2021) based on Phylomatic ver. 3 and the BLADJ algorithm in PYLOCOM (Webb et al. 2008) was used. The dry weight of seed kernels (i.e. embryo, endosperm, and cotyledon) was used as a measure of seed morphology (Shibata et al. 2010). Seed dry weight captures both the attractiveness of seeds to predators and the capacity of germinants to penetrate the litter layer. We used published estimates of mean fecundity per capita (Masaki et al. 2019; Wijenayake et al. 2023). Seedling survival under the canopy was used as a measure of species capacity to form seedling banks. Specifically, we assessed the 2-year survivorship of seedlings after emergence, as survivorship is extremely low during the first year after emergence due to the high vulnerability of newly emerged seedlings, but improves after the second year (Masaki and Nakashizuka 2002). Thus, the total survivorship of seedlings during the first 2 years is likely a good indicator of the capacity to form seedling banks. Values for 2-year survivorship were obtained by multiplying the annual survivorship of new seedlings (age <1 year) by the annual survivorship of older seedlings (age ≥1 year), both of which were obtained from published data (Masaki et al. 2021). We used the maximum DBH as a proxy for the longevity of adults (Osumi and Masaki 2023). DBH values were obtained from census data collected in the 6-ha plot in 1989.

In the analyses below, we often used Pearson’s parametric coefficient of correlation, which requires a normal distribution of variables. Therefore, we transformed the variables by raising each to the power of q to normalize their distributions:

(10) $\text{transformed\hspace{0.17em}}\mathit{x}=\{\begin{array}{cc}{x}^q& \left(\mathit{q}>0\right)\\ -x^q& \left(\mathit{q}<0\right)\end{array}$(10)

To seek the appropriate value of q, we transformed variables by raising them to the power of a value between − 10 and 10 in steps of 0.001. The normality of the transformed variable was assessed at each step using the Kolmogorov – Smirnov test via the ks.test function in R, and the value of q that maximized the p value was used for transformation. We also logit-transformed P_TOTAL and 2-year seedling survivorship and log-transformed other variables to assess normality; however, transformation failed to yield better results (i.e. a higher p value) for any derived indices and variables. The q-values used for transformations are summarized in Supplemental Material 3.

The effects of phylogeny on the power-transformed values of P_TOTAL and T_AVERAGE were evaluated using the phylogenetic signal K (Blomberg et al. 2003). K was calculated using the phylosig function in the R package Geiger v2.0.7 (Harmon et al. 2020). Pearson’s correlation coefficients between transformed P_TOTAL and T_AVERAGE were estimated both with (r_PIC) and without (r) phylogenetic information. The former was calculated using the pic function in the R package ape v5.4 (Paradis et al. 2019) and a phylogenetic tree obtained from Masaki et al. (2021). Because five Fagaceae species had biased T_AVERAGE values (0.5 years), we also tested the correlations with Fagaceae species excluded. Pearson’s correlation coefficients between seed performance (P_TOTAL and T_AVERAGE) and trait variables (seed dry weight, per capita fecundity, survivorship of seedlings, and maximum DBH) were estimated in the same manner.

Results

The P_TOTAL and T_AVERAGE estimates and values of specific traits are summarized in Table 1. The extrapolated values of P_TOTAL and T_AVERAGE were not substantially different from those derived from P_1–5, except for Styrax obassia, which had a slightly greater extrapolated value. Thus, we confirmed that the derived P_TOTAL and T_AVERAGE based on P_1–5 were not biased and they were used in subsequent analyses.

Estimates of P_TOTAL varied widely, from 0.009 for Fagus japonica to 0.219 for Acer rufinerve (Table 1). P_TOTAL was generally lower for species in the families Betulaceae and Fagaceae than for other taxa. Estimates of T_AVERAGE varied to a lesser degree, from 0.5 year for Fagaceae species to approximately 2.4 years for Styrax obassia (Table 1). The trends for T_AVERAGE were similar to those for P_TOTAL with respect to taxa. Phylogenetic signals were conspicuous for P_TOTAL (marginally significant) and T_AVERAGE (significant) when all 16 species were included (Table 2); however, the signal became statistically non-significant when Fagaceae species were excluded. P_TOTAL and T_AVERAGE were not significantly correlated, irrespective of the inclusion of Fagaceae or phylogeny (Figure 2, Table 3).

Figure 2. The relationship between T_AVERAGE and P_TOTAL. Full names of species are provided in Table 1.

Table 2. Values (Blomberg’s K) and significance levels (p values; in parentheses) of the phylogenetic signal in soil seedbank performance, which were power-transformed to normalize distribution.

Derived indices	All species		Without Facaceae
PTOTAL	0.339	(0.096)	0.532	(0.462)
TAVERAGE	0.383	(0.098)	0.624	(0.290)

Table 3. Coefficients of the correlations between P_TOTAL and T_AVERAGE. p values are shown in parentheses. Each index was power-transformed to normalize distribution.

Correlation coefficient	All species		Without Fagaceae
r	0.33	(0.215)	0.19	(0.581)
rPIC	−0.37	(0.179)	−0.40	(0.258)

Species-specific estimates of P₁–P₅ are shown for non-Fagaceae species in Figure 3. Betula grossa exhibited a relatively high probability of emergence (>0.001) only in the first spring after seed dispersal. The probability of emergence remained relatively high for 1.5–2.5 years for three species of Carpinus, and longer for C. cordata. Carpinus cordata exhibited a unique pattern, whereby the probability of emergence peaked at 1.5 years after seed dispersal. Among Acer species, A. rufinerve maintained a high probability of emergence throughout the study period, whereas other species maintained high probabilities for 2.5–3.5 years. The other three species maintained high probabilities of emergence until 3.5–4.5 years. Styrax obassia consistently exhibited the highest probability of emergence among the 16 species for ≥1.5 years.

Figure 3. Estimates of P₁–P₅ (0.5–4.5 years) for non-Fagaceae species. Full names of species are provided in Table 1.

The relationships between the derived estimates (P_TOTAL and T_AVERAGE) and traits (seed morphology and three life history traits) are shown in Figure 4, and the correlations between variables are summarized in Table 4. Seed weight had a non-significant positive correlation with P_TOTAL. Exclusion of Fagaceae species or inclusion of phylogeny made the relationship marginally significant. Seed weight was negatively correlated with T_AVERAGE, but this result was driven by the presence of outlier Fagaceae species; the correlation was positive when Fagaceae species were excluded. The correlation became non-significant when phylogeny was included. Overall, seed weight was not substantially correlated with P_TOTAL or T_AVERAGE.

Figure 4. Relationships between traits (seed weight, per capita fecundity, 2-year survivorship of seedlings, and maximum DBH) and the performance of seeds in the soil seedbank. Full names of species are provided in Table 1.

Table 4. Coefficients of correlations between traits (seed dry weight, per capita fecundity, 2-year seedling survivorship, and maximum DBH) and the performance of seeds in the soil seedbank. All of the derived indices and trait variables were power-transformed (supplemental material 3) to normalize their distributions. p values are shown in parentheses.

Derived indices	Correlation coefficient	Seed weight				Fecundity				Seedling survival				Maximum DBH
		All species		Without Fagaceae		All species		Without Fagaceae		All species		Without Fagaceae		All species		Without Fagaceae
PTOTAL	r	0.31	(0.249)	0.57	(0.065)	-0.51	(0.042)	-0.44	(0.171)	0.47	(0.069)	0.26	(0.437)	-0.47	(0.063)	-0.67	(0.023)
	rPIC	0.47	(0.079)	0.51	(0.130)	-0.03	(0.920)	0.28	(0.428)	0.23	(0.417)	-0.46	(0.184)	-0.19	(0.507)	-0.23	(0.527)
TAVERAGE	r	-0.36	(0.166)	0.66	(0.029)	-0.22	(0.414)	-0.85	(0.001)	0.38	(0.147)	0.87	(0.001)	-0.71	(0.002)	-0.39	(0.232)
	rPIC	-0.22	(0.424)	-0.11	(0.765)	-0.58	(0.024)	-0.75	(0.013)	0.47	(0.077)	0.86	(0.001)	-0.24	(0.385)	-0.17	(0.646)

Per capita fecundity exhibited a significant negative correlation with P_TOTAL for all species, but the correlation became non-significant following exclusion of Fagaceae species and inclusion of phylogeny. Per capita fecundity was negatively correlated with T_AVERAGE, and the significance level increased when Fagaceae species were excluded or when phylogeny was included. Overall, fecundity exhibited no clear relationship with P_TOTAL and a significant negative relationship with T_AVERAGE.

The relationship between 2-year survivorship of seedlings and P_TOTAL was non-significant following the exclusion of Fagaceae species or phylogeny. By contrast, positive relationships with T_AVERAGE were frequently significant or marginally significant when Fagaceae species were excluded or when phylogeny was included. Overall, 2-year survivorship of seedlings, a measure of species’ capacity to form a seedling bank, was positively correlated with T_AVERAGE.

Maximum DBH was negatively correlated with P_TOTAL. Exclusion of Fagaceae species did not substantially impact the results; however, inclusion of phylogeny weakened the correlation. Maximum DBH was also negatively correlated with T_AVERAGE, but exclusion of Fagaceae species and inclusion of phylogeny made the relationship insignificant. Overall, maximum DBH, an index of the longevity of adult trees, was not significantly correlated with P_TOTAL or T_AVERAGE.

Discussion

In this study, we tested two models and decided not to use one of them (see Supplemental Material 2). The estimates of T_AVERAGE were robust and roughly equivalent with the both models. For P_TOTAL, although the estimates with the model used were probable, those from the model we did not use were 1.4–3.3 times higher, probably due to several arbitrary assumptions. Caution is still needed because the P_TOTAL estimates themselves from the model used might lack robustness. Nonetheless, the estimates of P_TOTAL correlated well between the models (r = 0.952, p = 6.12 × 10⁻⁶; see Supplemental Material 2 for details), implying that the relative hierarchical relationship of P_TOTAL among species was not affected by the model assumptions and that the important results (i.e. the phylogenetic signals of P_TOTAL and T_AVERAGE, and the correlations among P_TOTAL, T_AVERAGE, seed morphology, and life history traits) should remain the same regardless of which model is chosen.

Phylogenetic signals were substantial in both P_TOTAL (marginally significant) and T_AVERAGE (significant), but became non-significant when Fagaceae species were excluded (Table 2). This implies that Fagaceae species, in which seed lifespans were fixed at 0.5 year, substantially influenced our assessment of phylogenetic effects. Correlation among seed banking capacity and trait variables was often affected by phylogeny as shown previously (e.g. Barak et al. 2018). Some apparent correlations without phylogeny were no longer significant after the inclusion of phylogeny, and some non-significant correlations became significant when phylogeny was taken into account (Table 4). Thus, phylogeny constrained interspecific variation in seed-banking capacity in different ways.

An unexpected trend in our results was the variation in P_TOTAL and T_AVERAGE among species in the same genus or family (Table 1). Carpinus cordata and Betula grossa exhibited substantially lower P_TOTAL than other species in the same family, whereas Acer rufinerve and Castanea crenata exhibited considerably higher P_TOTAL than other species in the same genus and family, respectively. Carpinus cordata and Acer amoenum exhibited substantially higher T_AVERAGE than their congenerics. Such variations among closely related species are expected to render the phylogenetic signals non-significant, implying that the performance of seeds in the soil seedbank is not tightly constrained by phylogeny.

Our derived values for T_AVERAGE indicate that some species, including Styrax obassia, Cerasus leveilleana, Cornus controversa, Carpinus cordata, and Acer rufinerve exhibited higher seed-banking capacity; seeds of these species persisted for an average of approximately 1.5 years or more in the soil (Table 1 and Figure 2). A separate study noted that the seeds of S. obassia can remain viable for more than 10 years, far longer than our estimate, when buried deep in the soil (Ozawa 1950). It is likely that this discrepancy is an artefact resulting from differences in evaluation methods; we estimated average persistence, whereas the previous study assessed potentially maximum persistence. It is also possible that the results of this study reflect high seed mortality near the soil surface, as reported in a study of Cornus controversa at the same study site (Masaki et al. 1998). Predation during the post-dispersal phases undoubtedly affects seed longevity (Baker 1989; Pickett and McDonnell 1989) and could result in seed banks in these species being transient.

With respect to the species-specific values of P_TOTAL, Styrax obassia and Acer rufinerve had higher values than other non-Fagaceae species (Table 1), which implies that these species likely survived better than other species for a short period (within a few years) after dispersal. No study has indicated that the seeds of these species are protected by either physical (i.e. hard seed coat) or chemical (i.e. toxins) mechanisms, and it is unclear what mechanisms promote the persistence of these seeds in the soil seedbank. While the acorns of Quercus species contain a toxic compound (tannic acid; Shimada and Saitoh 2003) that is considered protective against herbivory, the seeds of Fagus and Castanea do not contain this compound. Despite this, Quercus did not exhibit conspicuously higher P_TOTAL than Fagus or Castanea. The stable annual seed production of Quercus (Supplemental Material 1) might both preclude predator (mainly wood mice) satiation and promote acclimatization to the toxin (Shimada et al. 2006).

T_AVERAGE was not negatively correlated with P_TOTAL; rather, we found no significant correlation between the two indices (Figure 2 and Table 3). Large variation within the same taxon may have hindered the detection of trade-offs between the capacity to persist in the soil seedbank and the capacity to evade mortality due to natural enemies. Based on the results of this study, a longer stay in the soil did not necessarily entail restrained seed germination and growing accumulation of seeds in the soil.

There were no significant correlations between P_TOTAL or T_AVERAGE and seed morphology (i.e. seed weight) (Table 4). Our T_AVERAGE results did not conform to empirical studies that have reported that smaller seeds are frequently stored in soil seedbanks (Jaganathan et al. 2015), whereas larger seeds are not. T_AVERAGE varied less than P_TOTAL (Table 1), obscuring the relationship between T_AVERAGE and seed morphology. The P_TOTAL results also differed from those of empirical studies that reported that larger seeds are more attractive to herbivores. This discrepancy may be driven more strongly by the performance of small seeds than by that of larger seeds. Abundant autumn leaf fall in the deciduous forest at our study site has led to the formation of a thick litter layer, which strongly inhibits seedling emergence (Masaki et al. 2007; Shibata et al. 2010). Thus, species-specific seed weight is not associated with the capacity of a species to form soil seedbanks in temperate deciduous forests.

Overall, we found that some life-history traits can compensate for shorter T_AVERAGE and lower P_TOTAL (Table 4). Per capita fecundity was negatively correlated with T_AVERAGE, implying that abundant seed rain might compensate for a low seed-banking capacity. For example, seeds of Betula grossa, the most fecund species in our study, were dispersed throughout the study stand, likely increasing the probability of encountering safe sites (Nakashizuka et al. 1995). Maximum DBH was often substantially negatively correlated with both T_AVERAGE and P_TOTAL; however, this trend disappeared when phylogeny was taken into account, which implies that evolutionarily acquired adult longevity did not compensate for reduced performance of seeds in the soil seedbank. Contrary to our expectations, 2-year seedling survivorship was positively correlated with T_AVERAGE. In theory, light-demanding, shade-intolerant species have a limited capacity to form a seedling bank in closed forests, and should instead bank their seeds in the soil to enable seedlings to exploit canopy gaps when they appear. However, this apparently plausible trade-off was not present in our study stand; rather, the significant, positive correlation between 2-year survivorship of seedlings and T_AVERAGE implies a synergistic relationship. This result implies that there is not necessarily a trade-off between seed-banking capacity and seedling-banking capacity in temperate forests. It is also possible that a low banking ability for shade-intolerant species can be compensated by other traits such as high dispersal ability. Further studies are required to determine whether the observed relationship is universal.

Our results imply that the seed-banking capacity of the tree species at our study site is generally weak. This supports findings from studies of other old-growth forests that implied that the soil seedbank often comprises forest generalist species and differs compositionally from standing trees, which are mostly forest specialists, and that soil seedbanks do not contribute to the in-situ regeneration of canopy trees (Naka and Yoda 1984; Gasperini et al. 2022). Late-successional species, which are the primary components of old-growth forests, are thought to have a relatively low capacity to bank their seeds in the soil (Pickett and McDonnell 1989). Most of our focal species had small T_AVERAGE values, highlighting the limited role of soil seedbanks in old-growth forests.

Despite this, we conclude that soil seedbanks may be an effective regeneration strategy for canopy trees under certain circumstances. The T_AVERAGE values of Carpinus cordata, Cerasus leveilleana, Acer amoenum, Cornus controversa and Styrax obassia were >1.3 years, implying that banked seeds may compensate for years with little to no seed production. By contrast, the very low T_AVERAGE of Quercus species (0.5 years) is unlikely to be disadvantageous, because these species produce seeds almost every year (Supplemental Material 1). These instances are examples of how soil seedbanks may contribute to natural regeneration processes in forests. However, if adult trees are lost due to harvesting or natural disturbance, the disadvantages of seed banking may quickly become apparent, as some species may fail to establish offspring due to a lack of seeds. Thus, in the context of timber harvesting, seed banking may not be useful. In this context, seed banking is a reliable strategy for forest regeneration in a conservation context, but not for sustainable forestry operations. However, if a species has very high fecundity and a small portion of their dispersed seeds escape high post-dispersal mortality and persist in the soil for a long time, then soil seeds may contribute to regeneration after large-scale disturbances, even with small P_TOTAL and T_AVERAGE. To confirm this hypothesis, the annual survivorship and germination ratio of soil seeds, which are components of P_TOTAL and T_AVERAGE, must be assessed separately over a long period. Such studies will provide a more comprehensive understanding of the importance of seed banking for tree species in temperate forests.

Data deposition

No applicable data.

Geolocation information

N 36°56′, E 140°35′, 610 m above sea level

Acknowledgements

The authors thank Dr. Hiroko Kurokawa for her support.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Some of the data used in this study are openly available in Dryad at https://doi.org/10.5061/dryad.mpg4f4r05

Supplementary data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/13416979.2024.2327979

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Additional information

Funding

This work was supported by the JSPS KAKENHI under Grants 21H04946.