Strict mast fruiting for a tropical dipterocarp tree: a demographic cost–benefit analysis of delayed reproduction and seed predation

Authors

  • Marco D. Visser,

    Corresponding author
    1. Van Hall Larenstein, Wageningen University and Research Centre, Larensteinselaan 26a, 6880 GB Velp, The Netherlands
    2. Department of Experimental Plant Ecology, Institute for Water and Wetland Research, Radboud University Nijmegen, Heyendaalseweg 135, 6525 AJ Nijmegen, The Netherlands
    Search for more papers by this author
  • Eelke Jongejans,

    1. Department of Experimental Plant Ecology, Institute for Water and Wetland Research, Radboud University Nijmegen, Heyendaalseweg 135, 6525 AJ Nijmegen, The Netherlands
    Search for more papers by this author
  • Michiel van Breugel,

    1. Van Hall Larenstein, Wageningen University and Research Centre, Larensteinselaan 26a, 6880 GB Velp, The Netherlands
    2. Centre for Tropical Forest Science, Smithsonian Tropical Research Institute, Av. Roosevelt 401, Balboa, Ancón, Panamá, Republic of Panamá
    Search for more papers by this author
  • Pieter A. Zuidema,

    1. Institute of Environmental Biology, Ecology and Biodiversity, Utrecht University, Padualaan 8, 3584 CH Utrecht, The Netherlands
    Search for more papers by this author
  • Yu-Yun Chen,

    1. Department of Natural Resources and Environmental Studies, National Donghwa University, Hualien, 974 Taiwan
    Search for more papers by this author
  • Abdul Rahman Kassim,

    1. Natural Forest Division, Forest Research Institute Malaysia, Kepong, Kuala Lumpur, Malaysia
    Search for more papers by this author
  • Hans de Kroon

    1. Department of Experimental Plant Ecology, Institute for Water and Wetland Research, Radboud University Nijmegen, Heyendaalseweg 135, 6525 AJ Nijmegen, The Netherlands
    Search for more papers by this author

  • Present address: Forest Ecology and Management, Centre for Ecosystem Studies, Wageningen University, Box 47, 6700 AA Wageningen, The Netherlands.

Correspondence author. E-mail: m.visser@science.ru.nl

Summary

1. Masting, the production of large seed crops at intervals of several years, is a reproductive adaptation displayed by many tree species. The predator satiation hypothesis predicts that starvation of seed predators between mast years and satiation during mast years decreases seed predation and thus enhances tree regeneration.

2. Mast fruiting comes at demographic costs such as missed reproduction opportunities and increased density-dependence of recruits, but it remains unknown if predator satiation constitutes a sufficiently large benefit for masting to evolve as a viable life-history strategy. So far, no studies have quantified the net fitness consequences of masting.

3. Using a long-term demographic data set of the dipterocarp Shorea leprosula in a Malaysian forest, we constructed stochastic matrix population models and performed a demographic cost–benefit analysis.

4. For observed values of mast frequency and seed predation rates, we show that strict masting strongly increases fitness compared with fruiting annually. Model results also show that the demographic costs of mast fruiting are very low compared to the demographic losses due to seed predation in a scenario of annual fruiting. Finally, we find that mast fruiting would still be selected for even at low levels of seed predation and when including additional costs such as decreased adult growth rates, limiting crop size and density-dependent seedling survival.

5.Synthesis. Our results are consistent with the predictions of the predator satiation hypothesis: mast fruiting increases fitness for a range of seed predation levels. Under seed predation pressure annually fruiting species are at a strong disadvantage and as a result a mast fruiting strategy may swiftly confer a fitness advantage. Our study shows that demographic modelling allows the weighing of fitness benefits and costs of life-history phenomena such as strict masting.

Introduction

Mast fruiting, or the synchronous variable production of large fruit crops (‘masts’) among years of smaller crops, is a widespread phenomenon occurring throughout the world (Sork, Bramble & Sexton 1993; Shibata, Tanaka & Nakashizuka 1998; Rees, Kelly & Bjørnstad 2002; Övergaard, Gemmel & Karlsson 2007) known to strongly impact the dynamics of both plants (Crone & Lesica 2004) and animals (Jones et al. 1998). This phenomenon has intrigued ecologists for a long time and a number of hypotheses have been proposed to explain the evolution of mast fruiting. These hypotheses start with the notion that irregular fruiting may increase pollination efficiency, seed survival and/or seedling establishment (Kelly 1994; Kelly & Sork 2002).

In large parts of tropical Southeast Asia, hundreds of tree species from dozens of families display a distinct form of masting termed ‘strict masting’ (Kelly 1994): the synchronous production of large fruit crops after years of hardly any seed production (Wich & Van Schaik 2000; Cannon et al. 2007) resulting in clear separation between mast and non-mast years (unlike more gradual forms of masting displayed elsewhere; Kelly 1994). In Southeast Asia, mast years occur unpredictably at intervals of 2–10 years (Appanah 1985; Ashton, Givnish & Appanah 1988). These masting events involve entire tree communities, extend over huge areas (Appanah 1985; Curran et al. 1999; Numata, Yasuda & Okuda 2003) and swamp forests with seeds and seedlings (Fox 1972; Curran & Leighton 2000).

The most widely accepted explanation of masting, and of strict masting in Southeast Asia in particular, is the predator satiation hypothesis (PSH; see Ashton 1988; Ashton, Givnish & Appanah 1988; Sun et al. 2007) originally proposed by Janzen (1974). This hypothesis predicts that mast fruiting enhances seed survival because: (i) synchronous production of large seed crops causes predator satiation and (ii) fruiting at long multiyear intervals prevents build-up of seed predator populations through starvation. Thus, mast fruiting is expected to decrease the destructive impact of seed predators (see also Holling 1959).

However, mast fruiting comes at a cost. Compared with tree species that fruit annually, mast-fruiting species have fewer reproductive opportunities and a higher risk of death of mature individuals between reproductive events (Bull & Shine 1979; Waller 1979). All else being equal, these costs would reduce population fitness of mast-fruiting species. Other noteworthy costs due to masting include increased density-dependent mortality (Hett 1971) or decreased diameter growth rates (Eis, Garman & Ebell 1965; Norton & Kelly 1988), which further reduce population fitness of masting species. To compensate for these costs, mast-fruiting species need to produce more seeds per reproductive event, which exhibit a higher per capita survival chance. Thus, masting can only be a viable strategy when the benefits more than compensate for the costs associated with delayed reproduction.

A full assessment of the fitness consequences of masting requires a cost–benefit analysis at the level of the entire life cycle, using demographic analyses based on long-term data (Kelly & Sork 2002). So far, such a full demographic cost–benefit evaluation is lacking for mast fruiting tree species. Empirical field studies have only addressed benefits to early juvenile stages such as increased seed survival during mast years (Curran & Leighton 2000; Curran & Webb 2000; Maycock et al. 2005; Sun et al. 2007). Evolutionary studies only used certain fitness components such as increased survival or reproductive success (Silvertown 1980; Sork 1993; Sork, Bramble & Sexton 1993; Kelly & Sullivan 1997; Herrera et al. 1998; Kerkhoff 2004). Existing demographic models of Southeast Asian species only included larger individuals (>10 cm d.b.h.; e.g. Boscolo & Vincent 1998; Sist, Picard & Gourlet-Fleury 2003), ignoring mast-fruiting events, leaving out the crucial early reproductive phases and not accounting for costs associated with masting. Thus, there is a clear need to use demographic data on the entire life cycle to evaluate the demographic costs and benefits of masting.

Here, we construct stochastic matrix models for the Southeast Asian tree Shorea leprosula using life-history data enabling integration over the entire life cycle of a masting species. We use these models to evaluate whether, at our study site, the demographic benefits of strict masting outweigh the costs of missed reproduction opportunities at the current level of seed predation and mast frequency. Secondly, we investigate the conditions under which mast fruiting is more advantageous than annual fruiting, testing the hypothesis that masting only results in higher fitness if seed predation intensities are high (Silvertown 1980). By quantifying costs and benefits of mast fruiting and evaluating fitness consequences, we provide insights into the conditions under which strict masting can be a viable life-history strategy for tree species in Southeast Asian forests.

Materials and methods

Study species

Shorea leprosula Miq. (nomenclature follows Ashton 1982) is a dominant, emergent (often more than 50 m in height) tree species of the Dipterocarpaceae family. It is common in the lowland forest of West Malaysia and locally abundant (up to 4% of stems >30 cm d.b.h. in our study site). The species flowers sporadically throughout the year and gregariously at longer intervals (Soerianegara & Lemmens 1993). Flowers are hermaphroditic and pollinated by thrips (Appanah & Chan 1981) and other insect species (Sakai et al. 1999). Seeds are dispersed primarily by gyration (Ashton 1988). Shorea leprosula and other dipterocarps are known to endure intense predation (Nakagawa et al. 2005) from a wide range of vertebrate and invertebrate seed predators (Curran & Leighton 2000; Sun et al. 2007). Shorea leprosula is locally known as ‘Meranti Tembaga’; it belongs to the commercially popular red meranti group (Newman, Burgess & Whitmore 1996) yielding light hardwoods (415–685 kg m−3; Choo & Lim 1983).

Study site

Demographic data for S. leprosula were derived from the 50-ha permanent forest dynamics plot (FDP) in Pasoh Forest Reserve, Negri Sembilan, Peninsular Malaysia (2°58′ N, 102°18′ E). The 1000 × 500 m study plot is located in the centre of 600 ha of primary lowland dipterocarp forest (i.e. dominated by trees of the Dipterocarpaceae family). Within the plot all stems >1 cm d.b.h. have been tagged, measured, mapped and identified (Manokaran et al. 1990), including approximately 3000 S. leprosula trees. Since the establishment in 1986 there have been three recensuses at approximately 5-year intervals (1990, 1995–1996 and 2000–2001). See Condit (1995) and Manokaran et al. (2004) for more information on Pasoh Forest Reserve and the FDP. Seedling recruitment, height growth and mortality have been monitored annually in 1024 one-m2 plots inside the Pasoh FDP, established between August 2001 and May 2002. Recruitment of approximately 5000 seedlings was recorded in censuses carried out directly after the 2005 mast fruiting event. During the 2005, mast fruiting event a whole-plot survey of the reproductive status of all dipterocarp trees ≥30 cm d.b.h. was conducted. Additionally, pre- and post-dispersal predation rates were quantified for S. leprosula in 55-m2 quadrates in which the fate of seeds and diaspores were identified (see Sun et al. 2007 for methodological details).

Modelling population dynamics

The population was divided in seven size-based classes ranging from <1 to >70 cm d.b.h. (Table 1). Class borders were based on the grouping of individuals with (statistically) similar demographic rates, ensuring a minimum sample size of 40 per class (see Appendix S1 in Supporting information). A seed stage was not included as dipterocarp seeds usually germinate within a few days and lack dormancy (Burgess 1972; Fox 1976; Raich & Khoon 1990).

Table 1.   Stage classification for Shorea leprosula, sample size (N) over all census years, participation or the proportion of individuals reproducing (2005), the average amount of germinated seedlings per adult (2005), mean growth rates and annual survival rates (seedling first-year survival was estimated to be 0.31). The values in this table are averages (the transition matrices are presented separately in Table S2)
Stage classSize (mm)Sample sizeParticipationFecundity (n adult−1)*Growth (mm year−1)Survival
  1. *Expected 2005 mast crop size at zero seed predation (τ = 0).

  2. †Height growth in mm year−1.

1<9.95076003.42†0.67
29.9–29.91906001.310.87
329.9–49.9325002.800.93
449.9–149.9496004.360.96
5149.9–299.9118007.330.98
6299.9–699.91280.78442.76.690.98
7>699.9460.782623.44.800.98

To model the size and structure of populations through time, we used stage-classified matrices (Lefkovitch 1965; Caswell 2001). In general, these models have the form n(t+1) = Atn(t), where n(t) and n(t+1) are vectors containing the population structure, At is an m × m, non-negative matrix with elements that represent transition rates among m life-cycle stages. In the stage-classified model, matrix elements are built from the vital rates growth (γ), survival (σ) and fecundity (F), which are described in detail below and summarized in Table 1.

Survival and growth

The vital rates survival (σ) and growth of survivors (γ) were estimated for individuals ≥1 cm d.b.h. based on data from 1986 to 2001. Rates for individuals <1 cm d.b.h. (the first stage class) were derived from census data following a cohort of seedlings up to 2 years after the 2005 mast event (three census points). Within each stage class, annual growth rates were estimated by the slope of a linear regression of d.b.h. or height against time in years (γj; see Appendix S1). Size-specific survival rates (σj; individuals ≥1 cm d.b.h.) were estimated following a model selection procedure (sensuHuisman, Olff & Fresco 1993; see Appendix S1) using logistic regression. Annual survival rates were derived from the multiple-year census interval by taking the t-root of the survival rate over the t-years of the census interval. Survival rates for individuals <1 cm d.b.h. were calculated as the proportion of seedlings surviving their second year. The height at which seedlings are expected to reach the second stage class (>1 cm d.b.h.) was estimated using allometric relation of height against d.b.h. (Michaelis–Menten model; R2 = 0.96, n = 33; 1 cm d.b.h. corresponded to 1.73 m height). From these vital rates, matrix elements were calculated as follows: progression (i.e. the subdiagonal of the matrix) is given by the probability of surviving and growing to the next stage class: Gij = σj × γj × c−1, where c is the stage class width (see Table 1). Stasis (i.e. the diagonal of the matrix) is defined by the probability of surviving and remaining in the same stage class: Pij = σjGij. Parameter definitions are included in Table 2 and an example of how the matrix elements are arranged can be found in Table S1 (Supporting information).

Table 2.   Model parameters, definitions and estimation methods
ParameterDefinitionEstimation method
πThe annual fruiting frequency (or the probability of a mast)Varied between 1 and 0 with 50 equally spaced steps in the model
μYears since previous fruitingE(μ) = π−1
τThe intensity of seed predation expressed as a factor of ϕVaried between 0 and 6 with 50 equally spaced steps in the model
κThe total pre- and post dispersal predation rate at Pasoh forest reserve during the 2005 mast eventProportion of Shorea leprosula seeds destroyed by seed predators (details in Sun et al. 2007)
FjThe amount of seedlings that survive seed predation for each tree sized jEquation 2
GijProgression, the probability of growing to the next class
 Gij = σj × γj × c−1; where c is the stage class width (Table 1)
PijStasis, probability of remaining in the same classPij = σjGij
γjSize specific growth rate (mm year−1)See online appendix S1
σjSize specific survival rateSee online appendix S1
σsSeedling first-year survival rateProportion of seedlings surviving their first year
ϕThe annual averaged number of seedlings produced per tree, in the absence of seed predationEquation 1
SAverage amount of newly germinating seedlings (see comment above)The average seedling density multiplied by the crown area of a tree size j
υjProportion of individuals size j that participate in reproductionEstimated from the 2005 plot-wide flowering survey

Crop size and predation

We used information on seedling recruitment during a masting year to estimate the hypothetical annual reproductive output of our study. We did so by assuming that crop size is directly related to the amount of accumulated reserves in a tree and that the amount of additional reserves that is stored by mature individuals does not vary over time. This is supported by the notion that reproductive output in both (annually fruiting) perennials and mast fruiting species (including dipterocarps) is strongly dependent on the stored reserves (Chapin, Schulze & Mooney 1990; Ichie et al. 2005; Yasumura, Hikosaka & Hiroseb 2006). Thus, if a tree is capable of producing ϕ seedlings annually (with ϕ representing the number of seedlings that is produced by a tree that fruits annually in absence of seed predation), postponing reproduction to the μ-th year after the previous fruiting will allow the accumulation of enough assimilates to reproduce with size μϕ at the next reproductive opportunity. It follows that a mast-fruiting tree producing μϕ seedlings every μ years has an annual reproductive potential of ϕ seedlings. We estimated ϕ, from the 2005 mast event, using data on seedling recruitment, seed predation, amount of flowering individuals and the estimated mast frequency.

The annual fruiting frequency (or probability of a mast; πPasoh) specific for Pasoh forest (Western Peninsular Malaysia) is 0.16 (which was determined with data from 38 years of monitoring; Wich & Van Schaik 2000). The proportion of reproducing individuals (proportion of individuals that flowered; υ = 0.78) was unrelated to tree size (logistic regression, P = 0.154, individuals >30 cm d.b.h.). The seed-survival rate from both pre- and post-dispersal seed predators during the 2005 mast event (κ) was 56% (see also Sun et al. 2007). Assuming that the vast majority of seedlings germinate directly under the tree crown (Ashton 1988), the number of germinating seedlings per adult (Sj) was estimated as the average seedling density multiplied by the expected crown area of a tree of size j. The relation between crown area and tree size (d.b.h.) was estimated through a linear regression (R2 = 0.76, P = 0.0011) for 12 individuals of ≥30 cm d.b.h. measured in the field. Seedling first-year-survival σs was calculated as the proportion of surviving individuals 1 year after germination. We estimated ϕ from the 2005 mast fruiting event, where seed predators showed signs of satiation (Sun et al. 2007), using (see also Table 2 for definitions):

image(eqn 1)

where ϕj is the mean number of 1-year-old seedlings produced by an annually fruiting j-sized tree in the absence of seed predation and πPasoh is the fruiting frequency. The reproductive threshold is assumed to be 30 cm d.b.h. as there are no observations of reproducing individuals of smaller size. Equation 1 transforms the reproductive rates obtained from the 2005 mast event into annual rates and adds to this the seedlings that were lost due to seed predation. Here, we do assume that the probability of germination and seedling survival is unaffected by seed predators.

In the model, we defined a seed predation coefficient τ, which represents the fraction of ϕ that seed predators are capable of destroying. When including seed predation, the number of surviving offspring for any level of τ can then be calculated as:

image(eqn 2)

where F1j is the number of 1-year-old seedlings produced by a tree in size category j (first-row fecundity elements of the transition matrix) when τϕ potential seedlings are lost due to seed predation and μ is the time since previous fruiting, which is expected to be E(μ) = πPasoh−1. Figure 1 shows the relationship for F1j for different combinations for τ and π. The function is bounded by 0, as negative fecundity is not allowed. Within the above model, we compare fruiting strategies at equal reproductive allocation between annually and mast fruiting strategies, that is, the average number of seedlings produced over time is independent of masting frequency.

Figure 1.

 (a) The amount of surviving offspring in the model (Fj) against μ (time since previous fruiting), when ϕ (number of seedlings expected to survive at zero predation) equals 100, at different levels of τ (seed predation coefficient). The grey horizontal line indicates the annual average ϕ. (b) Amount of offspring lost due to pre- and post-dispersal seed predation plotted against the crop size per tree; predators are assumed to be satiated when μϕ ≥ τϕ. All offspring beyond this point survives.

Simulations

Three annual-transition matrices were constructed, one for each 5-year census period between 1986 and 2001. We thus assumed that conditions varied more between those 5-year periods than within. Still, temporal variation might have been underestimated, also because we had seedling data after a single mast event only. In our stochastic simulations, the three transition matrices were varied at random with equal probability. Fruiting years occurred in the main simulations with probability π = 0.16, simulating the erratic nature of masting events in West-Peninsular Malaysia (Ashton 1988; Wich & Van Schaik 2000). Note that mast fruiting can take place in two consecutive years (Numata, Yasuda & Okuda 2003; Övergaard, Gemmel & Karlsson 2007). In the model, fecundity (F1j) was either set to 0 in non-fruiting years or calculated from equation 2 in a mast year (i.e. the size of the seed crop depended on μ, the number of years since the previous mast). Population growth as simulated with the model results in a stochastic population growth rate (λs). We estimated λs by calculating the long-term growth rate (Caswell 2001; Tuljapurkar, Horvitz & Pascarella 2003):

image(eqn 3)

where T is a large number of simulation years, NT and Np are the population at time T and p respectively. To optimize computing efficiency, T was set to a value between 5000 and 10 000 depending on the mast frequency within a single simulation. The cut-off point p was set to 200, based on the dampening ratio of the mean matrix (1.07; see Caswell 2001; Grant & Benton 1996), to correct for any transient effects (Caswell 2001).

We use λs as a measure of fitness (Charlesworth 1980; de Jong 1994; Caswell 2001; Metcalf & Pavard 2007), which can be viewed as the average population fitness (Fisher 1958). Increases or decreases in fitness between masting (inline image) and annual reproduction (inline image) were defined as Δλ (=inline image), calculated at a given level of τ. A positive value of Δλ then suggests that selection favours mast reproduction.

We first compared the fitness of the annual reproduction (π = 1) and the current mast reproduction strategy (π = 0.16) at different levels of predation. We then compared the fitness of the population at the current seed predation level at Pasoh forest with a scenario of zero seed predation at differing levels of mast frequency. Here, the current expected seed predation level (τ) at Pasoh forest was calculated as E(τ) = (1−κ)/π; (0.44/0.16 = 2.75). Seed predators at Pasoh forest are therefore estimated to be capable of destroying 2.75ϕ (i.e. 275% of the potential annual seed crop). We also evaluated the required increase in F to keep λs = 1 for both annually and mast-fruiting strategies at increasing predation levels.

We subsequently evaluated the combined effects of predation and irregular reproduction for a range of different combinations of π (mast frequency) and τ (predation level) on λs by calculating fitness landscapes. This was performed through calculation of λs for all π between regular fruiting (π = 1) and π = 0 with 50 equally spaced steps. The predation factor τ was varied between 0 and 6 (with six equalling more than twice the current predation levels), again with 50 equally spaced steps. This resulted in a total of 2500 (50 × 50) iterations in π and τ. All estimates of λs were then calculated as mean values after 15 simulations as exploratory work showed that this was sufficient for credible estimates of λs. These 15 simulations were also used to derive standard deviations for the estimates of λs. The calculated fitness landscapes then allowed evaluation of selection towards or against reproducing annually by calculating Δλ landscapes (hereafter ‘selection landscapes’).

Elasticity analysis

To quantify how much each matrix element contributed to the stochastic population growth rates (with π = 0.16) and (π = 1), we performed a stochastic elasticity analysis (Tuljapurkar, Horvitz & Pascarella 2003). The stochastic elasticity values of all elements of a matrix model add up to 1, allowing for direct comparison of the relative importance of these matrix elements for inline image and inline image.

Robustness analysis

In the previous sections, we evaluated a simplified model of mast fruiting where only the cost of postponing reproduction was included (i.e. loss of reproductive opportunities and death of mature individuals between masts). In this section, we evaluate how robust a masting strategy is when including further plausible and observed costs of and limitations to masting. We therefore calculated fitness landscapes for an additional four different scenarios:

(A) Fruiting might cost more than just stored assimilates. Large crop sizes are thought to occur at the cost of decreased growth rate of adult individuals (Eis, Garman & Ebell 1965), a phenomenon not uncommon for mast-fruiting species in general (Norton & Kelly 1988) with evidence of dramatic growth arrestment in dipterocarps (Primack & Chai 1988 cited in Ashton 1988). In the first scenario, we therefore set Gij (progression) matrix elements to 0 in mast years and Pij (stasis) elements equal to σj for adult individuals.

(B) Accumulation of resources in the simplified model is a function of the time since the previous fruiting (μ). However, this implies that trees have an unlimited storage capacity, which seems unlikely (see also Isagi et al. 1997; Rees, Kelly & Bjørnstad 2002; Satake 2004). Rather, there may be a maximum value of accumulation. For the second scenario, we assumed that the current average mast return time [π−1 (0.16−1) ≈ 6 years] is related to a physical limitation of tree storage potential and that this average value is the best available estimate of a maximum storage value. Therefore, no trees accumulate resources beyond a 6-year limit in the second scenario and then await a fruiting year. Fruiting before reaching this maximum storage level does remain possible. Note that here the reproductive allocation (average number of seedlings produced over time) differs between annually and mast-fruiting strategies.

(C) The production of large seed crops is expected to induce higher density-dependent mortality of seedlings (Harms et al. 2000). We estimated the strength of density dependence on seedling survival by examining the relationship between the density (D0) of germinated seedlings directly after the 2005 mast event to their density (D1) 1 year later for all 1024 seedling plots at Pasoh forest. A linear regression was fitted to the relationship of the log-transformed values of D0 and D1. A fitted slope takes values smaller than one if the per-seed probability of survival is negatively related to seed density (see also Harms et al. 2000). Results showed that the slope was significantly smaller than one, indicating substantial density-dependent mortality for Shorea leprosula seedlings at Pasoh forest (slope values = 0.84, R2 = 0.65, P < 0.001). In the third scenario, we incorporated this observed density-dependent relationship by making seedling first-year survival (σs) a function of the average seedling density directly after a mast. Note that the incorporation of this type of density-dependence in our model reduces population growth but does not lead to a constrained population growth rate (λ = 1) with time, as the above evaluated density-dependence operates only on σs, seedling first-year survival (directly after mast years).

(D) In the final scenario, we included all costs and limitation included in scenarios A, B and C.

Results

Costs of delayed reproduction and seed predation

The highest stochastic population growth rate (mean ± SD from 15 simulations; inline image = 1.039 ± 4e−4) was found for an annually fruiting strategy (π = 1) in combination with no seed predation (τ = 0) (Fig. 2a, b). When only evaluating the masting costs of missed reproductive opportunities (by varying π) while the population experiences no seed predation (τ = 0), inline imageremains remarkably stable along a gradient of increasingly irregular reproduction (Fig. 2a, solid line). This indicates that the costs of delayed reproduction are very small for masting frequencies up to the current masting strategy at Pasoh forest. Along a gradient of decreasing mast frequency, inline imageonly drops sharply when masting becomes extreme (mast frequency π < 0.05), showing that delayed reproduction costs only become noticeable beyond the current frequency at Pasoh forest (πPasoh = 0.16).

Figure 2.

 (a) The population growth rate (λs) against decreasing fruiting frequency (π) for a scenario of zero predation (τ = 0; solid line) and at the current estimated predation level (τ = 3.5; dashed line). (b) The population growth rate (λs) against increasing levels of predation (τ) for annual reproduction (π = 1; solid line) and that of delayed reproduction at the current estimated fruiting frequency (π = 0.16; dashed line) (c) reproductive effort, as a percentage of the current fecundity (F) required to maintain a stable population (λs = 1). Here, the slope of the increase in F over predation is ±50 times greater for an annually fruiting strategy than for the current mast fruiting strategy at Pasoh forest. The vertical line in a–c indicates the estimated current levels of τ and π, respectively, while the grey area includes values within the 95% confidence intervals for these estimates. The grey horizontal line in a and b indicates the position at which λ = 1 or the point of a stable population with zero growth and the current ϕ (100%) in c.

In contrast, the inline imageof an annually fruiting strategy drops rapidly along a gradient of low to high incidence of seed predation (τ = 0–6; Fig. 2b), passing the stable population point (λs = 1) at τ = 0.9 (>90% of annual crop lost). This shows the strong negative impact of seed predation in absence of a masting strategy. After the point of total recruitment failure, τ = 1, inline imagestabilizes, reaching its lowest attainable value (equal to the survival of the largest stage class; 0.979). Figure 2b also shows that selection in favour of masting is estimated to occur already at low-predation intensity (τ > 0.1). When comparing the costs of delayed reproduction (Fig. 2a) to that of seed predation (Fig. 2b), it becomes clear that delayed reproduction costs for S. leprosula are very small compared to the costs of seed predation when fruiting annually (Fig. 2b, solid line). This point is made evident when comparing the (hypothetical) required increase in fecundity (as percentage of ϕ) to maintain a stable population (λs = 1; Fig. 2c). Annually-fruiting species will require drastic increases in fecundity when compared with mast fruiting species (Fig. 2c). Thus, demographic costs of seed predation reduce fitness more strongly than those of delayed reproduction alone.

Mast fruiting versus annual fruiting

The estimated inline image (mean ± SD from 15 simulations; 1.031 ± 7e−4) of the current masting strategy (π = 0.16) of the S. leprosula population at Pasoh forest at current seed predation levels (τ = 2.75) was considerably higher than that of an annually fruiting strategy at equal predation levels (inline image = 0.979 ± 8e−5, Fig. 2b; which is the rate of adult survival).

At the current predation intensity (τ = 2.75), more regular reproductive strategies (with π close to 1) will achieve the lowest λs (<1), indicating declining population sizes (Fig. 2a, dashed line). A stable population growth rate is only attained with an increasing inter-mast period, i.e. π < 0.6. For the current masting strategy at Pasoh forest (π = 0.16), inline imagedeclines more slowly with predation intensity than inline imageof an annually fruiting strategy (Fig. 2b), only dropping below λs = 1 at τ > 12.5 (i.e. when 1250% of the annual seed crop is destroyed). At such high-seed predation intensity, trees have a 10% chance per masting event to have accumulated resources over a sufficiently long period (μ ≥ 13 years) to produce a seed crop that satiates predators. Stochastic elasticity analysis revealed only a very small shift in the elasticity patterns between both fruiting strategies (Table S3 and Fig. S1 Supporting information).

Fitness and selection landscapes

As shown in Fig. 2, the fitness landscapes in Fig. 3 also showed largest λs for an annually fruiting strategy for a scenario with zero predation (bottom left corner in Fig. 3a). However, λs declined rapidly with predation at low values of π (<0.5) and much more slowly at higher values of π (>0.5). In cases, where seed predators are able to destroy the entire annual seed crop (τ > 1), masting is the only strategy yielding a population increase (λs ≥ 1). With intensifying predation, a stable population growth rate (λs = 1) could only be attained with increasingly long inter-mast periods (increasingly lower π; Fig. 3a).

Figure 3.

 (a) Contour plots showing the joint relationship of seed predation (τ) and delayed reproduction (π) on the population fitness (λs). (b) The strength of selection (Δλ) towards delayed reproduction. Both graphs are the result of 2500 (50 × 50) iterations in π and τ. Estimates of both λs and Δλ are mean values after 15 simulations (totalling 37 500 simulations). Within a, higher values of λs equal greater fitness at differing levels of π and τ. Within b, positive values of Δλ indicate selection for a mast reproduction strategy while negative values imply selection for annual reproduction. Within each graph the white line represents the ‘crest’ or the maximum value of λs or Δλ at each level of τ. The vertical and horizontal solid lines are the estimates of the current values of mast frequency (πPasoh) and pre- and post dispersal seed predation (τPasoh) respectively. The area between the horizontal dashed lines lies within the 95% confidence intervals of the current observed mast frequency (πPasoh).The simulations here are for the simplified model, only considering the pure demographic costs of lost reproductive opportunities and the death of adults.

In accordance with the above results, selection landscapes (Δλ; Fig. 3b) showed that selection for mast fruiting was the most extreme beyond the τ > 1 level. The optimal strategy at every given level of τ (white line in Fig. 3a,b) implied that delayed reproduction was already favourable even at low levels of τ. The intersection between the current estimates of π and τ occurred at a point where selection for even more delayed reproduction was strong, according to our simplified baseline model.

Additional costs of masting

In the above analysis, we evaluated the situation where the only cost of mast fruiting was limited to that of postponed reproduction (i.e. lost reproductive opportunities and death of mature individuals between masts). To test the robustness of the above patterns to increased limitations and costs of mast fruiting, we applied four additional scenarios. When the production of large fruit crops causes tree growth to stop during mast-fruiting years (scenario A, Fig. 4a), selection towards increasing delays in reproduction is strong, occurring even at low levels of seed predation. This shows that reducing adult growth leads to small shifts in the selection landscapes and that such costs would play a very limited role in the selection of mast fruiting. In contrast, when assimilate storage is limited (scenario B), stronger changes occur (Fig. 4b). Limited storage directly affects the optimal strategy, greatly diminishing the advantages of very long inter-mast periods. The optimal mast frequency (white line, Fig. 4b) was related to the maximum crop size (here six times annual crop) and limited storage efficiency could therefore reduce the optimal length of the inter-mast period. The third scenario (C), in which density-dependent seedling survival was taken into account, led to a disappearance of masting benefits at low-seed predation (Fig. 4c). An annually fruiting strategy now remains favourable up to τ = ±0.6 (i.e. ≤60% of the annually potential seed crop predated). Negative density-dependent seedling survival also decreases the advantage of very large seed crops and associated extreme intermast periods. Finally (scenario D), when including all the above-mentioned costs, we see that mast fruiting only becomes beneficial at high-seed predation intensities. The optimal fruiting strategy at current predation levels is also limited to a point well within the confidence intervals of the current mast strategy (πPasoh) displayed at Pasoh forest (Fig. 4d).

Figure 4.

 Selection landscapes as in Fig. 3b for scenarios A–D in the robustness analyses. (a) Selection landscape with growth stops in mast years (scenario A). (b) Limited storage of accumulates (scenario B). (c) Density-dependent seedling survival (scenario C). (d) All costs combined (a–c; scenario D). Here, the white line represents the maximum value of Δλ at each level of τ. The vertical and horizontal solid lines are the estimates of the current values of mast frequency (πPasoh) and pre- and post-dispersal seed predation (τPasoh) respectively. The area between the horizontal dashed lines lies within the 95% confidence intervals of the current observed mast frequency (πpasoh).

Discussion

Balancing costs and benefits of masting

This study is the first to quantify the balance of demographic costs and benefits of mast fruiting over the entire life cycle of a masting tree species. For our study species S. leprosula, we showed that masting is a low-cost strategy (Fig. 2a) to escape from heavy-seed predation. Selection landscapes revealed that masting was favourable for S. leprosula at various predation levels, but mast-fruiting was especially beneficial at high-seed predation, i.e. when seed predators are able to remove the entire annual seed crop (τ ≥ 1). Accounting for plausible costs of masting did not change these conclusions: reduced growth during masts, negative density-dependent seedling survival, limited storage or their combined effect could not outweigh the demographic benefits of masting.

We show that at existing predation levels, masting was highly beneficial allowing population growth rates to be only slightly less than those of an annually reproducing population experiencing no predation. This is a surprising result because mast fruiting was commonly perceived to reduce fitness compared with annually fruiting species through ‘lost opportunities of reproduction’ (Silvertown 1980; Kelly 1994). Waller (1979), for instance, predicted that mast-fruiting species must increase life-time reproductive output to maintain population growth rates compared with annually fruiting species. Our results are only consistent with this prediction when seed predation is ignored (as in Waller 1979). When seed predation is included, we show that annually-fruiting trees require drastic increases in life-time reproductive output to maintain positive population growth (λ > 1; Fig. 2c) when predation increases, while even decreased life-time reproductive effort can still result in positive fitness for masting trees (Fig. 2c). These results show that annually-fruiting species are hit disproportionately hard by seed predation and – as a result – the cost described by Waller (1979) may be trivial when seed predation is taken into account.

Our study shows that demographic costs due to negative density-dependent seedling survival and due to limited storing of reserves may be of much greater importance than those due to the ‘lost opportunities to reproduce’. Both negative density-dependent seedling survival and limited storage capacity were shown to have the strongest impact on the optimal fruiting frequency and may prevent lower mast frequencies to evolve. The strength of negative density-dependence in seedling survival was estimated from empirical data and proved strong enough by itself to move the optimal strategy more closely to the currently observed fruiting strategy. On the other hand, little is known about the limits of assimilate storage for dipterocarps, making this potentially equally important cost of masting an interesting topic for future studies.

A possible cost of masting not investigated here is the risk of masting in a year with unfavourable conditions for seedling establishment and growth. A known benefit of annual fruiting is that species can spread environment-related risks over time, a strategy often referred to as bet-hedging (Philippi & Seger 1989). However, this risk may be minor if masts occur non-randomly such that seedling establishment coincides with favourable conditions. A recent study by Williamson (2002) suggests such a positive relationship, which would render a potential cost into a benefit of masting. Uncertainty in the quantification of demographic costs of mast fruiting limits our understanding of this phenomenon and deserves further study in long-term population monitoring.

It seems that the life history of S. leprosula is well-suited for delayed reproduction. Waller (1979) showed that the costs of delayed reproduction is a function of adult survival; for every mature individual that dies between mast years ever greater compensation is required at the next reproductive episode. In the case of S. leprosula, delayed reproduction costs are low because annual survival rates for reproducing adults are high (0.98), with concomitant high-elasticity values (Table S3 and Fig. S1) and long life span. The approximate average conditional life span (Cochran & Ellner 1992) at first reproduction (>299 mm d.b.h.) is estimated to be 96 years, based on our study data. Such high-survival rates are, however, typical for many species of tropical trees (Condit, Hubbell & Foster 1995).

Another important adaptation which reduces the cost of mast-fruiting was introduced by Rees, Kelly & Bjørnstad (2002), who argued that a long-lived seed or seedling bank is an effective way to ‘store the gains’ of reproductive episodes. Shorea leprosula was found to have relatively high (67%) annual survival rates of (established) seedlings. In addition, dipterocarps display highly synchronized seed fall across species (while flowering occurs sequentially; Ashton, Givnish & Appanah 1988) combined with rapid germination (Burgess 1972; Fox 1976; Raich & Khoon 1990), both life-history characteristics that likely reduce the impact of seed predation (Curran & Webb 2000). A final possible evolutionary advantage of mast-fruiting is the prevention of seed predator specialization when seeds are produced at long intervals (Janzen 1974; Satake & Bjørnstad 2004), for which there is evidence as dipterocarps seem to lack host-specific enemies (Medway 1972).

In all, it appears that the life-history strategy of S. leprosula and that of related dipterocarp species seems well-adapted to maximize seed predator satiation and minimize costs of mast fruiting with long-living seedling banks, high levels of adult survival, synchronized fruiting and rapid seed germination. These adaptations and the low-demographic costs of masting that we found here are all consistent with the predator satiation hypothesis as an explanation of strict masting in Southeast Asia. Seed predators are known to have a major influence on population dynamics, structure and diversity of plant communities (Janzen 1971; Harper 1977; Hulme 1998; Paine & Beck 2007; Lewis & Gripenberg 2008), and our study suggests that seed predation can also influence evolutionary landscapes and induce clear selection pressure. Seed predation is extremely harmful to plant fitness and any strategy that allows escaping from seed predation likely increases plant fitness (e.g. escape by dispersal; Schupp & Fuentes 1995; Howe & Smallwood 1982). Strict mast fruiting is a life-history strategy that allows such escape from seed predation for long-living plants like trees. Our results suggest that the predator satiation may drive selection of mast fruiting, simply because heavy-seed predation causes disproportionately strong fitness penalties for annually fruiting species.

Why do not all tropical trees exhibit mast fruiting?

Mast fruiting can evolve when returns are maximized and costs are minimized. Our simulation results show that high returns occur at intense-seed predation when fitness benefits are greatest (τ > 1, Fig. 3), while low costs are maintained through high adult and seedling bank survival (Waller 1979; Rees, Kelly & Bjørnstad 2002; Figs 2–4). However, these conditions are common for tropical rain forests throughout the world, while strict mast fruiting is essentially restricted to Southeast Asia (Kelly 1994). In any (tropical) forest, extreme seed predation tends to be a rule rather than an exception (>90% loss, Janzen 1971; Crawley 1992; Fenner & Thompson 2005), adult tree survival (of large canopy species) is uniformly high (Condit, Hubbell & Foster 1995) and long-living seed or seedling banks are common (Vazquez-Yanes & Orozco-Segovia 1993; Baraloto & Goldberg 2004). Shorea leprosula, in addition, is a typical tropical tree species with demography and life history (growth, mortality, population growth rates and elasticity distribution patterns) very similar to those of many tropical forest tree species (Swaine, Lieberman & Putz 1987; Zuidema & Boot 2002; Zuidema et al. 2010). Therefore, both the general conditions for masting and the demography of our study species seem to be applicable to a wide range of tropical tree species. A fundamental question is, therefore, why is strict mast fruiting unique for Southeast Asia when it is such a beneficial strategy to overcome heavy seed predation?

Our model suggests that selection in favour of mast fruiting over annually fruiting is strongest when (i) fitness benefits from predator satiation are high along an axis of predation (Fig. 2b) and (ii) costs along an axis of delayed reproduction are low (Figs 2 and 3). Mechanisms that deter the evolution of masting will either operate through decreasing benefits from (i) or increasing costs along (ii). First, predator satiation depends on the intensity and spatial extent of a mast event (e.g. Curran & Leighton 2000) and is therefore foremost a function of the fruiting behaviour of the surrounding forest. The success of masting, as shown by our results (e.g. Figs 2–4), depends on community composition. Any benefits from predator satiation along (i) are thus mostly absent when the surrounding forest fruits annually, making in situ evolution of mast fruiting unlikely (Lalonde & Roitberg 1992). Ashton, Givnish & Appanah (1988) suggested that dipterocarps invaded from the seasonal tropics (present-day India) from an ancestral masting population, implying that dipterocarps already exhibited mast fruiting, likely in a less extreme form, when entering the aseasonal Southeast Asian tropics.

Secondly, seed–animal interactions probably play multiple roles in the evolution of mast fruiting. If seeds are destroyed by predators, selection forces may act to increase the per seed survival rate by mass fruiting and long inter-mast periods. If instead the consumer is not destructive but a mutualist, an increased rate of fruit consumption may in fact be beneficial.

Seed dispersal is a principal factor influencing plant fitness (Schupp & Fuentes 1995; Wenny & Levey 1997; Nathan & Muller-Landau 2000) and may be an alternative strategy to escape predation (Howe & Smallwood 1982). Reliance on animal dispersers may therefore increase the threshold for the evolution of strict masting (see also Silvertown 1980), as strict masting will likely cause competition, satiation and eventual loss of animal dispersers (see e.g. Wheelwright 1985; Herrera et al. 1998; Bas, Pons & Gomez 2006). It is noteworthy that the vast majority of Neotropical tree species do not mast (Nordon et al. 2007) and are dispersed by vertebrates (Howe & Smallwood 1982), while masting dipterocarps in Southeast Asia are not dispersed by animals (Curran & Leighton 2000; Curran & Webb 2000). Note that tree species exhibiting much less extreme forms of masting, e.g. oak trees producing fluctuating seed crops but maintaining some production every year (e.g. Kelly & Sork 2002), have developed mutualistic relationships with vertebrate dispersers (den Ouden, Jansen & Smit 2004).

Once a population successfully displays masting, other mechanisms theoretically could increase the net benefit of masting, promoting its persistence. Mast fruiting itself could decrease needs for defensive chemicals in seeds (Numata et al. 1999), while lower reproductive allocation needs (Fig. 2c) may have facilitated shifts in resource allocation to other variables limiting plant fitness, effectively increasing competitive ability. Furthermore, loss of specialist seed predators would alleviate Janzen–Connell type interactions, a major mechanism limiting species’ competitive ability (Gillet 1962; Janzen 1970; Connell 1971; Hammond & Brown 1998; Leigh 2004; Muller-Landau & Adler 2007; Lewis & Gripenberg 2008). The unique dominance of strict masting species, especially the dominant dipterocarps but also including other families (Janzen 1974; Ashton 1988; Ashton, Givnish & Appanah 1988), certainly bears witness to the success of mast fruiting in tropical Southeast Asia.

In this study, we show that masting is a highly beneficial strategy for tropical trees that experience heavy seed predation. We were able to reach that conclusion using demographic data and stochastic population models. These tools allow for a complete analysis of demographic costs and benefits of a life-history strategy such as masting (Poorter et al. 2005). Future demographic research can play an important role in helping to answer other exciting questions related to the evolution and success of (strict) mast fruiting in tree communities.

Acknowledgements

We thank the Forest Research Institute Malaysia (FRIM) for logistical support in the field. We are grateful for the helpful comments of four anonymous reviewers. In addition, we thank W. van der Werf, P.A. Jansen and S. Tuljapurkar for their comments and discussions. M.D.V. acknowledges funding from the Smithsonian Tropical Research Institute short-term fellowship program. E.J. acknowledges funding from the Netherlands Organization for Scientific Research (NWO-veni grant, 863.08.006). M.v.B. acknowledges support from the HSBC Climate Partnership. P.A.Z. was supported by ERC grant 242955. We acknowledge S.P.Hubbell, S.J. Wright and I-Fang Sun for their effort to establish the Pasoh forest dynamics plot seed and seedling monitoring project with financial aid from US NSF (grant no.: DEB-0108388). The Pasoh forest 50-ha dynamics plot research was in part supported by NSF grant DEB-0075334 to P.S. Ashton and S.J. Davies.

Ancillary