Male fecundity and pollen dispersal in hill dipterocarps: significance of mass synchronized flowering and implications for conservation


Correspondence author. E-mail:


1. To understand pollinator–plant symbioses and to conserve forest resources effectively, we need an understanding of pollen dispersal patterns and the heterogeneity (spatial and temporal) of male fecundity.

2. We investigated pollen dispersal patterns of the hill dipterocarp species Shorea curtisii using a modified Neighbourhood model. Seeds were collected following a sporadic general flowering (GF) event in 2002 and two mass GF events in 1998 and 2005. Seed paternity was then determined to investigate pollen dispersal patterns and the heterogeneity of male fecundity.

3. Pollen dispersal distances were longer, and the effective number of males lower, in the sporadic GF of 2002, than in the mass GF events of 1998 and 2005.

4. There was high between-individual variation in male fecundity, and small trees contributed less than large trees to the analysed seed pool. The results indicate that selective logging may adversely affect the reproduction of such species due to the removal of large trees from the reproductive population.

5.Synthesis. The large number of effective male individuals during the mass GF events provides evidence of Allee effects, suggesting that the mass GF events are of evolutionary significance and may contribute to the maintenance of genetic diversity. Future forestry practice needs to take into account variation in male fecundity, pollen dispersal and the mating system to ensure sustainable regeneration.


Synchronized flowering and subsequent mast seeding, whereby many paleo-tropical tree species synchronize their reproductive efforts, are widely reported in aseasonal Asian tropical forests (Ashton, Givnish & Appanah 1988; Sakai 2002). These phenomena are often referred to as general flowering (GF) events, because many species belonging to different taxa flower simultaneously, even without a seasonal climate. Major components of the aseasonal Southeast Asian Tropical forests are species of the family Dipterocarpaceae (Ashton 1982; Symington 2004), and GF intensity is known to fluctuate between events, in terms of both the number of dipterocarp species involved and the proportions of individuals of each species that flower (Numata et al. 2003). Various studies in Tropical forests have shown that low-density populations of some tree species have few pollen donors and reduced outcrossing rates, providing indirect evidence that density can potentially affect reproductive output through pollen limitation (e.g. Murawski et al. 1990; Murawski & Hamrick 1991; Nagamitsu et al. 2001). As pollen movement is a major component of gene flow, the effects of low tree density on pollinator abundance and behaviour may have adverse effects on the reproductive output of predominantly outcrossing trees, such as dipterocarps, and may alter the genetic structure of the next generation. Several studies (Ghazoul, Liston & Boyle 1998; Courchamp, Clutton-Brock & Grenfell 1999) have also shown that the adverse effects of reduced tree density on pollination and reproductive output may cause Allee effects, that is reduction in individual survival or breeding output at low population sizes (Stephens, Sutherland & Freckleton 1999). Furthermore, Allee effects may occur in populations with low flowering densities in weak or ‘sporadic’ GF events, as well as in populations with low densities of adult trees. Thus, as Allee effects increase the probability of extinction (Gilpin & Soule 1986), intense GF events may reduce extinction risks, by increasing the density of both flowering trees and their pollinator populations.

In peninsular Malaysia, the timing of GF events is heavily dependent on strong drought in the absence of El Niño conditions (Numata et al. 2003). In addition, several studies have detected effects of climate change in flowering phenology (e.g. Menzel & Fabian 1999; Miller-Rushing & Primack 2008) and plant–pollinator interactions (e.g. Harrison 2000; Memmott et al. 2007). Thus, as substantial climate change is anticipated (e.g. Meehl et al. 2007), further knowledge of temporal variation in mating processes (including variations in male fecundity, pollen dispersal and mating systems) during GF events of differing magnitudes, and ultimately their interactions with climatic variables, is required. Furthermore, in recent decades, human activities have caused considerable destruction of tropical forests. To develop sustainable management strategies for what remains of these ecosystems, it is important to understand the impact of deforestation on pollen dispersal patterns and gene flow (Lowe et al. 2005). The dipterocarp forests of Western Malaysia are among the richest in the world in terms of flora and fauna (Whitmore & Burnham 1975). They are also the most productive tropical forest types in Asia, with considerable timber value (FAO 2001). Although lowland dipterocarp forests have been extensively exploited in the Malaysian peninsula (with the exception of some forest reserves), hill dipterocarp forests generally remain intact, especially in the mountains that form the backbone of the peninsula. However, harvesting of dipterocarp lumber, known as Lauan, has now shifted to the hill dipterocarp forests (Sist et al. 2003). While hill dipterocarp forests are being harvested more sensitively than the lowland dipterocarp forests, actions to ensure their sustainable regeneration need to take into account aspects of reproduction such as variations in male fecundity, pollen dispersal and the breeding system, because tree species with reduced flowering tree density that have few and/or poorly dispersing pollinators are particularly susceptible to Allee effects (Ghazoul, Liston & Boyle 1998).

Recently, a model-based method has been developed to estimate pollen dispersal patterns using dispersal kernels (Burczyk, Adams & Shimizu 1996; Oddou-Muratorio, Klein & Austerlitz 2005), and to quantify the variability of individual male reproductive success in plant populations using paternity data (Burczyk & Prat 1997; Oddou-Muratorio, Klein & Austerlitz 2005; Klein, Desassis & Oddou-Muratorio 2008). Here, we present analyses of variations in pollen dispersal kernels and male fecundity during one sporadic and two mass GF events which aim to validate the hypothesis that differences in flowering tree density during such events influence pollen dispersal patterns, pollination parameters and reproductive outputs. Variations in male fecundity, including its relationship with tree size, during the three GF events are also reported. Information on these variations and correlations is important both from an evolutionary perspective and for developing sustainable tropical forestry practices, because trees of particular classes are removed in selective logging, for example.

Materials and methods

Research Plot, Sampling Strategy and DNA Extraction

The Semangkok forest reserve located in Selangor state, 60 km north of Kuala Lumpur, Malaysian Peninsula (2°58′N, 102°18′E), is a designated hill dipterocarp forest conservation area. In 1992, Niiyama et al. (1999) established an undisturbed, 6-ha permanent plot (200 m × 300 m) on a narrow ridge and steep slope, ranging from 340 to 450 m above sea level. All trees in the plot with a diameter at breast height (dbh) >5 cm were tagged. Samples of the inner bark or leaves of all Shorea curtisii individuals with a dbh >20 cm (144 individuals, including 17 individuals growing adjacent to the plot) were collected; these trees were considered to be the candidate pollen donors (Fig. 1). Mass GF events were observed in the hill dipterocarp forests of the Malaysian Peninsula in 1998 and 2005. In addition, sporadic flowering was observed in 2002, but flowering of a small number of trees seldom occurred during non-GF years. All individuals that flowered in the plot during the three events were carefully observed and recorded directly from the ground; however, the observations were not repeated within individual GF events. We collected seeds under the canopies of seven, 11 and 10 selected mother trees in 1998, 2002 and 2005, respectively. Although most flowering trees produced seeds, we selected mother trees that did not have overlapping canopies (to minimize seed migration from neighbours), produced sufficient seeds for sampling and were neither neighbours nor located at margins of the plot. The number of seeds produced in each GF year we analysed is presented in Table 1. Genomic DNA was extracted directly from the embryos of seeds and from the leaves or the inner bark of the potential pollen donors, using the procedure described by Murray & Thompson (1980). The extracted DNA from the adult trees was further purified using a High Pure PCR Template Preparation Kit (Roche, Basel, Switzerland). After RNA digestion, DNA from the embryo of seeds was diluted to a concentration of 2 ng μL−1.

Figure 1.

 The location of the Semangkok Forest Reserve and the distribution of adult Shorea curtisii adult trees (with a dbh >20 cm) in the 6-ha undisturbed study plot. The black circles and their diameter represent locations and dbh of the adult trees, respectively. The black squares indicate locations of S. curtisii trees on the ridge beside the plot that were also sampled. It should be noted that other adult trees grow adjacent to the plot, but were not sampled. The lines show pollen routes to two mother trees that crossed with the highest numbers of pollen donors detected in the seed pool in the sporadic (D390) and mass general flowering events (F040). The arrows represent immigrant pollen dispersal from outside the plot. The widths of the lines and arrows correspond to the number of pollen dispersal events detected, as indicated in the scale to the right of the figure.

Table 1.   Total number of seeds analysed for the modelling, mean number of detected allogamous pollen donors inside the plot and mean rates of categorical paternity, such as immigration, selfing and allogamous seeds sired by pollen donors revealed by paternity analysis
 Total number of seeds for analysisMean number of detected allogamous pollen donorsMean rate of immigrationMean selfing rateMean rate of allogamous seeds sired by pollen donors inside the plot
A   m s 1 – − s
  1. GF, general flowering.

1998 GF33916.000.24190.06490.6932
2002 GF40911.640.22490.14180.6333
2005 GF74420.600.32120.05910.6196

Molecular Analysis

All samples were genotyped with respect to 10 microsatellite markers previously developed by Ujino et al. (1998) and Lee et al. (2004, 2006). Polymerase chain reaction (PCR) amplifications were carried out in total reaction volumes of 10 μL using a GeneAmp 9700 (Applied Biosystems, Carlsbad, CA, USA). The PCR mixtures contained 0.2 μM of each primer, 1× QIAGEN Multiplex PCR Master Mix (Qiagen, Hilden, Germany), and 0.5–3 ng of template DNA. The temperature profile was as follows: 15 min at 95°C, then 30–35 cycles of 30 s at 94°C, 90 s at 50–57°C and 90 s at 72°C, followed by a 10-min extension step at 72°C. Amplified PCR fragments were electrophoretically separated using a 3100 genetic analyzer (Applied Biosystems) with a calibrated internal size standard (GeneScan ROX 400HD, Applied Biosystems). The genotype of each individual was determined from the resulting electropherograms by Genotyper software.

Microsatellite Marker Diversity, Paternity Assignment and Mating System

Before assigning paternity, offspring genotypes that conflicted with the assumed maternal tree genotypes were excluded from the offspring genotype array. After excluding anomalous offspring, we used categorical allocation in combination with an exclusion procedure to identify candidate paternal trees. The paternity of each offspring was determined by likelihood ratios, and their confidence levels (>95%) were derived using cervus ver. 3.0 software (Marshall et al. 1998; Kalinowski, Taper & Marshall 2007). For the likelihood tests in CERVUS, we created 100 000 simulated offspring genotypes from 600 potential paternal candidates, with a mistyping rate of 0.1% for categorical allocation. However, if the paternal candidates identified by the likelihood procedure had more than two loci mismatches in the simple exclusion procedure, we assumed that the paternal tree of the offspring was located outside of the plot. The difference in LOD score between the first and second candidates was not significant for two seeds in 1998, 10 in 2002 and 2 in 2005. After seeds that could not be allocated a single paternal candidate were excluded, 339 seeds in 1998, 409 in 2002 and 744 in 2005 were used in the modelling (Table 1).

Modelling Pollen Dispersal and Variance of Male Fecundity

The neighbourhood model has been widely used to estimate pollen dispersal parameters (Burczyk, Adams & Shimizu 1996; Burczyk & Koralewski 2005), the effects of ecological variance on mating processes and the effects of phenotypic traits on male mating success (Burczyk & Prat 1997; Burczyk et al. 2002; Oddou-Muratorio, Klein & Austerlitz 2005). Recently, Klein, Desassis & Oddou-Muratorio (2008) proposed a Bayesian approach to estimate parameters of pollen dispersal and pollen pool composition with male fecundity. An alternative approach is to estimate the parameters for a dispersal kernel using maximum likelihood methods to determine the probability of each assigned paternity for all the analysed seeds (Devlin, Clegg & Ellstrand 1992; Robledo-Arnuncio & Gil 2005; Tani et al. 2009). Here, we applied the model developed by Klein, Desassis & Oddou-Muratorio (2008), but the latter approach with Bayesian analysis was applied to estimate the dispersal kernel and male fecundity parameters, and effects of tree size.

Male fecundity

We used the method presented by Klein, Desassis & Oddou-Muratorio (2008) to determine the variation in male fecundity. The male fecundity of each mature tree j in the plot is denoted Fj and assumed to follow a log-normal distribution of mean 1 and variance Σ2; hence,


Therefore, the logarithm of mature tree fecundities follows a normal distribution:


The variance of male fecundity is related to the ratio of the observed density of pollen donors (dobs) to the effective density of pollen donors (dep), defined as the number of equifertile pollen donors per unit area, providing a probability of co-paternity before dispersal equal to that observed (Austerlitz & Smouse 2002). This relationship can be written (Oddou-Muratorio, Klein & Austerlitz 2005; Klein, Desassis & Oddou-Muratorio 2008):


In tropical forests, not all of the mature trees bloom evenly, even during mass GF events. This might lead to a small proportion of mature trees being dominant pollen donors and the majority contributing much less pollen. Therefore, we chose a distribution of large variance with a long tail.

Dispersal kernel

We applied an exponential power dispersal kernel to the model, using the probability of pollen travel from its origin (0,0) to be present in the pollen cloud at position (x, y) (Wright 1943; Clark 1998). This approach has been used in many previous studies (Austerlitz et al. 2004; Oddou-Muratorio, Klein & Austerlitz 2005; Klein, Desassis & Oddou-Muratorio 2008) in the form:


where Γ is the classically defined gamma function (Abramowitz & Stegun 1964) and inline image is the pollination distance. The parameter b is a shape parameter, affecting the tail of the dispersal distribution and a is a scaling parameter (see details in Clark et al. 1999; Austerlitz et al. 2004). The mean distance (δ) travelled by a pollen grain under the kernel p(a,b) (Clark, Macklin & Wood 1998) is given by:


Mating model and probability of paternity assignment

As in classical mating models (Adams & Birkes 1991; Burczyk et al. 2002), we divided seeds from the ith mother tree into three classes: si, the proportion fertilized by pollen from the ith mother tree (self-fertilization); mi, the proportion fertilized by pollen from an unrecorded donor outside the plot (immigration); and (1 − mi − si), the proportion fertilized by any pollen donor candidates from within the plot. Some of the local pollinations (1 − mi − si) could not be distinguished from pollination by immigrant pollen grains, which we defined as cryptic gene flow. However, we ignored the cryptic gene flow in the analysis because the level of cryptic gene flow should be small, given the high exclusion rates (Table S1 in Supporting information). Unlike the neighbourhood model, the proportions were obtained directly from paternity analysis. That is, the proportion si was obtained from nij(s)/Ai, where nij(s) is the number of seeds of the ith mother tree whose assigned paternal donor (j) was the same adult tree as the mother tree (selfing) and Ai is the number of seeds analysed from the ith mother tree. The proportion mi was obtained from inline image, where inline image is the number of seeds whose paternal donor was not detected in the plot. Therefore, the proportion (1 − mi − si) of seeds from the ith mother tree was assigned to the pollen donors growing in the plot. The ratio of the jth pollen donor candidate’s mating contribution to the ith mother tree in relation to the total number of allogamous seeds whose pollen donor was detected in the plot was inline image, where nij is the number of seeds from the ith mother tree sired by the jth paternal candidate. The expected probability of the ith mother tree’s seeds sired by the jth pollen donor was modelled as:


(model 1), where qij is the expected rate of the allogamous ith mother tree’s offspring sired by the jth pollen donor. It should be noted that the normalizing constant of the dispersal kernel cancels out, and that only one or zero candidate fathers were assigned to each offspring.

Evaluation of male fecundity variation with tree size

Two approaches were used to evaluate whether tree size had an effect on male fertility. First, to investigate the possibility of a selection gradient of tree size to male fertility, rather than treating the male fecundity parameter for the jth tree (Fj) as a random effect (as in model 1), in a modified model (model 2) the male fecundity of each adult tree in the plot was expressed as a product of its power-transformed diameter at breast height (dbh) and Fj (random effect). Therefore, male fecundity of the jth pollen donor was defined as inline image, where Dj and c are the diameter at breast height of the jth tree and a parameter quantifying the influence of tree size on male fecundity, respectively. We used normalized values of the logarithm of dbh for estimating the parameters. When dj = ln (Dj), inline image, where s is the standard deviation of the d of the 127 candidate pollen donor trees in the plot. Hence, the expected probability that the ith mother tree’s seeds were sired by the jth pollen donor was modelled as (model 2):


This mixed model was modified from Morgan & Conner (2001) and was considered a random effect mixed model. If the 95% posterior credibility interval for c does not include the value of zero, there is a significant relationship between tree size and male fertility. The 17 individuals in the set of candidate pollen donors located outside of the plot were excluded from the data set because their dbh was not recorded.

The male fecundity of trees of different sizes was also evaluated on the basis of distributional differences between dbh classes for Fj estimates from model 1 calculated on the basis of 10 000 MCMC iterations (explained below). The adult trees in the plot were categorized into six dbh classes (1, 20–35 cm; 2, 35–50 cm; 3, 50–70 cm; 4, 70–85 cm; 5, 85–110 cm and 6, >110 cm). The average estimated Fj estimates of the adult trees of each dbh class were calculated for each MCMC iteration. Then, the 95% credibility intervals of the average differences between pairs of dbh classes were calculated. If the credibility intervals included the value zero, we assumed that the male fecundity of trees of the two dbh classes was not significantly different.

Estimation of posterior probabilities of the parameters

The conditional likelihood function for M mother trees was expressed as:


The full posterior distributions are:


The prior parameter values of Fj are assumed to be independent and identically distributed with the same prior distribution. The log-scale male fecundity, p(f|σ2), has a normal distribution with mean −σ2/2 and variance σ2. The inverse of the hyper-parameter (σ2) was assumed to follow a Gamma distribution with values of 0.1 for both the shape and rate parameters, as this represents little a priori information for model 1. In contrast, the hyper-parameter (σ2) was assumed to follow a uniform distribution with values of 0 for the shape and 100 for the rate parameters of model 2. The tree size parameter, c, was assumed to be normally distributed with values 0 for the shape and 0.01 for the rate parameters of model 2. For the mutually independent dispersal kernel parameters, a and b, we assumed Gamma (0.01, 0.01) for both models prior to analysis.

Bayesian estimation using the Markov Chain Monte Carlo (MCMC) algorithm

Bayesian analysis calculates posterior distributions of parameters, producing conditional distributions that are updated on the basis of observations (Clark & Gelfand 2006). Prior distributions were first defined and then modified according to the observations relating to the probability of the paternal origin of the seeds. The re-parameterization to fit the data was performed by MCMC sampling using jags software on the r platform (Plummer 2003). The R script can be provided by the authors upon request. The MCMC procedure for the model was run for 10 000 iterations after a burn-in of 2000 iterations. The MCMC convergence was determined on the basis of observations after every six iterations, according to the behaviour of the chains with respect to all estimated parameters, which were visualized using Convergence Diagnostic and Output Analysis (Best, Cowles & Vines 1995). The value of Gelman and Rubin’s convergence diagnostic was estimated to validate the convergence of MCMC for each parameter (Gelman & Rubin 1992).

Effective number of pollen donors

Given the parameter values of model 1 for male fecundity (F = {Fj}= 1…N) and the dispersal kernel p(a,b; x,y), the proportion of pollen from each pollen donor j in the pollen pool of each mother i originating from all known fathers is given by the mass-action law:




Here, self-fertilization was not taken into account for the estimation of πij. The effective number of pollen donors (Nep) and the effective number of pollen donors for each mother tree (Nepi) in the whole pollen cloud sampled by the mother trees in each GF year was determined. These statistics were calculated as the inverse of the average probability of paternal identity of randomly drawn pairs of seeds from within the seeds collected during each GF year (Klein, Desassis & Oddou-Muratorio 2008):




where ni is the number of seeds analysed from the ith mother tree and n is the total number of seeds analysed in each GF year.


Diversity of Microsatellite Markers and Paternity Analysis

On average, 12.3 alleles, ranging from 5 to 17 alleles per locus, were detected in the S. curtisii adult tree samples. The observed and expected heterozygosities ranged from 0.347 to 0.866 (average, 0.709) and from 0.378 to 0.847 (average, 0.721), respectively and no significant Hardy–Weinberg disequilibrium was observed. The high variability associated with the markers resulted in a very high total exclusion probability for identifying the second parents of offspring in the paternity analyses (0.999756).

The mi pollen rates were virtually synonymous with the rate of pollen dispersal from outside the plot (immigrant), because the high paternity exclusionary power of the microsatellite markers reduced the rate of cryptic pollen dispersal to a negligible level. The detected rates of pollen dispersal from outside the plot varied between mother trees. The immigrant rates for some mother trees (located at edges of the study plot or in other clumps of S. curtisii) exceeded 30%. The selfing rates (si) estimated from the paternity analysis showed that outcrossing predominated in the species (the mean value of si was 0.0831). The proportion of allogamous seeds sired by pollen donors inside the plot (1 − mi − si) is informative for estimating the parameters. More than 60% of the 1492 seeds analysed across the three GF events were assigned to pollen donors inside the plot, allowing us to estimate parameters associated with pollen dispersal and male fecundity (Table 1).

Pollen Dispersal Kernel Estimates

The 95% credible interval for the dispersal distance δ was 30.5–165.0 m in 1998, 48.5–56.3 m in 2002 and 39.4–126.6 m in 2005, with means of 65.0, 81.6 and 67.1 m, respectively (Table 2). Hence, the mean dispersal distance δ was greater in 2002 than in 1998 or 2005, although the 95% credible intervals overlapped. When the means of the posterior distributions for the dispersal parameters were applied, the shapes of the dispersal kernels for 1998 and 2005 were similar, approximately following an exponential distribution with shape parameter b close to 1 (Tables 2 and S3). However, the mean value of shape parameter b for 2002 was 1.165, resulting in a platykurtic probability distribution that differed from those for 1998 and 2005 (Fig. 2). Nevertheless, the 95% credible intervals of the parameter b showed that the shapes of dispersal kernels did not differ significantly from the exponential kernel in any of the three GF years (Tables 2 and S3).

Table 2.   Posterior mean, standard deviation, median and 95% Bayesian credibility of dispersal kernel and male fecundity parameters, estimates of population density and effective pollen donors estimated from the model 1 and estimate of tree size parameter c from the model 2
ParameterMeanSDMedian2.5% lower97.5% upper R*
  1. GF, general flowering.

  2. *Gelman and Rubin’s convergence diagnostic.

1998 GF
 a 33.389166.72234.29448.04522.8311.02
 b 1.0140.1271.0050.7911.2871.02
 c 0.9180.2030.9140.5301.3301.00
 δ 65.03367.89030.534164.984
 d obs/dep11.55410.9864.24562.931
 N ep 32.06427.21016.06949.704
2002 GF
 a 52.994417.36253.64867.68641.4081.00
 b 1.1650.1281.1590.9321.4371.00
 c 1.2030.2641.1930.7161.7541.00
 δ 81.55283.16548.540156.276
 d obs/dep24.98523.4917.250183.619
 N ep 29.88824.34914.98349.859
2005 GF
 a 36.738256.67437.38347.63527.8861.00
 b 1.0420.0921.0370.8751.2351.01
 c 1.0220.1661.0180.7061.3591.00
 δ 67.14669.86339.405126.604
 d obs/dep10.57510.1734.67736.056
 N ep 39.13734.41122.16654.814
Figure 2.

 Estimated normalized pollen dispersal kernels from model 1. Grey, black and dotted lines indicate dispersal kernels (derived from the posterior means of a and b) for the 1998, 2002 and 2005 general flowering events, respectively.

Estimates of the Variance of Male Fecundity

Large variance in male fecundity was observed in all three GF events. However, the variance revealed by the dobs/dep ratio during the sporadic 2002 GF event was nearly double that during the 1998 and 2005 GF events. The posterior mean, median and 95% credible intervals of the dobs/dep ratios are summarized in Table 2. The dobs/dep ratio exhibited a long-tail distribution during all three GF events. A few individuals were very fecund, especially in the 2002 event (when the maximum posterior median of relative fecundity was 7.9), but most individuals within the study plot exhibited very low fecundity (with posterior median Fj values <0.1). Most of the individuals with high fecundity (>1) were in flower during the field observations. However, many individuals that were observed flowering in the field exhibited low fecundity (Fig. 3 and Table S3).

Figure 3.

 Box plot of posterior median (central diamonds), 50% (box) and 95% (line) Bayesian Credibility of 144 individual male fecundities during the three general flowering events (model 1). The black circle indicates flowering of adult trees confirmed by field survey.

Credibility intervals of c exceeding 0 were obtained for all three GFs (Table 2), indicating a relationship between tree size and individual male fecundity. The distributions of the posterior probability of Fj were significantly lower for the small dbh classes (<50 cm) than for the large dbh classes (70 to >110 cm) during all three GF events, and for all pairs of classes except 35–50 cm class and >110 cm class during 1998. During the sporadic GF event in 2002, there were no exceptions to this general finding, but during the 2005 mass GF, the distribution was significantly lower for class 20–35 cm class than for 50–70 cm class, while the distributions for 35–50 cm and 50–70 cm classes were similar (Fig. 4). The significance of the posterior distributions of 50–70 cm class and the smaller dbh classes (<50 cm) differed among the three GF events.

Figure 4.

 95% Bayesian credibility of pairwise subtraction of mean Fj estimates between tree size classes estimated from model 1. Individuals were assigned to one of six dbh classes, as follows: (1) 20–35; (2) 35–50; (3) 50–70; (4) 70–85; (5) 85–110; and (6) more than 110 cm.

Effective Pollen Donors and Densities of Pollen Clouds

The effective numbers of pollen donors inside the study plot during the 3 years were estimated to be Nep = 27.21, 24.35 and 34.31 (posterior medians) for 1998, 2002 and 2005, respectively. The estimates were larger for the 1998 and 2005 GF events than for the 2002 GF, amounting to approximately a fifth of the recorded number of individuals in the study plot. Smaller numbers of pollen donors (c. 24) contributed to mating during the sporadic 2002 GF event (Table 2). It should be noted that the variation in the number of effective pollen donors could have been affected by differences in the sets of mother trees sampled during the 3 years. Therefore, we compared the effective number of pollen donors of ith mother trees (Nepi) sampled during more than one GF event; two sampled during both the mass 1998 and the sporadic 2002 GF events, and one sampled during both the sporadic 2002 and the mass 2005 events. For these trees, the Nepi was larger during the mass GF events than during the sporadic event. However, the differences were within the 95% credibility intervals.


Pollen Dispersal Patterns

To date, the pollen dispersal distances of about 20 tree species in tropical forests have been measured (reviewed in Ward et al. 2005; Dick et al. 2008). In general, pollen dispersal seems to be negatively correlated with the density of reproductive trees (Stacy et al. 1996). However, many studies have also shown that pollinators affect the pollen dispersal distance (reviewed in Dick et al. 2008). It has frequently been observed that self-fertilization of species pollinated predominantly by weakly flying species increases when the density of reproductive trees is low, as in selectively logged forests, during sporadic GF events or when mother trees are isolated (Murawski, Gunatilleke & Bawa 1994; Fukue et al. 2007; Naito et al. 2008; Tani et al. 2009). This is in sharp contrast to tropical tree species that depend on pollinators with strong flight (Dick, Etchelecu & Austerlitz 2003; Cloutier et al. 2007; Lourmas et al. 2007). It has been reported that the main pollinators of Shorea species are thrips with weak flight (Appanah & Chan 1981), although small beetles make some contribution (Momose et al. 1998; Sakai et al. 1999). The short pollination distance and rapid decline in pollination probability with distance of Shorea species also indicate that thrips are probably their main pollinators.

Tani et al. (2009) conducted paternity analysis and used a mating process model to estimate pollen dispersal kernels for two lowland Shorea species that are thought to have similar pollinators to S. curtisii. The average pollen dispersal distances (δ) were estimated to be 314 m for Shorea leprosula and 852 m for Shorea parvifolia. However, in the current study, δ (posterior mean) was estimated to be much shorter: 65 m in 1998, 82 m in 2002 and 67 m in 2005. The densities of reproductive trees were 15.0 (1998), 7.5 (2002) and 12.5 (2005) trees/ha, substantially higher than those of S. leprosula (0.75 trees ha−1) and S. parvifolia (0.23 trees ha−1) in the lowland dipterocarp forest examined, in the 2002 GF, by Tani et al. (2009). These densities near the mother trees are higher than the regional values because the species clusters on ridges, and the high densities of S. curtisii in their favoured ridge habitats may lead to shorter pollen dispersal distances.

Evolutionary Significance of Mass GFs for Conservation of Genetic Diversity

When the density of reproductive trees was low, during the sporadic GF event in 2002, a platykurtic dispersal kernel was apparent, with pollen dispersed over longer distances than during the mass GF events. Under these conditions, the probability of thrips encountering another reproductive tree over a short distance was decreased, resulting in a broader dispersal kernel. The fluctuation in the density of reproductive trees also affected the number of effective males; the sporadic GF event was associated with a lower number of effective males than either of the mass GF events, although the 95% credible intervals of Nep between the GF of 2002 and those of 1998 and 2005 overlapped. Furthermore, mass GF promotes a sharp increase in pollinator populations, and hence, their pollination activity among reproductive trees (Sakai 2002). Consequently, the larger numbers of effective males in mass GF events reduces Allee effects. Our observations indicate that synchronized flowering effectively promotes successful pollination, even in relatively monotypic hill dipterocarp forest, and thus, mass GF events are important for maintaining genetic diversity. In contrast, sporadic flowering might increase risks of inbreeding due to the restricted numbers of mating partners. Accordingly, mean selfing rates were higher in the sporadic GF event than in the mass flowering events (Tables 1 and S2).

Forest Management and Conservation

Dipterocarp forests were among the first tropical forests in which sustainable timber production was attempted (Wyatt-Smith, Panton & Mitchell 1963). By the mid-1970s, harvesting had shifted to the hill dipterocarp forests. In the late 1970s, the Malayan Uniform System (Wyatt-Smith, Panton & Mitchell 1963) had already been abandoned in hillier terrain, and the selective management system (SMS) was introduced. In this system, the felling regime is based on a pre-felling inventory, which prescribes a flexible cutting limit with a minimum dbh (non-dipterocarps, 45 cm; dipterocarps, 50 cm), and ensures that sufficient residual trees are left to form the next cut in c. 30 years time (Thang 1987). However, neither the MUS nor SMS take account of mating and pollination processes. Our results demonstrate a significant positive relationship between male fecundity and tree size (Tables 2 and S4). The small trees (which would be left after logging by these regimes) had a low probability of flowering during GF events, and lower male fecundity than medium or large trees. Even when the small trees did produce flowers, the mean individual fecundity was significantly lower for trees <35 cm dbh than for the intermediate size (50–85 cm dbh) during the three GF events described herein. The second small trees (35–50 cm dbh) also produced less pollen than other classes, although this difference was not always statistically significant (Fig. 4). These findings indicate that selective logging may reduce pollination rates, as the residual trees are small (<50 cm under the SMS regime). Furthermore, the short pollen dispersal distances are likely to expose mother trees to allogamous pollen deficits when the density of reproductive trees is reduced. Nevertheless, Appanah & Manaf (1994) observed that small residual trees flowered after selective logging and that their seeds often escaped predator foraging, surviving at high rates. This raises the possibility that removing the larger trees might favour more homogeneous participation of the smaller trees in reproduction and thus reduce the excessive contribution of a few large males. However, they may produce pollen and seeds at lower rates than larger trees in undisturbed forests. Furthermore, evidence of reduced outcross mating has been obtained from a selectively logged forest adjacent to our study plot (Obayashi et al. 2002).

In future studies, we must clarify the genetics of trees in selectively logged forest throughout their growth stages and identify the effects of genetics on growth performance (genetic effects of inbreeding) and the demography of selfed and outcrossed individuals produced by residual trees in selectively logged forests. The information generated should then be incorporated into management programmes for hill dipterocarp species.


The authors appreciate Drs. S. Ibrahim, T. Okuda, T. Gotoh and R. Tabuchi for the management of projects, Mrs. G. Jaafar, Y. Marhani, R. Ponyoh, Y. Baya and A. G. Mohd-Afendi for their field assistance and Ms. D. Mariam, Ms. Nor Salwah A. W. and Ms. Nurl H. M. for their assistance with DNA extraction. We also thank Ms. K. Obayashi and Dr. T. Katsuki for providing some of the DNA material and Ms. M. Koshiba and Ms. K. Ito for their laboratory and administrative support. We also thank anonymous two referees for valuable comments that allowed us to improve the analysis and manuscript. The study was partly supported by a JIRCAS-FRIM joint project and the Global Environment Research Program of the Japanese Ministry of Environment (grant no. E-4).