INTRODUCTION
The tropical rain forests of South-East Asia are among the most diverse terrestrial ecosystems on Earth (Soepadmo Reference SOEPADMO, Primack and Lovejoy1995). The species diversity has been explained with regard to coexistence of plant species based on habitat heterogeneity where each species is best suited to a particular habitat (Tilman & Pacala Reference TILMAN, PACALA, Ricklefs and Schluter1993), but whether the habitat heterogeneity maintains the high tree species richness is still controversial. A number of hypotheses have been proposed and these can be categorized into equilibrium and non-equilibrium categories (Connell Reference CONNELL1978). The equilibrium category was explained by partitioning of habitat or regeneration niches (Ashton Reference ASHTON1969, Grubb Reference GRUBB1977). In agreement with these hypotheses, an association of tree species with physical habitats in species-diverse tropical forests and interspecific differences in these habitat associations are commonly observed in tropical rain forests (Clark et al. Reference CLARK, CLARK and READ1998, Comita et al. Reference COMITA, CONDIT and HUBBELL2007, Davies et al. Reference DAVIES, PALMIOTTO, ASHTON, LEE and LAFRANKIE1998, Debski et al. Reference DEBSKI, BURSLEM, PALMIOTTO, LAFRANKIE, LEE and MANOKARAN2002, Harms et al. Reference HARMS, CONDIT, HUBBELL and FOSTER2001, Itoh et al. Reference ITOH, YAMAKURA, OHKUBO, KANZAKI, PALMIOTTO, LAFRANKIE, ASHTON and LEE2003, Noguchi et al. Reference NOGUCHI, ITOH, MIZUNO, SRI-NGERNYUANG, KANZAKI, TEEJUNTUK, SUNGPALEE, HARA, OHKUBO, SAHUNALU, DHANMMANONDA and YAMAKURA2007, Sri-Ngernyuang et al. Reference SRI-NGERNYUANG, KANZAKI, MIZUNO, NOGUCHI, TEEJUNTUK, SUNGPALEE, HARA, YAMAKURA, SAHUNALU, DHANMMANONDA and BUNYAVEJCHEWIN2003, Svenning Reference SVENNING1999, Webb & Peart Reference WEBB and PEART2000, Yamada et al. Reference YAMADA, TOMITA, ITOH, YAMAKURA, OHKUBO, KANZAKI, TAN and ASHTON2006). On the other hand, non-equilibrium hypotheses were explained by the Janzen–Connell effect (Connell Reference CONNELL, den Boer and Gradwell1971, Janzen Reference JANZEN1970) and the unified neutral theory (Hubbell Reference HUBBELL2001). The Janzen–Connell effect predicts that reduced recruitment near reproductive conspecifics due to host-specific pests will enhance creation of space for other plant species, and whether other plant species can occupy this space depends on chance factors determined by the density of productive trees close to the empty space. Hubbell's model predicts that hyper-diverse communities such as rain forest will principally be determined by neutral factors and successfully explains patterns of relative species abundance in neotropical plant communities.
Although many previous studies found evidence that tree species conform to the equilibrium theory, some studies indicated that limited or habitat-biased distribution patterns may be the ephemeral or transient result of a population's history of seed dispersal and immigration (Losos Reference LOSOS1995, Primack & Miao Reference PRIMACK and MIAO1992). Harms et al. (Reference HARMS, CONDIT, HUBBELL and FOSTER2001) concluded that the effects of physical factors may contribute little to maintenance of species richness, since so many species exist in the same ecological habitat. In contrast, Noguchi et al. (Reference NOGUCHI, ITOH, MIZUNO, SRI-NGERNYUANG, KANZAKI, TEEJUNTUK, SUNGPALEE, HARA, OHKUBO, SAHUNALU, DHANMMANONDA and YAMAKURA2007) suggested that even among the species which associated with same ecological habitat, the sympatric species slightly differentiate the position of maximum abundance of each population along a habitat gradient to achieve coexistence. Since our knowledge of species-habitat associations in tropical rain forests is inadequate, an integration of information concerning species-habitat associations in many tropical forests is needed (Yamada et al. Reference YAMADA, TOMITA, ITOH, YAMAKURA, OHKUBO, KANZAKI, TAN and ASHTON2006). As a consequence of the above mechanism and biotic interactions (e.g. seed dispersal and host-specific predation), spatial structuring of each species emerges in the tropical rain-forest community. Therefore, studying the spatial organization of such communities should improve our understanding of the mechanisms responsible for the diversity of tropical rain forest (He et al. Reference HE, LEGENDRE and LAFRANKIE1996).
Fagaceae, which comprises species of apparently limited dispersal ability (Ashton Reference ASHTON1988), are one of the codominant families in montane forest in South-East Asia (Fujii et al. Reference FUJII, NISHIMURA and YONEDA2006, Symington Reference SYMINGTON1943, Wyatt-Smith Reference WYATT-SMITH1963). Ashton (Reference ASHTON1988) noted that the altitudinal distribution of Fagaceae and Dipterocarpaceae in Malesia was suggestive of interfamilial competitive exclusion. However, the ecology of Fagaceae in Malesia is poorly investigated (Noguchi et al. Reference NOGUCHI, ITOH, MIZUNO, SRI-NGERNYUANG, KANZAKI, TEEJUNTUK, SUNGPALEE, HARA, OHKUBO, SAHUNALU, DHANMMANONDA and YAMAKURA2007), although some taxonomic, geological and morphological studies have been undertaken recently (Cannon Reference CANNON2001, Fujii et al. Reference FUJII, NISHIMURA and YONEDA2006, Kamiya et al. Reference KAMIYA, HARADA, OGINO, CLYDE and LATIFF2003). In the present study, we investigated the spatial patterns of Fagaceae in association with their ecological habitat in a mid-elevation hill dipterocarp forest in Sumatra.
Fagaceae show greatest dominance and relatively high diversity in Sumatran mid-elevation hill dipterocarp forest (Fujii et al. Reference FUJII, NISHIMURA and YONEDA2006, Nishimura et al. Reference NISHIMURA, YONEDA, FUJII, MUKHTAR, ABE and KANZAKI2006a, Reference NISHIMURA, YONEDA, FUJII, MUKHTAR, ABE, KUBOTA, TAMIN and WATANABEb). This study aimed to elucidate factors determining the coexistence of species of Fagaceae in a mid-elevation hill dipterocarp forest in Ulu Gadut, Sumatra. To provide a critical test of niche differentiation we studied congeneric species (mainly Lithocarpus and Quercus), which minimized interspecific differences in dispersal syndrome and the effects such differences would have on spatial patterns (Debski et al. Reference DEBSKI, BURSLEM, PALMIOTTO, LAFRANKIE, LEE and MANOKARAN2002). The following hypotheses were addressed: (1) Fagaceae are potentially dispersal-constrained and will thus show an aggregated spatial pattern in the forest compared with that of wind-dispersed codominant species; (2) therefore the distribution of juvenile trees will be associated with same habitat as adult trees; (3) since Ulu Gadut is a topographically complex site, the habitats occupied by Fagaceae will be associated with specific topography; (4) consequently, habitat-associated Fagaceae will aggregate in the preferred habitat; and (5) differences in habitat preference may play an important role in the coexistence of Fagaceae species in the forest.
METHODS
Site description
The study area comprised 3.99 ha within a 6.55-ha plot in a well-developed, mid-elevation hill dipterocarp forest in Ulu Gadut, Padang, West Sumatra (00°53′S, 100°21′E). The site has a humid climate with a remarkable mean annual precipitation of about 6000 mm. The vegetation of the forest stand is essentially hill dipterocarp forest sensu Symington (Reference SYMINGTON1943) with no species exhibiting dominance. The forest stand has been damaged by small-scale local logging (non-mechanized logging by hand), which is one reason why dipterocarp species did not show dominance. The research plot was divided into 1568 quadrats (5 m × 5 m, horizontal distances) and a PVC post was placed at the corners of each quadrat. The relative elevation at each post was determined by land survey. Detailed description of the vegetation and topography of the site is presented in Nishimura et al. (Reference NISHIMURA, YONEDA, FUJII, MUKHTAR, ABE and KANZAKI2006a).
Tree census
All trees of 10 cm diameter at breast height (dbh) or greater were mapped, marked with a numbered aluminium tag and the dbh was measured (above any buttresses if present). Each tree was identified to species level at the Herbarium Bogoriense to clarify the species composition of the forest stand. All voucher specimens are lodged in the Kagoshima University, Japan. The heights of 857 trees (25% of the total tree number) of a range of dbh were measured using a measuring pole and Haga altimeter.
Target species
Fagaceae species are one of the major components of this forest stand. The species number, abundance and basal area dominance of this family ranked sixth, sixth and first, respectively, out of 64 families in the 6.55-ha plot (Nishimura et al. Reference NISHIMURA, YONEDA, FUJII, MUKHTAR, ABE and KANZAKI2006a). At least 23 Fagaceae species grew in the plot. Most of these are canopy tree species up to 30–40 m in height (Table 1). Among these, 10 target species from the genera Castanopsis (one sp.), Lithocarpus (six spp.) and Quercus (three spp.) were selected for use in this study based on their density in the plot, i.e. those species exceeding 10 individual trees of dbh ≥ 1 cm in the inner 3.12-ha area (as the outermost quadrats of 3.99-ha were unavailable for calculations of slope convexity). Scientific names used follow those in the revision by Soepadmo (Reference SOEPADMO and van Steenis1972) and the check-list by Govaerts & Frodin (Reference GOVAERTS and FRODIN1998). Detailed morphological and ecological descriptions of the species in Malesia are given by Soepadmo (Reference SOEPADMO and van Steenis1972).
Table 1. Status of 23 Fagaceae species within a 6.55-ha plot in a mid-elevation hill dipterocarp forest in Ulu Gadut, West Sumatra. The number of individuals and tree density were based on trees with dbh ≥ 1 cm. Basal area, mean diameter at breast height (dbh), maximum dbh and maximum height (H) were based on all trees with dbh ≥ 10 cm.
Five canopy or emergent codominant tree species, comprising one Anacardiaceae and four Dipterocarpaceae species, were selected for comparison. Each species can be a competitor of Fagaceae in terms of sharing the same or an upper stratum. Three of the species currently show high dominance and the other two species formerly exhibited high dominance in the plot. Swintonia schwenkii (T. & B.) T. & B. (Anacardiaceae) and Parashorea lucida (Miq.) Kurz (Dipterocarpaceae), the main emergent trees in the plot, are among the most common species at present. Hopea dryobalanoides Miq. (Dipterocarpaceae) is also one of the most common canopy trees. Some mature trees of Shorea maxwelliana King and Shorea atrinervosa Sym. were once the target of small-scale local logging (Nishimura et al. Reference NISHIMURA, YONEDA, FUJII, MUKHTAR, ABE and KANZAKI2006a). These two species were formerly among the emergent and dominant species in the plot. Kochummen (Reference KOCHUMMEN and Whitmore1989) and Ashton (Reference ASHTON and van Steenis1982) were used for identification of Anacardiaceae and Dipterocarpaceae, respectively.
All of the target Fagaceae species produce nuts, which are only shed beneath the canopy of the mother tree, whereas the five codominant species have winged fruits dispersed by wind. Fruit morphology suggests that the nuts are zoochorous through caching by small mammals after barochory and that the winged fruits are anemochorous probably with zoochory as a secondary dispersal mechanism.
All individuals, including newly established seedlings, of the 15 target species in the plot were mapped, marked with a numbered plastic tape, and the dbh and tree height were measured. In this study, trees greater than 1-cm dbh were defined as the tree stage and included pole and mature trees. Trees less than 1 cm dbh were categorized into two additional developmental stages, namely the large-sapling stage (dbh < 1 cm and height ≥ 50 cm) and small-sapling stage (height < 50 cm).
Seed dispersal of the target species was not observed in this study, therefore it is not possible to specifically discuss seed dispersal of these species. However, as Webb & Peart (Reference WEBB and PEART2000) noted, seedling distributions are strongly influenced by seed dispersal patterns, so the distribution patterns of plants at early growth stages should reflect seed dispersal patterns. On this basis, we estimated the dispersal constraint of these species using data for the small-sapling stage, which are expected to be less affected by biotic interaction than the tree stage. The median distances between small saplings (height < 50 cm) and the nearest reproductively mature conspecific tree (dbh ≥ 30 cm) were calculated for each species in the 6.55-ha plot to estimate dispersal constraint.
Data analysis
Univariate spatial patterns of the target species were analysed using Ripley's 
(t) function (Ripley Reference RIPLEY1977). The function (t) (λ = intensity) is defined as the expected number of target species within distance t of an arbitrary target species. The unbiased estimate of (t) is defined as:
where n is the number of target species in a plot; |A| denotes plot area; Uij is the distance between the ith target species and jth target species in A; It(u) is equal to 1 if u ≤ t and 0 otherwise; wij is the proportion of the circumference of a circle with its centre at the ith target species and with radius uij that lies within A; and summation is for all pairs of target species not more than t apart (Diggle Reference DIGGLE1983, Ripley Reference RIPLEY1977).
Square-root transformation of 
(t), as suggested by Besag (Reference BESAG1977), was applied in this study to detect spatial patterns for the tree stage. (t) is defined as:
A value of (t) = 0 indicates that the spatial pattern at distance t is random. Values of (t) > 0 indicate clumped distributions. Values of (t) < 0 indicate regular distributions. Significance of this function was determined with Monte Carlo simulations (Besag Reference BESAG1977, Besag & Diggle Reference BESAG and DIGGLE1977, Marriott Reference MARRIOTT1979). For this analysis, the null hypothesis is complete spatial randomness. A total of 10 000 simulations were performed to create 95% confidence intervals. (t) was examined for every 1-m interval from 0–60 m for distance t.
The inclination and compass direction of the slope was calculated for each 10 m × 10-m quadrat following the plane regression method of Yamakura et al. (Reference YAMAKURA, KANZAKI, ITOH, OHKUBO, OGINO, CHAI, LEE and ASHTON1995) using the elevation data for the four corners of each quadrat. The relative elevation of each quadrat was obtained by averaging the elevation at each corner. The surface relief of the slope was expressed using the index of slope convexity (IC) proposed by Yamakura et al. (Reference YAMAKURA, KANZAKI, ITOH, OHKUBO, OGINO, CHAI, LEE and ASHTON1995). A positive value of the index means that the slope relief is convex. IC could not be calculated for the marginal quadrats because the elevation outside the target quadrat is needed.
Since the study site was disturbed previously by local logging, we used the canopy height as one of the habitat variables that might explain the distribution of target species. Each 10 m × 10-m quadrat contained nine points in a grid of 5-m quadrats, and the height of the highest crown at each point was recorded. Crowns up to 15-m high were measured with a height-measuring pole. For crowns above 15 m in height, we used the height of the nearest measured tree to estimate relative height visually. The average canopy height of the quadrat was defined as the average value of the height of these nine points.
To detect the significance of habitat association of the target species, a torus-translation procedure was used based on that of Harms et al. (Reference HARMS, CONDIT, HUBBELL and FOSTER2001). This procedure consisted of moving the true habitat map about a two-dimensional torus by 10-m increments in the four cardinal directions (Harms et al. Reference HARMS, CONDIT, HUBBELL and FOSTER2001). More maps can be generated by each of three translations: 180° rotation, mirror image, and 180° rotation of the mirror image. On this basis a true habitat map and 1247 simulated habitat maps were produced. For the test of association, each of the 1248 habitat maps was overlain by the true distribution of trees and the relative density of each species was calculated for each habitat. Within the 6.55-ha plot, a 3.12-ha area was utilized for analysing the association of target species across the three developmental stages. The tree density of target species in the study site is rather lower than in previous studies (Aiba et al. Reference AIBA, KITAYAMA and TAKYU2004, Harms et al. Reference HARMS, CONDIT, HUBBELL and FOSTER2001, Yamada et al. Reference YAMADA, TOMITA, ITOH, YAMAKURA, OHKUBO, KANZAKI, TAN and ASHTON2006) that used the same analytic method. However, there are no compelling differences in how this method assesses the degree of habitat association between low-and high-density species (Harms et al. Reference HARMS, CONDIT, HUBBELL and FOSTER2001), therefore we proceeded to use this method to analyse the data.
Spatial interactions among species that showed the same habitat preference at the tree stage were analysed using the bivariate 1, 2(t) function, a transformation of function 1,2(t) (Besag Reference BESAG1977, Lotwick & Silverman Reference LOTWICK and SILVERMAN1982, Ripley Reference RIPLEY1977):
The 1,2(t) function indicates the spatial independence among two species at distance t. To examine statistical significance, we used simulated 1,2(t) values at the limit of the 2.5% tails of 10 000 torus randomizations (mean ± 1.96 SD) for 95% confidence intervals. If the sample statistic remains within the bounds of the confidence interval at any given t, the null hypothesis of spatial independence is not rejected, but when it exceeds the upper (or lower) boundaries the sampled points are regarded as an attractive (or repulsive) pattern (Diggle Reference DIGGLE1983).
Fisher's exact test was used to compare the median distance between small saplings and the nearest reproductively mature conspecific tree. False discovery rate (FDR)-corrected alpha (Verhoeven et al. Reference VERHOEVEN, SIMONSEN and MCINTYRE2005) was applied for the multiple comparisons across species.
RESULTS
Spatial pattern
The spatial patterns of ten Fagaceae (Figure 1) and five codominant wind-dispersed species (Figure 2) were investigated. Among Fagaceae, five species (Castanopsis rhamnifolia, Lithocarpus meijeri, L. reinwardtii, Quercus argentata and Q. gemelliflora) were significantly (P < 0.05) aggregated within most of the 60-m interval, whereas the other five species (L. hystrix, L. javensis, L. lucidus, L. macphailii and Q. oidocarpa) exhibited a rather random pattern (Figure 3). All of the codominant wind-dispersed species showed significantly aggregated distributions (Figure 4).
Figure 1. Spatial distribution of ten Fagaceae species in a 6.55-ha plot within a mid-elevation hill dipterocarp forest in Ulu Gadut, West Sumatra. Key to symbols: The circles indicate (from smallest to largest circle, respectively) the small sapling stage, large sapling stage, tree stage, and reproductively mature trees (dbh ≥ 30 cm). The dashed line indicates the 3.99-ha area used for the spatial pattern analysis. The contour interval is 10 m.
Figure 2. Spatial distribution of five codominant species in a 6.55-ha plot within a mid-elevation hill dipterocarp forest in Ulu Gadut, West Sumatra. See Figure 1 for explanation of the symbols.
Figure 3. (t) values for the ten species of Fagaceae at the tree stage in a 3.99-ha plot within a mid-elevation hill dipterocarp forest in Ulu Gadut, West Sumatra. The solid line represents actual (t) values for extant plants. The dashed lines represent the 95% confidence limits for the pattern expected from a random distribution of plant locations generated by 10 000 simulations. Values outside the limits indicate significant departure from a random distribution.
Figure 4. (t) values for the five codominant species at the tree stage in a 3.99-ha plot within a mid-elevation hill dipterocarp forest in Ulu Gadut, West Sumatra. See Figure 3 for explanation of the lines.
Habitat association
The results of torus-translation tests of habitat associations are shown in Appendix 1. Six Fagaceae species (C. rhamnifolia, L. hystrix, L. reinwardtii, Q. argentata, Q. gemelliflora and Q. oidocarpa) were significantly associated with one or more of the four habitat types at the small-sapling, large-sapling and/or tree stages. On the other hand, L. javensis, L. lucidus, L. macphailii and L. meijeri were considered to be independent of the four habitats tested in this study. Three of the codominant species (Hopea dryobalanoides, Shorea maxwelliana and Swintonia schwenkii) had significantly positive habitat associations.
For most of the species exhibiting significant habitat associations, the associations were not consistent at the three developmental stages. Among four topographic categories, IC was the biggest contributor to the habitat association. Among seven species that were associated positively with habitat at the tree stage, all of them were associated with IC. Although not statistically significant, some of the Lithocarpus species showed a tendency to be distributed on convex topography.
The population histograms for the three ridge-specialist Quercus species at the tree stage in relation to IC are shown in Figure 5. The histogram for Q. argentata is broader and with a less marked peak than for the other two Quercus species.
Figure 5. Relative frequency of three ridge-specialist Quercus species in relation to the index of slope convexity in a 3.12-ha plot in a mid-elevation hill dipterocarp forest in Ulu Gadut, West Sumatra. A positive IC value indicates convex local landforms and a negative value indicates a concave slope.
Spatial interaction between species
Spatial interaction between species was tested for each pair of congeneric species that showed the same habitat preference at the tree stage (Figure 6). Among congeneric Quercus species, spatial interaction between Q. argentata and Q. gemelliflora was negative or random throughout the 0–45-m distance interval. On the other hand, the interaction between Q. oidocarpa and Q. argentata was positive over a distance of 2–45 m. For Q. gemelliflora and Q. oidocarpa a mostly random interaction between 10 and 45 m and a negative interaction from 0 to 10 m was indicated.
Figure 6. 1, 2(t) value of bivariate distribution of each pairing between three Quercus populations at the tree stage. The solid line shows the actual 1, 2(t) values for extant plants. Dashed lines show 95% confidence limits for the pattern expected from an independent distribution of plant locations generated by 10 000 simulations.
Spatial interaction between Quercus and anemochorous species (Hopea dryobalanoides, Shorea maxwelliana and Swintonia schwenkii) that preferred the same habitat was tested. The interaction between Q. argentata and H. dryobalanoides was negative and that between Q. oidocarpa and S. schwenkii was random within 45 m, other than that the interactions between three ridge-specialist Quercus species and three ridge-specialist anemochorous species overlapped positively within 45 m. Spatial interaction among ridge-specialist anemochorous species was positively overlapping within 45 m in all species combinations.
Distance between small saplings and nearest reproductively mature tree
The median distances between small saplings (height < 50 cm) and the nearest reproductively mature conspecific tree (dbh ≥ 30 cm) are shown in Table 2. Although there were no significant differences among aggregated and random distributed species at the threshold distance (around 20 m), species with an aggregated distribution at the tree stage showed a tendency for small saplings to be closer to reproductively mature conspecific trees than for species with a random distribution pattern, except for L. reinwardtii. Among these species, the median distances of H. dryobalanoides, Q. argentata and Q. gemelliflora, each of which are habitat-associated and spatially aggregated species, were significantly smaller than the other species (except for S. schwenkii). On the other hand, the distance to reproductively mature conspecific trees was greater in spatially random species. The average median distance was significantly shorter in the aggregated group than in the spatially random group (Mann–Whitney U-test, n1 = 8, n2 = 5, U = 1.0, P < 0.01). However, comparison of the average median distance among Lithocarpus (37 m), Quercus (19 m) and anemochorous species (12 m) was not significant (ANOVA, df = 2, F = 3.5, P = 0.07).
Table 2. Median distance of small saplings (height < 50 cm) from the nearest reproductively matured conspecific tree (dbh ≥ 30 cm) in a 6.55-ha plot in a mid-elevation hill dipterocarp forest in Ulu Gadut, West Sumatra. Median distances with the same letter indicate that values do not differ significantly at the 5% significance level. False discovery rate-corrected alpha (Verhoeven et al. Reference VERHOEVEN, SIMONSEN and MCINTYRE2005) was used. Shorea atrinervosa and Shorea maxwelliana were omitted from this analysis because of a low sample size (n = 2).
DISCUSSION
Factors influencing spatial pattern
Of the ten Fagaceae species examined in the present study, five species exhibited an aggregated distribution pattern at the tree stage while the other five species showed a rather random spatial pattern. How did these spatial patterns come to arise? These ten species can be categorized into four groups based on their spatial pattern and habitat association (Table 3). It is likely that if a species has a specific habitat preference, the spatial pattern of the species will aggregate at the preferred habitat. For example, a previous study showed that 24 out of 25 bird-dispersed Aporosa species showed an aggregated distribution in the preferred habitat at Lambir, Borneo and/or Pasoh, Peninsular Malaysia (Debski et al. Reference DEBSKI, BURSLEM, PALMIOTTO, LAFRANKIE, LEE and MANOKARAN2002). The relationship between spatial pattern and habitat association of Q. argentata and Q. gemelliflora was in accordance with this assumption. However, for other species, this was partly rejected, as Q. oidocarpa and L. hystrix were distributed randomly but associated with a specific habitat. This implies that these species might exhibit superior survival at the preferred habitat after being widely dispersed. Castanopsis rhamnifolia, L. meijeri and L. reinwardtii, which are considered to be independent of habitat but have aggregated spatial patterns, might be able to establish irrespective of the specific habitat but their dispersal seems to be limited. The independence of habitat association and the random spatial pattern for three species (L. lucidus, L. javensis and L. macphailii) indicates that chance biotic interactions, such as secondary dispersal and/or post-dispersal survivorship, seem to be more important for determining the spatial pattern for these species. With regard to the wind-dispersed codominant species, all of the species showed an aggregated distribution pattern. Three of the species had specific habitat preferences while two species did not.
Table 3. Grouping of ten species of Fagaceae based on spatial pattern and habitat association.
In Fagaceae the nut falls to the ground directly beneath the mother tree and apparently has limited dispersal ability (Ashton Reference ASHTON1988) in contrast to the anemochorous species. Our results, however, showed that five Fagaceae species have a random distribution pattern whereas all of the wind-dispersed codominant species were aggregated. Enhanced seed dispersal reduces aggregation (Condit et al. Reference CONDIT, ASHTON, BAKER, BUNYAVEJCHEWIN, GUNATILLEKE, GUNATILLEKE, HUBBELL, FOSTER, ITOH, LAFRANKIE, LEE, LOSOS, MANOKARAN, SUKUMAR and YAMAKURA2000), thus an increasingly random spatial pattern is expected with increasing seed dispersal distance. Consequently, species with a random spatial pattern may be dispersed more widely than aggregated species. This implies that limitation of seed dispersal occurs in not only the five Fagaceae species but also in the codominant wind-dispersed species in the study plot. Comparison of the average median distance among Lithocarpus (37 m), Quercus (19 m) and anemochorous species (12 m) was not significant. However, there was a tendency for the median distance of Lithocarpus species to exceed that of either Quercus or even codominant anemochorous species. Seed dispersal mechanisms of Fagaceae in South-East Asia are poorly known (Cannon Reference CANNON2001), but it is thought that the nuts are scatter-hoarded by terrestrial squirrels (Corlett Reference CORLETT1998, Leighton & Leighton Reference LEIGHTON, LEIGHTON, Sutton, Whitmore and Chadwick1983, Vander Wall Reference VANDER WALL2001). We found quite a number of empty nuts with holes of L. javensis in a ground hollow that might have been carried and consumed there by a porcupine, which would indicate the existence of an animal disperser for at least some species of Fagaceae (S. Nishimura pers. obs.). Our results indicated that while the primary dispersal of Fagaceae nuts may be restricted to beneath the canopy of the mother tree, secondary dispersal by animals of spatially random species may occur over a longer distance than aggregated species in the study plot. If this is true, spatially random species must attract animals to disperse the nuts over greater distances. Animal preference for certain species might reflect differences in the concentration of polyphenols (condensed or hydrolysable tannin) in the nuts, which are thought to reduce the digestive efficiency of the nut predator (Vander Wall Reference VANDER WALL2001). It is also suggested that the longer seed germination period in Lithocarpus species, in particular, compared with that of Quercus and wind-dispersed species (Ng Reference NG1991; S. Nishimura unpubl. data) may increase the chance of the nuts being found by animals for gathering as a foodstuff for consumption and/or caching.
Species coexistence
Most seeds fall close to parent trees, therefore the density of seeds will tend to be higher in the preferred habitat of adult trees compared with other habitats. Thus seedlings may show associations with the same habitat as adults (Comita et al. Reference COMITA, CONDIT and HUBBELL2007), especially for potentially dispersal-constrained species such as the aggregated and habitat-associated species in the present study. However, if the associated habitat at the tree stage corresponds with the preferred habitat for the species, most ecological associations of these species will not develop during early growth stages in the study plot, thus the distribution of small saplings does not exactly correspond with the habitat of mature trees even it is an aggregated species with short median distances. A high frequency of seedling establishment beneath the mother tree would result in a high density of small trees in that habitat, but if negative density dependence outweighs the benefits of the habitat, survival in that habitat would be reduced (Comita et al. Reference COMITA, CONDIT and HUBBELL2007). This may account for the preferred habitat in early growth stages of Fagaceae differing from that of adult trees at Ulu Gadut. However, since those species associating with a specific habitat at the tree stage are retaining an aggregated distribution, seedlings may survive in topographically similar sites a limited distance from the parent tree.
Once the habitat is determined at the tree stage, how important is habitat association for coexistence of congeneric species? Previous studies that suggested the importance of topographical variation at spatial scales of approximately 50-ha showed opposite niche preference of congeneric species along a topographic gradient, e.g. two Dryobalanops species (Itoh et al. Reference ITOH, YAMAKURA, OGINO, LEE and ASHTON1997) and three Scaphium species (Yamada et al. Reference YAMADA, YAMAKURA, KANZAKI, ITOH, OHKUBO, OGINO, CHAI, LEE and ASHTON1997) at Lambir, Borneo. Gunatilleke et al. (Reference GUNATILLEKE, GUNATILLEKE, ESUFALI, HARMS, ASHTON, BURSLEM and ASHTON2006) also showed clear habitat partitioning of two Mesua species and three Shorea species which are growing sympatrically in a 25-ha plot in Sri Lanka. These results strongly suggest that topographical habitat differentiation determines the coexistence of congeneric species. Our results showed that the spatial distribution patterns of four Fagaceae species were positively associated with topography at the tree stage. Among these species, all Quercus species showed a similar dependency on topography and were associated with ridge sites. On the other hand, neutral associations were prominent for most Lithocarpus species, although a positive association with ridge sites was detected for L. hystrix. Positive association of three Quercus species with ridge sites had the effect of reducing confamilial competition from ten to three species, but habitat association alone does not provide sufficient support for the hypothesis that niche differentiation is the primary mechanism maintaining species diversity (Harms et al. Reference HARMS, CONDIT, HUBBELL and FOSTER2001, Tilman & Pacala Reference TILMAN, PACALA, Ricklefs and Schluter1993, Webb & Peart Reference WEBB and PEART2000), since so many species coexist in the same habitat.
Yamada et al. (Reference YAMADA, TOMITA, ITOH, YAMAKURA, OHKUBO, KANZAKI, TAN and ASHTON2006) suggested that coexistence of Heritiera species preferring the same habitat at Lambir may be related to their occupation of different forest storeys and hence differing regeneration requirements. However, Quercus species in the present study seemed to occupy the same stratum (Table 1). Davies et al. (Reference DAVIES, PALMIOTTO, ASHTON, LEE and LAFRANKIE1998) demonstrated the distribution of 11 sympatric Macaranga species along a light-intensity gradient at Lambir. Our results did not show a significant association between Quercus distribution and forest structure (average tree height), so the current distribution of Quercus species is likely to be independent of the light environment (but see Aiba et al. (Reference AIBA, KITAYAMA and TAKYU2004) that showed significant association with exposed canopy conditions for Q. cf. subsericea). In addition, the spatial patterns of Q. argentata and Q. oidocarpa overlapped significantly. Noguchi et al. (Reference NOGUCHI, ITOH, MIZUNO, SRI-NGERNYUANG, KANZAKI, TEEJUNTUK, SUNGPALEE, HARA, OHKUBO, SAHUNALU, DHANMMANONDA and YAMAKURA2007) suggested the importance of habitat divergence in the specialized habitat and showed that sympatric species may offset the position of maximum abundance of each population along a habitat gradient to achieve coexistence, which may contribute to the coexistence of sympatric Fagaceae at Doi Inthanon, Thailand. In the present study, three Quercus species showed a different distribution pattern to the slope IC (Figure 5). This subtle difference in their ecological preferences may permit them to coexist even in a similar habitat. It is suggested that positive association of Q. argentata with intermediate elevations may be one factor contributing to avoidance of complete overlap with Q. oidocarpa, which is monotonically associated with ridge habitats irrespective of elevational range. The reason why Quercus species preferred ridge habitats in our study is unknown but this topographic preference may be favourable for their pollination, which is reliant on wind. Therefore evolutionary constraints may at least partially account for the distribution of Quercus on the ridge rather than insect-pollinated genera such as Lithocarpus and Castanopsis. Though the distribution of each of the three Quercus species was not associated with forest structure (average tree height), it is likely that their current distribution also reflects the influence of logging. Spatial overlap with S. maxwelliana, a previous logging target, indicates that some of the three Quercus may have occupied the previous habitat of S. maxwelliana following logging. Thus the high dominance of Q. gemelliflora (second out of 465 species in the 6.55-ha plot), in particular, may partly reflect logging of the same habitat preferred by competitive species.
Yamada et al. (Reference YAMADA, TOMITA, ITOH, YAMAKURA, OHKUBO, KANZAKI, TAN and ASHTON2006) noted that a higher proportion of significantly positive associations between habitats was due to the topographic and edaphic complexity at Lambir. Condit et al. (Reference CONDIT, ASHTON, BAKER, BUNYAVEJCHEWIN, GUNATILLEKE, GUNATILLEKE, HUBBELL, FOSTER, ITOH, LAFRANKIE, LEE, LOSOS, MANOKARAN, SUKUMAR and YAMAKURA2000) documented that more species are associated with topographic features in rugged topography than in a topographically uniform site. The topography of the present study site was very complex. However, the neutral association with topography for the five Lithocarpus species indicated that topographic complexity is not the principal mechanism determining the coexistence of these species. This is in accordance with results for Lithocarpus species in Bornean lowland dipterocarp forest at Gunung Palung (Webb & Peart Reference WEBB and PEART2000) and partly with Bornean lower montane forest on Mt. Kinabalu (Aiba et al. Reference AIBA, KITAYAMA and TAKYU2004). The latter study reported that two of the four species studied showed a distribution independent of topographic association. However, in lower montane oak-laurel forest at Doi Inthanon, the distribution of all three Lithocarpus species studied was associated with habitat (Noguchi et al. Reference NOGUCHI, ITOH, MIZUNO, SRI-NGERNYUANG, KANZAKI, TEEJUNTUK, SUNGPALEE, HARA, OHKUBO, SAHUNALU, DHANMMANONDA and YAMAKURA2007). Niiyama et al. (Reference NIIYAMA, ABDUL RAHMAN, IIDA, KIMURA, AZIZI and APPANAH1999) also reported a significant habitat association for L. wallichianus from hill dipterocarp forest in Peninsular Malaysia. Niche partitioning is more likely to explain the coexistence of a high diversity of congeneric or confamilial species distributed in the same forest stand (Davies et al. Reference DAVIES, PALMIOTTO, ASHTON, LEE and LAFRANKIE1998, Yamada et al. Reference YAMADA, TOMITA, ITOH, YAMAKURA, OHKUBO, KANZAKI, TAN and ASHTON2006). However, Lithocarpus showed prominent topographic dependency at a site with low diversity at Doi Inthanon (0.2 species ha−1), whereas topographic association was independent of distribution at sites with higher diversity at Ulu Gadut (2.4 species ha−1). Coexistence of a large number of species independent of niche partitioning and coexistence in a non-equilibrium state are remarkable features of Lithocarpus in the Ulu Gadut study plot. This implies that, rather than contraction of the distribution over time due to higher survival in the preferred habitat, chance biotic interactions (e.g. the dispersal regime, or density- or frequency-dependent mortality from natural predation) may be important factors determining their distribution. The median distance (Table 3) of these Lithocarpus species also indicates that more intense disturbance by animals occurs at an early stage of development or before seed germination, consequently creating an indistinct distribution pattern in the forest stand. For these species, loss of seeds during secondary dispersal may be greater than in habitat-associated species with less widely dispersed seeds, but plants may be able to survive in a wider range of sites and habitats. However, more sophisticated studies, including seed dispersal observations and transplant experiments, are required in the future.
According to a previous study at Lambir (Palmiotto et al. Reference PALMIOTTO, DAVIES, VOGT, ASHTON, VOGT and ASHTON2004), Swintonia schwenkii and Hopea dryobalanoides showed significant associations with soil type, i.e. low-fertility udult ultisols and moderate-fertility humult ultisols, respectively, and the two species showed a distinct segregation pattern. However, the significantly overlapping distribution of the two species in the present study indicates that edaphic factors are less important for controlling their distribution pattern at Ulu Gadut. The present study, however, did not include edaphic features among the habitat variables. A number of studies have reported a strong association between tree distribution and soil type (Baillie et al. Reference BAILLIE, ASHTON, COURT, ANDERSON, FITZPATRICK and TINSLEY1987, Paoli et al. Reference PAOLI, CURRAN and ZAK2006, Phillips et al. Reference PHILLIPS, VARGAS, MONTEAGUDO, CRUZ, ZANS, SÁNCHEZ, YLI-HALLA and ROSE2003) or a combination of topography and soil type (Davies et al. Reference DAVIES, PALMIOTTO, ASHTON, LEE and LAFRANKIE1998, Harms et al. Reference HARMS, CONDIT, HUBBELL and FOSTER2001, Itoh et al. Reference ITOH, YAMAKURA, OHKUBO, KANZAKI, PALMIOTTO, LAFRANKIE, ASHTON and LEE2003, Svenning Reference SVENNING1999, Webb & Peart Reference WEBB and PEART2000, Yamada et al. Reference YAMADA, ZUIDEMA, ITOH, YAMAKURA, OHKUBO, KANZAKI, TAN and ASHTON2007). Therefore, consideration of edaphic factors should be a component of future investigations. It should also be noted that the limited area of the study site compared with that of other study sites in the tropics might influence the results, since it is unclear whether the distribution patterns of each species at the site are typical.
ACKNOWLEDGEMENTS
The authors thank the late Emeritus Professor Dr S. Kawamura (Kyoto University), Emeritus Professor Dr R. Ogushi (Kanazawa University), Professor Dr M. Rahman (former rector of Andalas University), and Dr A. Bakar (Head of Sumatra Nature Study Center, Andalas University) for their cooperation in this study under the Field Biology Research and Training Project in Sumatra; Emeritus Professor Dr H. Watanabe and his colleagues at Kyoto University are thanked for fruitful discussions; the staff of Herbarium Bogorience, especially Mr Wardi and Mr Agus, for assistance with plant identification; and Mr Irwan Jr., Pak. Mansul, Pak. Azi, Mr Edi, Mr Maison, Mr Nadi, Mr Deswendri, Mr Beni and Mr Oki for assistance with field work; Drs T. Shimamura and S. Nanami for instruction and assistance with spatial analysis; and Drs K. Abe, S. Kobayashi, M. Yanagisawa and K Tanaka for their support and encouragement. Part of this study was financially supported by the Japan International Cooperation Agency (1994–1997), Nippon Life Insurance Foundation (1995) and Global Environment Research Fund (S-2).
Appendix 1. Results of torus-translation tests for habitat association of 15 species across three developmental stages in the 3.12-ha plot in a mid-elevation hill dipterocarp forest in Ulu Gadut, West Sumatra. Four habitats were tested independently. Values represent the actual tree number observed in the field. Significant positive associations are denoted by ‘+’ and ‘++’ at significance levels of 5% and 1%, respectively. Significant negative associations are denoted by ‘-’ and ‘−’ at significance levels of 5% and 1%, respectively. The three developmental stages are: small sapling (SS), height < 50 cm; large sapling (LS), height ≥ 50 cm and dbh < 1 cm; and tree, dbh ≥ 1 cm. The small-sapling stages of Shorea atrinervosa and Shorea maxwelliana were not examined because of low numbers.
