INTRODUCTION
The high species diversity of tropical rain forests has fascinated biologists for more than six decades (Black et al. Reference BLACK, DOBZHANSKY and PAVAN1950). For example, 473 tree species with a diameter at breast height (dbh) ≥ 5 cm were recorded in a 1-ha plot in Amazonian Ecuador (Valencia et al. Reference VALENCIA, BALSLEV and PAZ Y MIÑO1994). Many different factors have been proposed to explain this extraordinary diversity (Chesson Reference CHESSON2000, Connell Reference CONNELL1978, Hubbell & Foster Reference HUBBELL, FOSTER, Diamond and Case1986, Wright Reference WRIGHT2002). One of the best supported (although not necessarily one of the most important) of these is niche differentiation associated with micro-topographic variation in drainage, moisture and nutrients (John et al. Reference JOHN, DALLING, HARMS, YAVITT, STALLARD, MIRABELLO, HUBBELL, VALENCIA, NAVARETTE, VALLEJO and FOSTER2007, Svenning Reference SVENNING1999, Yamada et al. Reference YAMADA, TOMITA, ITOH, YAMAKURA, OHKUBO, KANZAKI, TAN and ASHTON2006).
Ridges, slopes, flats and other topographic microhabitats in tropical rain forests have been shown to have distinct communities (Clark et al. Reference CLARK, CLARK and READ1998, Webb & Peart Reference WEBB and PEART2000). This phenomenon is caused by tree species being significantly associated with certain topographic positions (Clark et al. Reference CLARK, CLARK and READ1998, Harms et al. Reference HARMS, CONDIT, HUBBELL and FOSTER2001, Hubbell & Foster Reference HUBBELL, FOSTER, Diamond and Case1986). While obligate habitat restriction is rare, association with certain microhabitats appears common (Clark et al. Reference CLARK, CLARK and READ1998, Phillips et al. Reference PHILLIPS, NÚÑEZ, MONTEAGUDO, CRUZ, ZANS, SÁNCHEZ, YLI-HALLA and ROSE2003). Webb & Peart (Reference WEBB and PEART2000) also found that certain families seem to have affinities for certain microhabitats, while on the other hand niche partitioning has been detected among species in Bornean Sterculiaceae (Yamada et al. Reference YAMADA, TOMITA, ITOH, YAMAKURA, OHKUBO, KANZAKI, TAN and ASHTON2006) and Amazonian Myristicaceae (Queenborough et al. Reference QUEENBOROUGH, BURSLEM, GARWOOD and VALENCIA2007).
Islands present an interesting case, as they generally have biotas that are disharmonic and have lower species diversities than comparable continental habitats (Keppel et al. Reference KEPPEL, LOWE and POSSINGHAM2009, MacArthur & Wilson Reference MACARTHUR and WILSON1967). Compared with mainland habitats, this could result in larger realized niche space and hence lower habitat specificity (Roughgarden Reference ROUGHGARDEN1974). This would imply that communities in different topographic microhabitats should be less distinct. However, an alternative scenario would be that island species have similar niche width and habitat specificity as mainland species and, as a result, well-differentiated communities in different habitats with lower species diversity than mainland communities. In the Samoan archipelago, distinct ridge forest communities have been reported (Whistler Reference WHISTLER1980). Furthermore, forest structure and species composition have been found to vary considerably with respect to topography (Webb et al. Reference WEBB, STANFIELD and JENSEN1999).
In this study we investigate whether tree species composition varies among different microhabitats in a remote lowland rain forest on Viti Levu, Fiji. We (1) test whether different topographic habitats (ridge, slope and flats) in this island rain forest have distinct species compositions, (2) determine which species and families have distinct habitat affinities, and (3) compare our results with the continental island of Borneo and the oceanic island of Tutuila in Samoa.
METHODS
Study site
The Sovi Basin, located on the island of Viti Levu (10 388 km2) in the Fiji Group, covers some 200 km2, comprising lowland tropical rain forest (100–500 m asl) surrounded by mountain ranges (600–1300 m) in all directions (Figure 1). As a result, the basin has some of the most remote lowland rain forest in Fiji and has been protected in a partnership between local landowners, Conservation International and Fiji Water. The basin is composed of a mosaic of volcanic rocks that originated 20–40 Mya and transected by the Wainavobo and Wainivalu rivers, which join and exit as the Sovi River through a gorge in the east of the basin (Hirst Reference HIRST1965). Four climate stations adjacent to the basin report high annual precipitation, ranging widely between 3000 and 5000 mm y−1. This suggests that local topography greatly influences the amount of rainfall in the area. Although the Sovi Basin is currently (and has been for the last 100–200 y) uninhabited, it has a complex prehistoric settlement history, which is concentrated in the lower (eastern) and central parts of the basin. After European contact, the human population in Fiji and other Pacific Island countries decreased due to introduced diseases, and reduced warfare allowed the relocation of villages to more accessible locations (Bayliss-Smith Reference BAYLISS-SMITH and Ulijaszek2006). There is no evidence that the uppermost (western) reaches of the Sovi Basin were ever settled.
Location of study sites in the Sovi Basin in Fiji.
Data collection
During two surveys, the first from 5–17 May 2003 and the second from 14–20 March 2004 (totalling 20 d), we surveyed 19 plots (10 × 60 m in size) in different microhabitats (Table 1) throughout the Sovi Basin (Figure 1) on Viti Levu, Fiji's oldest (30–40 million y) and largest island. The four microhabitats differentiated in the field were flats (locations with a slope ≤10°), ridges (locations of higher elevation than the surrounding landscape and of a narrow (<30 m), linear form), ridge flats (like ridges but wider (usually ≥50 m)) and slopes (locations with a slope >10°). Plots were set up parallel to elevational contour lines, except for slopes (where plots were set up perpendicular to contour lines). All plots are remote (more than 6-h walk from the nearest settlement) and were selected by choosing homogenous old-growth (Clark Reference CLARK1996) forest in the mature phase (Martínez-Ramos et al. Reference MARTÍNEZ-RAMOS, SARUKHÁN, PIÑERO, Davy, Hutchings and Watkinson1988, Whitmore Reference WHITMORE1989). Two plots (site codes = DSF, WAS; see Table 1 for site codes) were only 10 × 30 m in size to ensure that they were located in homogenous forest and did not extend into different microhabitats.
Codes, characteristic features, plot sizes, tree species richness, percentage endemism (%E), basal area and density for all 19 study plots in the Sovi Basin. Topographic classes are defined in methods; average and standard deviation for species richness, basal area and density exclude DSF and WAS, which had a smaller plot size.
Within each plot species identity and dbh (measured at 1.4 m height) were recorded for each tree with dbh ≥10 cm. We also recorded the topographic microhabitat (flat, slope, ridge flat or ridge), angle of the slope and elevation for each plot. All tree species encountered were identified in the field. If a plant could not be identified, herbarium samples were collected and then deposited and identified at the South Pacific Regional Herbarium (SUVA). Identifications were carried out using Smith (Reference SMITH1979–1991) and Keppel & Ghazanfar (Reference KEPPEL and GHAZANFAR2006) and nomenclature follows these sources.
Data analysis
We calculated the total basal area of each tree (based on the measured dbh), which gives an indication of species dominance (Mueller-Dombois & Ellenberg Reference MUELLER-DOMBOIS and ELLENBERG2002), of each species and used the software R 2.10.1 (R-Development-Core-Team; http://cran.r-project.org/) for all analyses. We conducted Kruskal's non-metric multidimensional scaling (NMDS) based on the Bray–Curtis dissimilarity coefficient (B; if B = 0, sites have equal composition; if B = 1, sites have no species in common) of combined total absolute basal areas for each species in a plot using the metaMDS option to determine which sites shared similar species composition. Based on the NMDS plot, we excluded an outlying site (site code = DSF) from further analyses.
We then repeated the NMDS analysis without the excluded site and correlated the topographic microhabitat, slope and elevation of each site with the resulting NMDS plot using vector fitting (Dargie Reference DARGIE1984, Kantvilas & Minchin Reference KANTVILAS and MINCHIN1990), which allows quantification of the strength of relationships between environmental variables and species composition through the correlation coefficient (r 2). The significance of r 2 was calculated by producing P-values based on 1000 permutations. To facilitate vector fitting, we coded the topographic microhabitats based on inferred increasing drainage as follows: flat (as 1), slope (2), ridge flat (2.5) and ridge (3).
Based on the NMDS plot, we divided sites into three groups, which had different species composition and appeared to correspond to different drainage regimes: (1) ridges and steep (>30°) slopes (site codes = UWS, WAR, WAS, WVR, WVRS, WWR, WWS), (2) moderate (≤10–25°) slopes and ridge flats (site codes = AMRF, DERF, NORF, NOS, OGRF, UWRF, WDS, WNS), and (3) flats (site codes = UWF, WDF, WNF). We tested whether these three groups differed significantly in species composition using the multi-response permutation procedure (MRPP; Mielke et al. Reference MIELKE, BERRY and JOHNSON1976). To minimize the effect of chance occurrences of species in plots, we focused on species (genera and families) that occurred at least in half of the plots within at least one of the three groups to test which species were strongly associated with particular microhabitats. For these selected taxa we conducted an indicator species analysis (Dufrêne & Legendre Reference DUFRÊNE and LEGENDRE1997) to test the significance of association with a particular habitat.
RESULTS
Species composition
In the 19 plots, which totalled 1.08 ha in area, we recorded 1003 trees of 112 species in 76 genera and 46 families (Appendix 1). Syzygium (Myrtaceae) was the most diverse genus with seven species, followed by Calophyllum (Clusiaceae) and Palaquium (Sapotaceae) with five, and Myristica (Myristicaceae) with four species. The most species-rich families were the Clusiaceae, Euphorbiaceae and Sapotaceae with eight species each, followed by the Myrtaceae with seven species. Forty-six to 70% (mean = 60%) of species per plot were endemic to Fiji (Table 1). Basal area in the 10 × 60-m plots (Table 1) ranged between 1.42 and 6.09 m2 (mean = 3.24 m2) and density between 33 and 95 trees (mean = 61 trees). Although the site with the highest density also had the highest basal area (Table 1), there was no significant correlation (r 2 = 0.063, P = 0.250) between the two variables.
NMDS including all sites produced three clusters, differentiated based on species composition. One of the flats studied (DSF) was an extreme outlier and formed a unique vegetation type, dominated by two species, Retrophyllum vitiense (Podocarpaceae) and Calophyllum vitiense (Clusiaceae). Because the species composition of this site was very different from any of the other plots analysed, it was excluded from other analyses.
The NMDS plot of the reduced data set was very similar to that of the complete one (Figure 2) and produced the same three clusters of plots: ridges and steep (>30°) slopes (seven well-drained sites; left-hand side of Figure 2), moderate (≤20°) slopes and ridge flats (eight plots with moderate drainage; centre) and flats (three plots with poor drainage; right-hand side). MRPP confirmed these groups to be distinct communities (A = 0.1127, P < 0.001). Topographic position (r 2 = 0.532, P = 0.005) was the only significant explanatory vector for species composition (as represented by the NMDS plot). The slope (r 2 = 0.205, P = 0.172) and elevation (r 2 = 0.045, P = 0.590) of a plot were not significant.
Non-metric multidimensional scaling (NMDS) for 18 study plots (DSF excluded) in the Sovi Basin, Fiji, based on basal areas of trees with dbh ≥ 10 cm. Last letter (s) of site codes indicate topographic location (R = ridge, RF = ridge flat, S = slope, F = flat). See Table 1 for site codes and details of plots. Stress = 18.5, two convergent solutions found after three attempts. Fitted onto this is the only significant explanatory variable topographic microhabitat (Topo; r 2 = 0.532, P = 0.005).
Well-drained plots (ridges and steep slopes) were dominated by Palaquium hornei (Sapotaceae), Agathis macrophylla (Araucariaceae), Dacrydium nidulum (Podocarpaceae) and Haplolobus floribundus (Burseraceae). Moderate slopes were dominated by Gonystylus punctatus (Thymelaeaceae), Endospermum macrophyllum (Euphorbiaceae), Myristica gillespieana (Myristicaceae), Parinari insularum (Chrysobalanaceae) and Syzygium fijiense (Myrtaceae). Members of the Verbenaceae (Viticipremna vitilevuensis, Premna protusa, Gmelina vitiense) dominated the flats, with Dillenia biflora (Dilleniaceae) and Dysoxylum richii (Meliaceae) being other important components of this community (Table 2).
The 10 species with the greatest combined basal areas (m2, stated after species name) in the three topographic microhabitats for 18 study plots (DSF excluded) in the Sovi Basin, Fiji, including trees with dbh ≥ 10 cm. See Table 1 for site codes and details of plots and Appendix 1 for status (endemic/indigenous). RS (7 plots) = ridges and steep (>30°) slopes, MS (8 plots) = moderate (≤20°) slopes and ridge flats, FL (3 plots) = flats.
Habitat associations
Forty-one species occurred in at least half the plots of at least one particular microhabitat (Table 3). Of these species 23 (56%) showed significant habitat association at P ≤ 0.05. The remainder did not return significant P-values in the indicator species analysis. Of the 23 species that showed habitat association, 14 were associated with poorly drained habitats (categories FL and FL/MS in Table 3), five with well-drained habitats (categories RS and RS/MS in Table 3), and four with sites of intermediate drainage (category MS in Table 3).
Proportion of sample plots occupied and habitat associations of species, genera and families for 18 study plots (DSF excluded) in the Sovi Basin, Fiji, including trees with dbh ≥ 10 cm. See Table 1 for site codes and details of plots. The column for habitat association states the habitat for a taxon and the associated probability (P-value) obtained from the indicator species analysis. If values for one taxon also apply to higher taxonomic levels, taxa are listed with a slash (/). Taxa higher than species level are indented in the first column. RS (7 plots) = ridges and steep (>30°) slopes, MS (8 plots) = moderate (≤20°) slopes and ridge flats, FL (3 plots) = flats. NS = not significant in the indicator species analysis.
A few genera and families also showed association with certain habitats (Table 3). The families Burseraceae, Podocarpaceae and Sapotaceae, and the genera Calophyllum and Palaquium were significantly associated with well-drained habitats (categories RS and RS/MS in Table 3), while the families Lauraceae and Verbenaceae and the genera Dysoxylum and Macaranga (Euphorbiaceae) were associated with poorly drained habitats (categories FL and FL/MS in Table 3). The genus Myristica was associated with sites of intermediate drainage (category MS in Table 3).
DISCUSSION
Species richness and endemism of trees with dbh ≥ 10 cm in the lowland rain forest of the Sovi Basin are high. Values are similar to the total value reported for four 50 × 50-m plots (total = 1 ha) in Savura (124 species, 54.1% endemism), located about 25 km south-east of the Sovi Basin on the same island (Keppel et al. Reference KEPPEL, BUCKLEY and POSSINGHAM2010). This suggests that Fiji's lowland tropical rain forests have high diversity (about 100 tree species ha−1) and endemism (50–60%), compared with other Western Polynesian archipelagos (Keppel et al. Reference KEPPEL, BUCKLEY and POSSINGHAM2010, Webb & Fa'aumu Reference WEBB and FA'AUMU1999, Webb et al. Reference WEBB, STANFIELD and JENSEN1999, Reference WEBB, VAN DE BULT, CHUTIPONG and KABIR2006). Although some of these studies in the Samoan archipelago used somewhat different plot sizes, the total sample sizes are comparable and indicate a diversity of about 30–40 tree species ha−1).
Species composition appears to be very similar to the Savura site, with species of Myristica, Calophyllum, Garcinia, Syzygium, Palaquium and Gonystylus punctatus dominating in both locations (Keppel et al. Reference KEPPEL, NAVUSO, NAIKATINI, THOMAS, ROUNDS, OSBORNE, BATINAMU and SENIVASA2005). It therefore appears that a group of the same species (an oligarchy) dominates large stretches of lowland tropical rain forest on Viti Levu island. This phenomenon has also been found in Amazonian rain forests (Duivenvoorden Reference DUIVENVOORDEN1995, Pitman et al. Reference PITMAN, TERBORGH, SILMAN, NÚÑEZ, NEILL, CERÓN, PALACIOS and AULESTIA2001, Svenning et al. Reference SVENNING, KINNER, STALLARD, ENGELBRECHT and WRIGHT2004) and has been attributed to low environmental heterogeneity (Pitman et al. Reference PITMAN, TERBORGH, SILMAN, NÚÑEZ, NEILL, CERÓN, PALACIOS and AULESTIA2001). The topography and soils of lowland rain forests in south-west Viti Levu are indeed mostly similar, being composed of a (sometimes steeply) undulating landscape on red-brown clay soils of volcanic origin.
Our results show that more than 50% of all common species have clear habitat associations. This value is higher than the 35% (17 of 49 species using Monte Carlo randomizations) that showed significant association with topographic habitat in Borneo (Webb & Peart Reference WEBB and PEART2000) but similar to that for Samoa's 53% (9 of 17 species with 10 individuals or more using chi-squared tests). Similar to other studies in mainland rain forests (Clark et al. Reference CLARK, CLARK and READ1998, Phillips et al. Reference PHILLIPS, NÚÑEZ, MONTEAGUDO, CRUZ, ZANS, SÁNCHEZ, YLI-HALLA and ROSE2003) and in Samoa (Webb et al. Reference WEBB, STANFIELD and JENSEN1999), we found that obligate habitat association is rare but that many species show a significant tendency towards habitat association. Saurauia rubicunda (Actinidiaceae) and Dacrydium nidulum were the only common species restricted to a particular physiographic habitat in this study. While the former may indeed be mostly restricted to creek banks, the latter commonly occurs in habitats other than ridges and slopes outside the lowland tropical rain forest (Keppel et al. Reference KEPPEL, ROUNDS and THOMAS2006, Keppel & Tuiwawa Reference KEPPEL and TUIWAWA2007).
Considering that we only studied habitat associations in relation to topographic variation, which we assume to roughly correspond to soil drainage, and did not investigate other important factors such as soil nutrients (John et al. Reference JOHN, DALLING, HARMS, YAVITT, STALLARD, MIRABELLO, HUBBELL, VALENCIA, NAVARETTE, VALLEJO and FOSTER2007, Palmiotto et al. Reference PALMIOTTO, DAVIES, VOGT, ASHTON, VOGT and ASHTON2004), our study provides conservative evidence for habitat partitioning of rain-forest trees and hence for ecological determinism playing an important role in facilitating co-existence (Svenning et al. Reference SVENNING, KINNER, STALLARD, ENGELBRECHT and WRIGHT2004). Although our study does not permit direct comparisons because of different plot sizes and analyses, the realized niches of tree species in the island rain forest of this study and those in Samoa (Webb et al. Reference WEBB, STANFIELD and JENSEN1999) do not appear to be broader than on continental Borneo (Webb & Peart Reference WEBB and PEART2000). In all three studies (Webb & Peart 2009, Webb et al. Reference WEBB, STANFIELD and JENSEN1999, this study) a similar distinction of well-drained ridges/steep slopes, moderate slopes and flats was observed. The commonness of habitat specialization in Fijian and Samoan rain forest (Webb et al. Reference WEBB, STANFIELD and JENSEN1999) also implies that competition amongst tree species may play an important role in Pacific Island rain forest and cautions against generalizations about islands having lower levels of competition.
We detected some similarities at higher taxonomic levels (genus and family) between our results and those from Bornean (Webb & Peart Reference WEBB and PEART2000) and Samoan (Webb et al. Reference WEBB, STANFIELD and JENSEN1999) rain forests. While comparisons are difficult because of different methods and analyses employed, Calophyllum and Palaquium were found to be primarily associated with well-drained habitats in all three studies. This may suggest that taxa retain their niches when colonizing islands, which are often assumed to have more vacant niche space, and hence that niche conservatism plays a role in the assembly of island communities.
While Fiji lowland rain-forest communities of ridges and slopes have been previously described (Keppel et al. Reference KEPPEL, NAVUSO, NAIKATINI, THOMAS, ROUNDS, OSBORNE, BATINAMU and SENIVASA2005), the communities on stream flats have not. This community is dominated by distinct species, especially in the Verbenaceae, and forms a physiographic habitat type defined by having poorer drainage, experiencing brief, occasional floods after heavy downpours and as a result receiving regular deposits of alluvial materials. Because such stream flats are highly fertile and because human disturbance is often concentrated around watercourses, this community is likely endangered and good exemplars were restricted to the upper reaches of the Sovi Basin.
This study shows Fiji to be highly diverse compared with other Polynesian Pacific islands and provides some initial evidence for habitat partitioning among species in insular environments. While significant associations with topographic microhabitats are demonstrated for half of all common tree species, more detailed studies are required to quantify the strength of these associations. This would facilitate comparing the sizes of realized niches on islands and the mainland and an assessment, if lower species diversity does lead to broader realized niches.
ACKNOWLEDGEMENTS
The field surveys were part of the Pacific-Asia Biodiversity Transect (PABITRA) and the Sovi Basin biodiversity assessment. We would like to thank the mataqalis that own land in the Sovi Basin for permitting us to enter and study on their land and for assisting us with numerous logistic aspects of our study. We are especially indebted to those landowners who accompanied us in the field and assisted with data collection. We would also like to thank Mosese Moceyawa, Nunia T. Thomas and Baravi Thaman for their assistance in the data collection. Figure 1 was produced by Sala Curuki of the National Trust for Fiji, incorporating data provided by the Department of Lands, Suva, Fiji. Janet Franklin of Arizona State University provided many useful comments on a draft of this paper.
Total basal area, family and status (e = endemic; i = indigenous) for all tree species with a dbh ≥ 10 cm in seven plots on ridges and steep (>30°) slopes RS, eight plots on moderate (≤20°) slopes and ridge flats (MS), three plots on flats (FL) and one outlying plot on a flat (DSF), all located in the Sovi Basin. See Table 1 for site codes and details of plots.
