INTRODUCTION
Most of the studies of floristic variation along edaphic and topographic gradients have shown differences in composition between broad environmental classes, such as uplands and valleys, or between clayey and sandy soils. In a review of the mechanisms that could maintain the high diversity of tropical forests, Wright (Reference WRIGHT2002) concluded that topographic and/or soil specializations could not maintain the high species diversity because there are few ‘environments’, which could act as niche compartments. A similar conclusion was reached for specialization in light use (Hubbell et al. Reference HUBBELL, FOSTER, O'BRIEN, HARMS, CONDIT, WECHSLER, WRIGHT and DE LAO1999), and the hypothesis of light regeneration niches (Denslow Reference DENSLOW1980) has lost favour in recent years (Brown & Jennings Reference BROWN, JENNINGS, Newbery, Prins and Brown1998). However, few studies searched for community patterns within the broad environmental categories (Lieberman et al. Reference LIEBERMAN, LIEBERMAN, HARTSHORN and PERALTA1985, Svenning Reference SVENNING1999, Vormisto et al. Reference VORMISTO, PHILLIPS, RUOKOLAINEN, TUOMISTO and VASQUEZ2000).
Distinct habitats could result from specialization for specific zones along continuous environmental gradients, which was Whittaker's original definition of beta diversity (Whittaker Reference WHITTAKER1972). Such specialization may be hard to detect because mass effects (Clark Reference CLARK, Guariguata and Kattan2002, Shmida & Wilson Reference SHMIDA and WILSON1985, Zobel Reference ZOBEL1997) tend to blur the species limits. However, when analyses are based on continuous variables, some surprising results may arise, such as evidence of light partitioning in the absence of gaps (Montgomery & Chazdon Reference MONTGOMERY and CHAZDON2002). Species-rich communities structured by segregation across hydrological gradients were observed in English meadows, even when the spatial variation in soil hydrological conditions thought to cause this occurred in the absence of any obvious topographic variation (Silvertown et al. Reference SILVERTOWN, DODD, GOWING and MOUNTFORD1999). Therefore, it is possible that variation within broad topographic or soil classes provides opportunities for specialization which have yet to be documented.
Riparian zones around streams (Bren Reference BREN1993) are one of the broad habitat categories recognized in tropical rain forests (Clark et al. Reference CLARK, PALMER and CLARK1999, Webb & Peart Reference WEBB and PEART2000), including terra firme forests of central Amazonia (Valencia et al. Reference VALENCIA, FOSTER, VILLA, CONDIT, SVENNING, HERNÁNDEZ, ROMOLEROUX, LOSOS, MAGARD and BALSLEV2004). Variation in the riparian vegetation associated with streams is expected as a response to soil saturation, which changes from the stream margins to the slopes (Gregory et al. Reference GREGORY, SWANSON, MCKEE and CUMMINS1992), and to the light regime. Light penetration is expected to be higher at the stream margins, and to decrease towards the slopes. However, riparian environments are complex, because flooding patterns, water and litter accumulation, and light availability to the understorey can change rapidly in space and time.
Patterns of vegetation zoning along margins are well-documented for large rivers. In Amazonian floodplains, plant composition varies with flood duration, which depends on terrain elevation (Ferreira Reference FERREIRA1997, Reference FERREIRA2000; Ferreira & Prance Reference FERREIRA and PRANCE1998, Junk & Piedade Reference JUNK, PIEDADE and Junk1997, Keel & Prance Reference KEEL and PRANCE1979, Salo et al. Reference SALO, KALLIOLA, HÄKKINEN, MÄKINEN, NIEMELÄ, PUHAKKA and COLEY1986, Worbes Reference WORBES and Junk1997). This strong gradient selects plants according to their life cycle duration, growth rate and wood density. It is not known, however, if the weaker gradients on the margins of small streams can play the same role as selective forces shaping community composition.
Many of the early studies of plant ecology in the tropics were concentrated on tree species (Gentry Reference GENTRY1988, Hubbell Reference HUBBELL1979). However, environmental variations relevant for understorey plants may not be important for canopy trees (Wiens Reference WIENS1989). More recently, there has been a growing interest in understanding the ecology of palms, shrubs and herbs (Duque et al. Reference DUQUE, SANCHEZ, CAVELIER and DUIVENVOORDEN2002, Poulsen & Balslev Reference POULSEN and BALSLEV1991, Svenning Reference SVENNING1999, Tuomisto & Ruokolainen Reference TUOMISTO and RUOKOLAINEN1994), and surveys of these groups can be conducted more rapidly than for trees. Shrub species composition was shown to be more correlated with light incidence and edaphic conditions than was canopy composition (Duque et al. Reference DUQUE, SANCHEZ, CAVELIER and DUIVENVOORDEN2002, Duque Montoya Reference DUQUE MONTOYA2001). Therefore, the understorey may respond to finer gradients than do trees. Riparian zones have been shown to contribute to between-habitat beta diversity (Sabo et al. Reference SABO, SPONSELLER, DIXON, GADE, HARMS, HEFFERNAN, JANI, KATZ, SOYKAN, WATTS and WELTER2005), but there is no guarantee that the width of the ‘riparian zone’ is similar for groups with different life forms.
In this paper, we investigate how terrestrial herb species composition varies from stream margins to the uplands, and how it varies within the riparian zone, from stream margins to the edge of slopes, in Reserva Florestal Adolpho Ducke, Amazonas, Brazil. We tested the hypotheses that composition differs between the riparian zone and uplands, and that species composition varies within the riparian zone along a gradient in water saturation, flooding and light availability associated with distance from the stream.
STUDY SITE
The study was conducted at Reserva Florestal Adolpho Ducke of the Instituto Nacional de Pesquisas da Amazônia (INPA), located 26 km north-west of Manaus, central Amazonia (2° 55′–3° 01′S, 59° 53′–59° 59W, Figure 1). The reserve covers 10 000 ha (10 km × 10 km) of terra firme tropical rain forest, with a closed canopy 30–37 m high and emergents growing to 40–45 m. The understorey is characterized by abundant sessile palms, such as Astrocaryum spp. and Attalea spp. (Ribeiro et al. Reference RIBEIRO, HOPKINS, VICENTINI, SOTHERS, COSTA, BRITO, SOUZA, MARTINS, LOHMANN, ASSUNÇÃO, PEREIRA, SILVA, MESQUITA and PROCÓPIO1999). Mean annual temperature at RFAD is around 26 °C and mean annual rainfall is 2362 mm, with a dry season between July and October (Marques-Filho et al. Reference MARQUES-FILHO, RIBEIRO, SANTOS and SANTOS1981).
Reserva Florestal Adolpho Ducke and plot distribution within the study site. Triangles represent riparian plots and squares represent upland plots.
Soils are derived from Cretaceous sediments from the Alter do Chão formation. Topography is an important determinant of soil formation in central Amazonia. Soils are clayey on the ridges, predominately formed by oxisols, constituted by kaolinite, iron oxide and gibbsite (Chauvel Reference CHAUVEL1982). The clay fraction decreases as elevation decreases, and lowlands have predominately hydromorphic podzolic sandy soils (Chauvel et al. Reference CHAUVEL, LUCAS and BOULET1987, Ranzani Reference RANZANI1980). The maximum altitudinal variation is about 87 m between the tops of ridges (max. observed elevation 115 m asl) and the lowlands (min. observed elevation 28 m asl). However, maximum variation between ridge tops and adjacent streams is generally only about 30 m.
METHODS
Sampling design
We installed forty 2 × 100-m plots that followed altitudinal contours, to minimize internal variation in elevation (Magnusson et al. Reference MAGNUSSON, LIMA, LUIZÃO, LUIZÃO, COSTA, CASTILHO and KINUPP2005). Twenty plots were installed on well-drained soils, in the higher portions of the topographic gradient, at a minimum distance of 1000 m from each other. The other 20 plots were installed in the riparian zone along the lower portions of the topographic gradient, with the main axis parallel to the streams, at a minimum distance of 600 m from each other.
Riparian plots were at distances up to 100 m from the Acará and Bolívia streams, which constitute the western drainage of the reserve. The headwaters of both streams are in the reserve, and they are up to 7 km long and third order within the reserve limits. Minimum distance among the forty upland and riparian-zone plots was 300 m. Distances of riparian plots to the stream margin were randomly assigned to 0, 25, 50 or 75% of the width (from the stream margins to the edge of slopes) of the valley around the stream. Percentages of width were used to describe the environmental gradient instead of actual distances because drainage patterns vary according to the size of the drainage area (Hodnett et al. Reference HODNETT, VENDRAME, MARQUES FILHO, OYAMA and TOMASELLA1997). This means that a plot 20 m from the stream margin is relatively closer to the margin when it is in a wide drainage area than when it is in a narrow one. Four plots per distance class were established. Plots in the 0% class were split in two subplots of 1 × 100 m, one on each side of the stream.
Herb community sampling
The herb community sampled in this study was restricted to the obligate terrestrial species (sensu Poulsen Reference POULSEN1996), i.e. those that germinate and spend their entire life cycle on the ground. Hemi-epiphytes and epiphytes fallen to the ground were not considered.
Herbs were sampled between July 2001 and July 2002 in the uplands and between February and September 2004 in the lowlands. All individuals greater than 5 cm in height were counted and identified. The varieties of Lindsaea lancea (var. lancea and var. falcata) and Calathea mansonis (var. 1 and var. 2) were treated as species, as they were morphologically distinct. For species with clonal growth (Poaceae and the Marantaceae genera Calathea and Ischnosiphon) in which individuals are difficult to separate, each clump at least 20 cm from another was counted as an individual.
Identifications were based on the field guide Flora da Reserva Ducke (Ribeiro et al. 1999), identification keys (Alston et al. Reference ALSTON, JERMY and RANKIN1981, Kramer Reference KRAMER1957, Mori et al. Reference MORI, CREMERS, GRACIE, DE GRANVILLE, HOFF and MITCHELL1997, Steyermark et al. Reference STEYERMARK, BERRY and HOLST1995, Tryon & Stolze Reference TRYON and STOLZE1989a, b; Tuomisto & Groot Reference TUOMISTO and GROOT1995, Windisch Reference WINDISCH1996) and comparison with material deposited in the Herbarium of the Instituto Nacional de Pesquisas da Amazonia (INPA). Voucher material was deposited in the INPA Herbarium.
Environmental variables
Elevation was measured by a professional topographic surveyor using a theodolite in all plots. Distance from stream margin was obtained with direct measurements with a measuring tape and compass for riparian plots and with a cartographic map for upland plots.
In each riparian plot, environmental variables were measured at five points (0, 25, 50, 75 and 100 m) along the main axis, and summarized as averages. Width of the riparian zone was measured with a clinometer, measuring tape and compass from the water margin to the edge of slope, the latter being defined as the point where elevation was 2 m above stream margin. Canopy openness was measured with a convex spherical crown densiometer (Model-A, Forestry Suppliers Inc.) at 1.0 m height. Water-table levels were measured once in each riparian plot at the end of the dry season in 50-mm-diameter dipwells augered to about 1 m below the ground surface, which was sufficient to reach the water table between 15 and 20 October 2004.
Data analysis
Plots were ordinated by their dissimilarity in species composition with non-metric multidimensional scaling (NMDS), to reduce dimensionality and allow the visualization of major patterns structuring the community. Ordinations were carried out with the whole set of 40 plots to detect differences in the herb community between the riparian zone and uplands, and separately using only the 20 plots installed in the riparian zone to detect community variation within the riparian zone. Two ordinations were carried out with the plant data for each group of plots (40 or 20), one based on quantitative data and another using presence-absence data. Quantitative ordinations used the Bray–Curtis distance measure on site-standardised (percentage of each species in each site) data. Ordination of quantitative data was used to capture the patterns displayed by the most abundant species, as these will have the greatest quantitative contribution to the differences between sites. Presence-absence ordination was performed using the Sørensen index. This ordination tends to capture the patterns of the rarer species, because the more abundant species generally occur in most sites and therefore contribute little to the differences between sites.
As results of the analyses for presence-absence and quantitative data revealed the same patterns in all cases, we only present results for qualitative data for the overall community, and quantitative data for plots in the riparian zone. Scores of the NMDS ordinations, which represent the major patterns in herb community composition, were used as dependent variables in models of univariate or multivariate regression to test for the effects of environmental variables.
Ordinations were done in PCord version 4.25 and inferential analyses with the statistical package Systat version 8.
RESULTS
Herb community composition from the stream margins to the uplands
The ground-herb community sampled along the complete topographic gradient of Reserva Ducke was composed of 75 species or morphotypes, distributed in 22 families and 8332 individuals (Table 1). Four morphotypes could not be identified to the species level. Nineteen species were pteridophytes (8 families), 18 were Marantaceae, 10 were Cyperaceae, 6 Poaceae, 5 Araceae and 17 species of 10 other angiosperm families. Most individuals (6195) and species (61 of 75) were recorded within the riparian zone, of which 29 species did not occur elsewhere. Thirty-two species occurred along the complete topographic gradient and 14 species were recorded only in the uplands. Thirteen species occurred in only one plot (9 in the riparian zone and 4 in the uplands).
Number of individuals and frequency of ground herbs in 20 riparian and 20 upland plots at Reserva Ducke, Manaus, Brazil. FR = frequency in riparian plots, TR = total number of individuals in riparian plots, FU = frequency in upland plots, NU = number of individuals in upland plots, TI = total number of individuals, summed over riparian and upland plots.
Ordination with NMDS captured 77.7% of the variation in the original distances between plots in one dimension, for qualitative data. There were strong curvilinear relationships between the one-dimensional NMDS ordination and distance from the stream (Figure 2a) and elevation (Figure 2b).
Herb species composition represented by one-dimensional NMDS ordination scores of qualitative data plotted against distance from stream (a) and elevation (b). U represents plots located in uplands and R represents plots located in the riparian zone.
Riparian and upland plots formed distinct groups in both figures. Herb-species qualitative composition, represented by the NMDS ordination scores, changed significantly (r2 = 0.75, P < 0.001) along the gradients of distance from stream and elevation. Most variation occurred at distances from stream < 100 m (Figure 2a) and elevations < 70 m (Figure 2b), indicating that species composition in riparian plots was more heterogeneous than in upland plots.
Herb community variation within the riparian zone
The ground-herb community sampled within the riparian zone was composed of 61 species or morphotypes, distributed in 20 families and 6195 individuals. Most individuals (3659) occurred in the four plots installed immediately adjacent to the stream (Table 2).
Number of individuals of ground herbs in 20 riparian plots in Reserva Ducke, Manaus, Brazil. Plots are grouped in five classes, defined as increasing proportions of distance between the stream and the upper limit of the riparian zone.
Ordination of quantitative data with NMDS captured 70.7% of the variation in the original distances between plots in one dimension. Community quantitative composition, represented by the ordination in one dimension, changed significantly with distance from the stream (r 2 = 0.77, P < 0.001). The best fit, however, was a negative logarithmic function of distance, which explained 82.3% of the variation in composition (P < 0.001; Composition = −1.71–0.506 × log(distance from the margin)).
Species quantitative composition turned over continuously with the distance from the margin, but one group of species was clearly restricted to the margins of streams (Table 2). The gradual change in species composition was evident when species composition ordinated by quantitative dissimilarity was plotted against the relative distances from the margin (Figure 3a), but not for absolute distances (Figure 3b).
Scores of a one-dimensional NMDS solution of riparian plots based on Bray–Curtis distances plotted against distance from the stream margin, expressed in classes of proportion of riparian-zone width (a) or absolute distance values (b). Relative distance from the margin is represented by the size of dots: the larger the size, the closer to the margin.
We hypothesized that canopy openness and water-table depth should be the main environmental factors varying with distance from the stream that affect community variation. Also, distance alone is possibly an indicator of the probability of flooding. Therefore, we tested a model to explain community composition that included canopy openness, water-table depth and the log-transformed distances from stream margin. The effect of water-table depth was significant (water-table depth: t = −3.04, P = 0.008), as was the effect of the log-transformed distances from stream margin (log(distance from margin): t = −8.98, P < 0.001). Canopy openness did not contribute significantly to the model (t = 0.58, P = 0.57). This model explained about 90% of the variation in community quantitative composition captured by the ordination (R 2 = 0.897, F3,15 = 43.4, P < 0.001).
DISCUSSION
Species composition of the herb community differed along the topographic gradient in Reserva Ducke. Most species and individuals occurred within the riparian zone, and species did not occupy the riparian zone homogeneously, but changed continuously in response to environmental gradients associated with distance from the stream.
Gradients of change in plant species composition associated with soil or topography have frequently been documented in the tropics, including the herbaceous vegetation at Reserva Ducke (Costa et al. Reference COSTA, MAGNUSSON and LUIZÃO2005). However, in general, the environment is modelled as broad classes (Poulsen & Tuomisto Reference POULSEN, TUOMISTO, Camus, Gibby and Johns1996, Tuomisto & Ruokolainen Reference TUOMISTO and GROOT1994, Tuomisto et al. Reference TUOMISTO, RUOKOLAINEN, AGUILAR and SARMIENTO2003). When the environment is forced into a few gross habitat categories, it may appear that there is less opportunity for specialization than when the environment is modelled as continuous multivariate dimensions. This study has shown that understorey herb composition of the riparian zone around streams is distinct from the upland areas for about 100 m from stream margins and to about 70 m asl in elevation, the exact distance depending on the size of the stream. However, the only assemblage that is almost completely distinct occurs as a narrow band, a few metres wide, along the streams. The rest of the riparian zone appears to represent an ecotone with continuous change, most of which occurs out to a distance of about half the width of the riparian zone as we defined it, based on topography.
Our data do not indicate that understorey herbs recognize a sharp limit of the riparian zone. We used an arbitrary limit of 2 m above the stream level, which in Reserva Ducke is sufficient to attain areas with a very low probability of flooding. Out to this limit, the species composition showed a continuous change in the direction of the composition of the upland areas as summarized by the multivariate ordination. Two metres above stream height is probably higher than many researchers would consider the limit of the riparian zone, but species composition, as represented by the NMDS axes, still had not converged on that of the upland plots.
The logarithmic decay in compositional similarity indicates that differentiation of species composition was stronger at the stream margins, suggesting stronger differences on the environment as well. In the field, we observed unpredictable flooding after heavy rain events that inundated stream margins, and submerged all herbs in plots adjacent to streams. Most plant species are not able to survive in inundated areas due to soil anaerobic conditions (Larcher Reference LARCHER2003). The margins of large rivers are usually occupied by habitat specialists (Kalliola & Puhakka Reference KALLIOLA and PUHAKKA1988, Salis et al. Reference SALIS, TAMASHIRO and JOLY1994). Restriction of species to the margins of small water courses of tropical forests is probably obvious to most observers, but has not previously been documented.
It is possible that adaptations to submergence imply competitive disadvantages in other environmental conditions. Cyclanthus bipartitus, Mapania pycnostachya, Pepinia sprucei and Urospatha sagittifolia were highly abundant at stream margins, but rare in other parts of the topographic gradient. Spathiphyllum maguirei and Thurnia sphaerocephala occurred exclusively on margins, suggesting that they cannot survive in other areas.
The edge of the slopes have greater chance of receiving propagules from herbaceous species established in higher portions of the topographic gradient due to gravity. Species observed at the edge of the slopes in the present study were found associated with steep slopes (Calyptrocarya poeppigiana and Danaea elliptica) and with clayey soils (Pleurostachys sparsiflora, Triplophyllum dicksonioides and Calathea cannoides) in the study by Costa et al. (Reference COSTA, MAGNUSSON and LUIZÃO2005) in Reserva Ducke.
To understand the riparian zonation with distance from the margin it is necessary to understand how different environmental factors vary along the distance gradient. It is likely that important environmental gradients correlated with distance from the stream margin are associated with hydrology. Further understanding of species distribution on riparian zones requires determination of how flooding patterns in response to rain events change subsurface and overland flows along the drainage profile in ways that might affect plants.
One study using data from seven continents and including taxa ranging from Antarctic soil invertebrates to tropical rain-forest lianas and primates (Sabo et al. 2005) found that riparian zones had lower alpha diversity than surrounding areas, but contributed to regional species richness because they harboured different species. In contrast, the riparian zones in our study had higher alpha diversity of herb species, as well as harbouring unique species. Also, heterogeneity within the riparian zone was as great as differences between the riparian zone and surrounding areas. We suggest that distributions of other plant and animal groups, specially those with small home ranges, such as ants, may also show strong variation within riparian zones. Riparian zones are critical to water-resource conservation, and are sensitive to changes in land use (Bren 1993). Although riparian zones are legally protected in Brazil, they are frequently degraded, indicating that the complex factors leading to zonation around streams need to be understood to effectively manage these areas.
ACKNOWLEDGEMENTS
This study was financed by the Brazilian National Research Council (CNPq), through the MSc degree postgraduate scholarship to Debora P. Drucker, the Long Term Ecological Research (PELD) grant to Flávio J. Luizão and grant No 472799/03-7 to William E. Magnusson. The Reserva Ducke ILTER/PELD site is a site of the Brazilian Programa de Pesquisa em Biodiversidade (PPBio). We thank João Batista, José da S. Lopes and José T. do Nascimento for field assistance, Fernanda A. de Carvalho for help in the identification of Pteridophyta and Maria de Lourdes Soares for help in the identification of Araceae. This study was possible due to the previous work conducted by the Projeto Flora da Reserva Ducke.
