• Amazonia;
  • beta diversity;
  • chronological clustering.;
  • floristic composition;
  • Mantel test;
  • Melastomataceae;
  • Peru;
  • pteridophytes;
  • random walk model;
  • satellite imagery;
  • tropical rain forest


  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  • 1
    The floristic variation in Amazonian lowland forests is poorly understood, especially in the large areas of non-inundated (tierra firme) rain forest. Species composition may be either unpredictable as abundances fluctuate in a random walk, more-or-less uniform, or it may correspond to environmental heterogeneity.
  • 2
    We tested the three hypotheses by studying the floristic variation of two phylogenetically distant plant groups along a continuous 43-km long line transect that crossed tierra firme rain forest in northern Peru.
  • 3
    The observed floristic patterns were compared to patterns in the spectral reflectance characteristics of the forest as recorded in a Landsat TM satellite image. The topography of the transect was measured in the field, and surface soil samples were collected to document edaphic conditions. The two plant groups, pteridophytes and the Melastomataceae, were assessed in 2-m wide and 500-m long sampling units.
  • 4
    Floristic similarity (Jaccard index) between sampling units ranged from 0.01 to 0.71 (mean = 0.27), showing that some units were almost completely dissimilar while others were very alike.
  • 5
    Spatially constrained clustering produced very similar subdivisions of the transect when based separately on satellite image data, pteriophytes, and Melastomataceae, and the subdivisions were also related to topography and soil characteristics. Mantel tests showed that floristic similarity patterns of the two plant groups were highly correlated with each other and with similarities in reflectance patterns of the satellite image, and somewhat less correlated with geographical distance.
  • 6
    Our results lend no support to the uniformity hypothesis, but they partially support the random walk model, and are consistent with the hypothesis that species segregate edaphically at the landscape scale within the uniform-looking forest.


  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References

hypotheses on floristic patterns

In the past few decades, researchers have become increasingly interested in documenting and understanding the spatial structure and species composition of Amazonian lowland rain forests. There are a range of different views concerning the main factors that control plant species distributions in these forests, and the kind of general distribution patterns that follow. The ecological and floristic differences between such contrasting habitats as inundated vs. non-inundated forests, or forests on podzolized white sand soils vs. non-podzolized soils, are universally recognized. However, researchers differ fundamentally in their views on how species are distributed within these broad forest categories, especially within the non-inundated forests on non-podzolized soils, i.e. the ‘typical’ tierra firme rain forest.

Much of the current discussion is centred around the relative importance of three alternative views: (i) plant species are competitively equal, and the local species composition is in a state of ‘random walk’ as a result of local immigration and extinction; (ii) the forest is essentially homogeneous, and a small proportion of the species are competitively superior and dominate the forest over wide areas; and (iii) differences in soils within the forest are distinct enough to favour different species at different locations, and thus create numerous floristically differentiated forest patches.

If the plant species in a community come and go at random, as in the first view (Hubbell & Foster 1986; Condit 1996; Hubbell 1997, 2001; Chave et al. 2002; Condit et al. 2002), the variation in abundance of a given species is expected to show strong spatial autocorrelation due to dispersal limitation, but not to be systematically correlated with the abundances (or even presence) of other species. As a result, a species that is abundant at one site is likely to be both present and abundant at nearby sites but not at faraway sites, and the overall floristic similarity among sites will decrease monotonically with increasing inter-site distance. This decrease in floristic similarity is expected to be approximately linearly related to the logarithm of geographical distance (Hubbell 2001; Condit et al. 2002). The species are more or less equivalent ecologically, so any one of them can become abundant, rare, or locally extinct by chance, and floristic patterns are therefore not expected to correlate with patterns in local site conditions. Furthermore, one would not expect to find sharp floristic boundaries where several species appear and/or disappear simultaneously. Instead, species turnover would be gradual in space.

The second view maintains that species composition and abundance patterns are relatively constant over wide areas, although only a few species may be shared between sampling plots if plot size is small (Duivenvoorden 1995; Pitman et al. 1999, 2001; Terborgh et al. 2002), and that, following disturbance, species composition and abundance revert towards their prior state (Terborgh et al. 1996). Most species are expected to be widespread, and the species that become abundant at a given site are not a random subset of all the species present, but are likely to belong to a limited group of species that possess biological characteristics that enable them to compete successfully and dominate over large tracts of forest (Pitman et al. 1999, 2001). These dominant species are expected to be omnipresent in the forest, at least at the landscape scale, so their abundance patterns are not expected to show either spatial autocorrelation or correlation with patterns in local site conditions. Because the forest is considered homogeneous, no abrupt turnover zones in species composition are expected.

The third view maintains that the environmental variation in western Amazonia is pronounced enough to create floristically differentiated communities within the tierra firme forest. Spatial variation in species composition is expected, both in response to landscape-scale soil differences and in relation to local factors such as topography and associated soil catenas (Poulsen & Balslev 1991; Tuomisto et al. 1995; Ruokolainen et al. 1997; Svenning 1999; see also Lieberman et al. 1985 and Clark et al. 1995, 1998, 1999 for Central America, and Sabatier et al. 1997 for Guiana). Species abundances and floristic composition are expected to reflect spatial patterns in the environmental conditions, so that if there is spatial turnover in these patterns, corresponding and predictable turnover is also expected in the vegetation (Tuomisto & Poulsen 1996, 2000). In a patchy environment, spatial autocorrelation is expected to be a poor predictor of similarity in species composition because environmentally, and hence floristically, dissimilar sites can occur in close proximity (Poulsen & Tuomisto 1996; Ruokolainen et al. 1997). Because each species is expected to be most abundant where the environmental conditions are most favourable for it, the same dominant species are expected at sites with similar environmental conditions, while different dominants are expected at sites with differing environmental conditions (Tuomisto et al. 1998).

Similar discussion about the detailed variation in species composition within broadly defined vegetation types has been conducted elsewhere in the tropics, especially in south-east Asia, where field results have also been variously interpreted (Baillie et al. 1987). Poore (1968) concluded that, while rare species may be habitat specialists, the distribution of common species is determined by biotic interactions and chance. Others have found evidence for mainly edaphically determined distribution patterns (Austin et al. 1972; Ashton 1976; Baillie et al. 1987), or evidence of such patterns in some but not all of the forest types studied (Newbery & Proctor 1984).

landscape-scale sampling

To test the three hypotheses described above, it is necessary that field inventories are both extensive enough and detailed enough to reveal landscape-scale floristic and edaphic patterns. Extensive tree sampling has been carried out in western Amazonia (Gentry 1988; Duivenvoorden 1995; Ruokolainen et al. 1997; Ruokolainen & Tuomisto 1998; Pitman et al. 1999, 2001; Duque et al. 2002), but both sampling and species identification of trees are very laborious and time-consuming, and the number of species involved is very high, so it is difficult to obtain tree samples that are both spatially and floristically representative enough to give a detailed and reliable picture of landscape-scale species distribution patterns.

To be able to cover larger spatial extents in more detail, we have concentrated our sampling effort on two plant groups that are more easily observable and less species-rich than canopy trees: pteridophytes (ferns and fern allies) and the Melastomataceae (which are mainly shrubs and small trees). In earlier studies, both groups have been found to show roughly the same floristic patterns as trees (Ruokolainen et al. 1997; Ruokolainen & Tuomisto 1998; Vormisto et al. 2000), and hence we call them here indicator groups.

Because our field sampling did not cover trees, we wanted to preclude the possibility that the indicator groups chosen conform unduly with each other because of phylogenetic relatedness or similarities in life histories. We therefore studied groups that are both phylogenetically remote (pteridophytes vs. angiosperms) and have contrasting life histories and dispersal modes (pteridophytes have wind-dispersed spores and a sessile self-supporting gametophyte generation, while Melastomataceae are predominantly bee-pollinated and bird-dispersed). Any congruence in species composition patterns between such dissimilar groups is likely to reflect external factors that would also affect other plant groups.

Because Amazonian rain forests are spatially extensive and difficult to access, remotely sensed information has been used to help in recognizing plant communities and spatial patterns within them. Early successional forest, inundated forest, different kinds of swamp and forest on podzolized white sand soil, which are all clearly structurally different, can be readily recognized in satellite images (Kalliola et al. 1991, 1998; Tuomisto et al. 1994; Foody & Hill 1996; Novo & Shimabukuro 1997; Tuomisto 1998; Hill 1999; Saatchi et al. 2000). Even non-inundated tierra firme forests on non-podzolized soil, which mostly look homogeneous in aerial photographs, show considerable spectral patchiness in Landsat TM (Thematic Mapper) satellite images with variation at scales from hundreds of metres to kilometres (Tuomisto et al. 1995). It has been debated whether or not these satellite image patterns indicate differences in the vegetation that are related to soil differences (Condit 1996; Duivenvoorden & Lips 1998), but recent results have indicated that this indeed is the case (Ruokolainen & Tuomisto 1998; Tuomisto & Ruokolainen 2001; Tuomisto et al. 2003a).

We used a Landsat TM satellite image as a source of spatially continuous information for landscape-scale variation in the rain forest. The satellite measures reflectance of the ground cover, which in this case is mainly determined by the canopy trees, lianas and epiphytes. On the basis of earlier studies (Tuomisto et al. 1995; Ruokolainen & Tuomisto 1998; Vormisto et al. 2000; Tuomisto & Ruokolainen 2001; Tuomisto et al. 2003a) we propose that canopy patterns follow underlying edaphic conditions, and hence we use the reflectance values from the satellite image as a proxy for environmental variation. Soil samples were analysed to verify this relationship, but their spatial resolution was not sufficient to include them in the formal analyses.

Note that our indicator groups are understorey plants, and therefore have hardly any direct influence on the reflectance characteristics of the forest. Consequently, a correlation between understorey species distribution patterns and canopy reflectance patterns can only be found if the factors that determine the reflectance characteristics of the canopy are strongly correlated with those factors that determine the floristic composition of the understorey. This puts the hypothesis on environmental control of floristic patterns under a stringent test, while favouring the acceptance of the random walk hypothesis and the uniformity hypothesis.


We have recently tested the three hypotheses using widely spaced field sampling that ranged from southern Peru to Ecuador and Colombia (Tuomisto et al. 2003b). At such a wide spatial scale, the forests were clearly not homogeneous, and the data indicated that both random walk with dispersal limitation, and environmental factors are needed to explain floristic patterns. Condit et al. (2002) found that the dispersal limitation model was sufficient to explain their field data at distances between 0.2 km and 50 km. In the present study, our aim is to concentrate on this landscape scale, and to test the three hypotheses using data from a continuous 43-km long transect.

A single transect was used instead of discrete plots because continuous sampling allows observations of spatial change to be made and compared between data sets, and assessment of whether turnover is spatially continuous or occurs more rapidly at certain points. A transect can also be georeferenced more readily than separate plots, because it crosses tree-fall gaps where GPS (Global Positioning System) coordinates can readily be obtained. Our field survey extended over 43 km of forest, and the analyses of both field data and satellite image data were based on 500-m long sampling units. This geographical scale is such that it is able to detect patches of the size recognized by Tuomisto et al. (1995), who assessed spectral patchiness along 30-km long transects that were drawn on satellite images but not field-verified. The relatively coarse resolution further makes it unlikely that any correlation between patterns in satellite imagery and plant species composition is caused by ordinary forest succession in tree fall gaps, because these are typically much less than 500 m across.

We also asked how many of the observed plant species are actually distributed in a way that correlates with the reflectance patterns in satellite imagery. This question was answered by first classifying the sampling units of the transect on the basis of information from the satellite image, and then testing, for each species, whether or not its distribution was biased towards any of the recognized classes.

Materials and methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References

data collection

Fieldwork was carried out in Amazonian Peru in the forest reserve of the Amazon Center for Environmental Education and Research (ACEER) close to the confluence of the Sucusari and Napo rivers (Fig. 1). The climate in the area is tropical, humid and almost aseasonal. Mean monthly temperature in the nearby city of Iquitos is 25–27 °C throughout the year, and annual precipitation is about 3100 mm. No month receives less than 180 mm of rain on average, but about half of the years for which records exist experienced one or two months with less than 100 mm of rain (Marengo 1998).


Figure 1. Location of the study area in Peruvian Amazonia, and of the 43-km long transect near the confluence of the Napo and Sucusari rivers. The transect first runs 30 km to the east, and then 13 km to the north.

Download figure to PowerPoint

The area is about 100–200 m above sea level, and the landscape ranges from flat to hilly. The surface soil is formed of unconsolidated sediments of various origins and ages, including the mid-Miocene Pebas formation and more recent fluvial deposits. The geology of the area has not been studied in detail, but accounts of the geological history of the general region do exist (Hoorn 1993; Räsänen et al. 1998).

The vegetation in the study area consists mainly of closed-canopy non-inundated forests, although seasonally or sporadically inundated zones are found along all rivers and major creeks. No special edaphically defined vegetation types (such as forests on white sand soils) are known from the non-inundated area.

Soil and floristic studies were conducted in a continuous 43.38-km long transect. The transect followed the border of the ACEER reserve, which had been marked in the field a few months earlier by a crew of men from nearby villages, some of whom also joined us on this field expedition. The transect was georeferenced using GPS technology.

A visible mark was fixed every 50 m along the length of the entire transect. Sections of 100 m were used as sampling subunits for the floristic inventory. Pteridophytes and members of the family Melastomataceae (excluding Memecylaceae) were censused within an estimated 2.0 m to the left side of the transect. Presence-absence data were collected for each species. Collecting abundance data would have been too slow to be practicable; as the length of our field expedition was mainly limited by the quantity of provisions we were able to carry, we had to compromise between local detail and spatial extent.

For the purposes of the present paper, five consecutive subunits were fused to obtain 87 sampling units with an effective size of 0.1 ha (500 × 2 m). The last unit ended at the shore of the Apayacu River, and was shorter than the other units. The vegetation in the first sampling unit (the one closest to the Sucusari River) showed, in parts, characteristics typical of secondary forests. Because our purpose was not to study differences between secondary and old-growth forests, but rather variation within old-growth forests, this unit was excluded from the analyses presented. The final sample size was therefore 86 sampling units that covered almost 8.6 ha.

To facilitate pteridophyte observation and sampling, only individuals with at least one leaf longer than 10 cm were considered, and epiphytic and climbing individuals whose lowermost green leaves were higher than 2 m above ground were ignored. Voucher collections for all species of both pteridophytes and the Melastomataceae are deposited in herbaria in Peru (AMAZ), Finland (TUR) and USA (KSP; herbarium acronyms follow Holmgren et al. 1990).

The topographic profile of the transect was measured using a clinometer (Suunto, Vantaa, Finland). Measurements were taken every 50 m, and in between if the slope of the terrain changed significantly. Surface soil samples (the top 5 cm of the mineral soil) were collected at roughly 2.5-km intervals, such that two samples were taken from each location, one from the top of a hill and one from the bottom of the nearest valley. Each of the soil samples consisted of five pooled subsamples collected within an area of c. 5 × 5 m. Physical and chemical analyses were carried out using standard procedures (van Reeuwijk 1993). We report soil texture (percentage of coarse sand with particle size 0.25–2 mm) and the concentration of exchangeable bases (calcium, potassium, magnesium and sodium measured in 1 m NH4OAc at pH 7).

A Landsat TM image (path 6, row 62, 1 November 1987) covering the study area was obtained from the Landsat Pathfinder HTF project of the University of Maryland and NASA, USA. After the field work, the satellite image was rectified using ground control points that had either been obtained using a GPS in the field, or could be identified on a base map derived from Landsat MSS satellite images (IFG 1984). The transect was drawn on the rectified satellite image with the help of GPS coordinates and landmarks.

For each of the 500-m long sampling units, an area extending 200 m to either side was delimited on the satellite image. Such a large area was used for two reasons. First, it reduced the effect of hilliness on the results, as each unit was large enough to average out the differences in reflectance values between the sunlit and shaded sides of hills. Second, the error in GPS coordinates at the time of sampling may have exceeded 100 m, so a 200-m buffer in pixel sampling was deemed necessary to ensure that the transect was actually contained within the sampled area.

The original values for the pixels included within each of the delimited areas were extracted for analysis. Most of the areas had 211–218 pixels, but four were made smaller (135–180 pixels) to avoid including pixels with clouds or cloud shadows. Only data from bands 1–5 and 7 were used, the thermal infrared band 6 being excluded. ER-Mapper 5.5 software (Earth Resource Mapping, Egham, UK) was used for all satellite image analyses.

data analyses

Our numerical analyses aimed to reveal areas of major changes in pteridophyte and Melastomataceae species composition, and to clarify the extent to which differences in environmental factors (as inferred from pixel values in the satellite image) and geographical distance can be used to predict differences in floristic composition of the forest.

Almost all analyses are based on resemblance matrices, each of which consists of pairwise comparisons among all 86 sampling units using one or more descriptor variables. When a resemblance measure that had originally been calculated as distance (D) needed to be converted to similarity (S) or vice versa, the formula S = 1 − D was used.

Floristic similarity matrices were calculated using the Jaccard index [S = a/(a + b + c), where a is the number of species shared between the two sampling units, b is the number of species only found in the first unit, and c is the number of species only found in the second unit]. Three similarity matrices were calculated: one using pteridophytes, one using Melastomataceae and one using both plant groups combined.

Differences in pixel values between sampling units were expressed in Euclidean distance using mean pixel values calculated separately for each band for each unit. A total of 11 Euclidean distance matrices were calculated using one or more of these satellite-derived variables. Seven matrices were based on a single variable: either the mean pixel values of one of the six bands, or the green vegetation index [NDVI = (band 4 − band 3)/(band 4 + band 3)]. For the remaining matrices, two different combinations of satellite-derived variables were used [all six bands (1–5 and 7), or band 2, band 7 and NDVI]. The purpose of including just three variables in the latter case was to build a model with reduced collinearity. The visible wavelength bands 1–3 are highly intercorrelated, as are the infrared bands 4, 5 and 7, so only one band from each group was used in the reduced model. Bands 2 and 7 were chosen because they showed highest correlations with the floristic data in the Mantel test (see below). NDVI was chosen because it provides information from bands 3 and 4 but is less correlated with bands 2 and 7 than either band alone.

Both combinations of satellite-derived variables were first used to produce a distance matrix in which each of the included variables was given equal weight by standardizing to zero mean and unit variance. However, there is no reason to believe that equal weights would give an optimal relation to floristic variation, so for both variable combinations, a second distance matrix was constructed where each of the standardized satellite-derived variables were weighted individually. The weights were obtained from an equation of multiple regression on distance matrices that was obtained for each variable combination. The independent distance matrices were in both cases based on the satellite-derived variables, and the dependent distance matrix was the floristic distance matrix that included both pteridophytes and Melastomataceae. For both models we report the standard partial regression coefficients (B) and the coefficients of multiple determination (R2). The partial regression coefficients for each satellite-derived variable were used as weights in calculating the corresponding Euclidean distance matrix. Backward elimination was applied in the multiple regression analysis that initially included only band 2, band 7 and NDVI to make sure that each of the variables in the final model would have a statistically significant (P < 0.05 after Bonferroni correction) contribution to the amount of variance explained (see Legendre et al. 1994; Legendre & Legendre 1998).

Mantel tests of matrix correspondence were run to analyse the degree of predictability in the floristic patterns of the sampling units. First, a Mantel test was run to quantify the correlation between the floristic distances as measured separately with the two plant groups (pteridophytes and the Melastomataceae). Then, Mantel tests were run to find out to what degree the floristic distances correlated with distances in pixel values in the satellite image and with geographical distance. All possible pairwise tests involving one of the three floristic distance matrices and one of the nine unweighted satellite-derived distance matrices were run. The weighted satellite-based distance matrices were only used in one Mantel test, i.e. the one using the combined pteridophyte and Melastomataceae distance matrix that had been used in the multiple regression analysis that provided the weights. All three floristic distance matrices were also used in two Mantel tests involving geographical distance: one test used original geographical distances, and the other used ln-transformed geographical distances.

Partial Mantel tests were run to find out how much of the correlation between two distance matrices (such as floristic and satellite-derived) remained after taking into account the correlation with a third distance matrix (such as geographical).

For each Mantel test and partial Mantel test, we report the standardized form of the Mantel statistic, which corresponds to a Pearson correlation coefficient calculated between the two distance matrices in question. The statistical significance of each correlation was determined by a Monte Carlo permutation test using 999 permutations, which allows testing of the statistical significance at the P < 0.001 level for each individual correlation. When interpreting the results, it is important to keep in mind that Mantel's matrix correlation coefficient rM is not comparable with the linear Pearson's correlation coefficient rP, which is based on the original variables rather than distances. When both are calculated on the same univariate data, rM obtains clearly lower values than rP, although they generally show the same degree of statistical significance (Legendre 2000).

Cluster analyses were carried out to classify the sampling units on the basis of their floristic similarity (on the basis of pteridophytes, Melastomataceae, and the two combined) and on the basis of their similarity in satellite image pixel values (using the matrix that gave the highest Mantel correlation with the floristic matrix). Two different agglomerative clustering methods were applied, both of which use a proportional-link linkage algorithm. The connectedness level was always set to 0.5 (i.e. mid-way between single and complete linkage).

One of the clustering methods, chronological clustering, applies a unidimensional constraint and only allows the fusion of sampling units or groups of units if they are contiguous along the transect (Legendre et al. 1985; Legendre & Legendre 1998). This method gives non-hierarchical results, because the clustering procedure performs at each step a permutational test of significance and decides according to the user-defined alpha significance level whether or not to fuse the two groups under consideration. When two groups are maintained separate, the probability that they actually belong to the same random population is less than the value of alpha used. In the present study, three runs were made with each similarity matrix using alpha significance levels 0.001, 0.01 and 0.05.

The second clustering method is hierarchical, and allows the fusion of sampling units irrespective of their position along the transect. This method was used to obtain a hierarchical classification of the clusters that had been recognized in the chronological clustering of the satellite image data at alpha significance level 0.01.

To test whether the distribution patterns of the individual plant species are related to the clusters obtained from the hierarchical classification of the satellite image, we computed the indicator values of Dufrêne & Legendre (1997) for all species that were recorded in five or more of the 86 sampling units. The indicator value combines into a single index two properties of a plant species’ distribution: specificity and fidelity. Specificity of a species to a given cluster is the proportion of its occurrences that are within that cluster, and fidelity is the proportion of sampling units in that cluster that contain the species. The two values are multiplied by each other and by 100 to yield index values in percentages. Indicator values were calculated for each of the clustering levels that contained between one and eight clusters, and the statistical significance of each indicator value was determined by a Monte Carlo test using 999 permutations.

Principal coordinates analysis on the floristic distance matrix (pteridophytes and Melastomataceae combined) was conducted to visualize the floristic relationships between the 86 sampling units in an ordination diagram.

The R-package was used to compute the resemblance matrices and to run the Mantel tests, cluster analyses and ordination analysis (version 3 was used for chronological clustering, version 4 for the other analyses). The multiple regression analyses were run with Permute! 3.4 (alpha version). The indicator values were computed using the program IndVal 2. All these programs are available through the web site


  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References

general patterns

In general, two kinds of landscape were found within the transect: low lying areas with small hills, and a more elevated area with higher hills. A part of the latter was like a plateau with deeply dissected creek valleys. Soil analyses showed that the soils in the low lying areas contained a smaller proportion of coarse sand but more exchangeable bases than soils in the more hilly areas (Fig. 2a).


Figure 2. Spatial patterns along a 43-km long transect in Peruvian Amazonia. (a) Topographical profile of the transect, and some soil characteristics measured from 36 soil samples. Species richness of (b) pteridophytes and (c) Melastomataceae in 500-m long units of the transect. Chronological clustering of 500-m long units of the transect based on either (d) satellite image pixel values or species composition of (e) pteridophytes and the Melastomataceae combined, (f) pteridophytes only or (g) Melastomataceae only. Thin vertical lines emphasize the limits between transect sections obtained with α = 0.01 in (d). Hierarchical clustering of these sections is shown in Fig. 3.

Download figure to PowerPoint

A total of 221 species were recorded in the 43-km long transect (130 pteridophytes and 91 Melastomataceae). The number of species within a 500-m long sampling unit ranged from 9 to 41 for pteridophytes and from 10 to 25 for Melastomataceae (Fig. 2b,c).

Floristic similarity (Jaccard index, calculated using both plant groups) between sampling units ranged from 0.01 to 0.71, the mean being 0.27. The results obtained when both plant groups were taken separately were remarkably similar. Jaccard index values ranged from 0 to 0.75 (mean 0.27) for pteridophytes and from 0 to 0.80 (mean 0.26) for Melastomataceae. These values show that some of the units were floristically completely different while others were highly similar.

matrix correlations and regressions

The actual similarity patterns for the two plant groups were also very similar. The matrix correlation computed between pteridophytes and Melastomataceae was 0.70 (Mantel test, P < 0.001). When the effect of geographical distance on this correlation was partialled out, the correlation coefficient decreased only slightly (rM = 0.68 with linear geographical distance; rM = 0.65 with ln-transformed geographical distance, P < 0.001 in both cases). Partialling out the correlation with satellite-derived distances had a more notable effect on the fern–Melastomataceae correlation (rM = 0.62 with bands 1–5 and 7, rM = 0.60 with band 2, band 7 and NDVI, P < 0.001 in both cases).

The Mantel tests involving the distance matrices based on satellite-derived variables (bands 1–5 and 7, NDVI) gave rather similar correlation patterns with pteridophytes, Melastomataceae and the combined floristic data (Table 1). In all cases, the matrix correlations were statistically highly significant. The single satellite-derived variable that showed the highest matrix correlation with all three floristic data sets was band 7, whose correlations with the floristic distance matrices were about as high as those obtained when information from all bands was combined but unweighted. Even higher correlations were obtained when the satellite-derived distance matrix included the unweighted band 2, band 7 and NDVI.

Table 1.  Matrix correlations as measured with the Mantel test between differences in floristic composition and pixel values of a Landsat TM satellite image in Peruvian Amazonia. Euclidean distance was used to measure differences in pixel values. The Jaccard index was used to measure floristic similarity, and this was converted to a distance measure before performing the Mantel test. Statistical significance of each correlation coefficient was calculated using 999 permutations; ***P < 0.001, **P < 0.01. No significance values are reported for the Mantel tests involving weighted distance matrices, as the only purpose of running these tests was to check if weighting improved the correlation coefficient
Band 10.27***0.18***0.25***
Band 20.45***0.36***0.44***
Band 30.14***0.13**0.14***
Band 40.30***0.27***0.31***
Band 50.41***0.36***0.42***
Band 70.47***0.44***0.50***
Bands 1–5, 70.48***0.42***0.49***
Bands 2, 7, NDVI0.52***0.48***0.55***
Bands 1–5, 7 (weighted)0.56
Bands 2, 7, NDVI (weighted)0.57

The multiple regression tests, where the distance matrices based on each of the satellite-derived variables were used to explain the variation in the combined floristic distance matrix, produced the following models:

  • (−0.17 × band 1) + (0.42 × band 2) − (0.04 × band 3) + (0.12 × band 4) − (0.19 × band 5) + (0.49 × band 7) [R2 = 0.34, P < 0.001]
  • (0.29 × band 2) + (0.34 × band 7) + (0.14 × NDVI) [R2 = 0.33, P < 0.001]

In an attempt to optimize the Euclidean distance matrices, the partial regression coefficients from these models were used to weight each of the satellite-derived variables before calculating the corresponding weighted Euclidean distances. It was indeed found in both cases that higher correlations with the floristic distance matrix were obtained using the weighted than using the unweighted satellite-derived matrix (Table 1).

To test whether floristic distances between sampling units can be predicted more accurately using satellite image reflectance patterns or geographical distances, a series of Mantel tests and partial Mantel tests were run. It is clear from the results in Table 2 that satellite images were superior in this respect. Geographical distances yielded Mantel correlations between 0.25 and 0.39, while the satellite-based distance matrices yielded correlations between 0.42 and 0.55. Furthermore, in the partial Mantel tests the satellite-derived distance matrices had a larger decreasing effect on the correlations between geographical and floristic distances than the geographical distance matrices had on the correlations between satellite-derived and floristic distances. The ln-transformed geographical distances showed clearly more explanatory power than the linear geographical distances.

Table 2.  Matrix correlations as measured with the Mantel test and partial Mantel test between differences in floristic composition, differences in pixel values of a Landsat TM satellite image, and geographical distance and its logarithm in Peruvian Amazonia. Euclidean distance was used to measure differences in pixel values. Jaccard index was used to measure floristic similarity, and this was converted to a distance measure before performing the Mantel test. Statistical significance of each correlation coefficient was calculated using 999 permutations; ***P < 0.001
Geographical distance 0.25***0.26***0.27***
Geographical distance; bands 1–5, 7 partialled out−
Geographical distance; bands 2, 7, NDVI partialled out
ln(geographical distance) 0.35***0.38***0.39***
ln(geographical distance); bands 1–5, 7 partialled out 0.12***0.19***0.17***
ln(geographical distance); bands 2, 7, NDVI partialled out 0.13***0.19***0.17***
Bands 1–5, 7 0.48***0.42***0.49***
Bands 1–5, 7; geographical distance partialled out 0.42***0.34***0.42***
Bands 1–5, 7; ln(geographical distance) partialled out 0.36***0.27***0.35***
Bands 2, 7, NDVI 0.52***0.48***0.55***
Bands 2, 7, NDVI; geographical distance partialled out 0.47***0.42***0.49***
Bands 2, 7, NDVI; ln(geographical distance) partialled out 0.43***0.37***0.44***

clustering and ordination

The satellite-derived matrix that yielded the highest Mantel correlation coefficient in Table 1, i.e. the weighted matrix with band 2, band 7 and NDVI, was chosen to be used in the cluster analyses. To check the robustness of the results, chronological clustering at α = 0.01 was performed using all four Euclidean distance matrices that combined several satellite-derived variables. The differences between the outcomes were minor, so using one of the other matrices for the rest of the analyses would not have changed the overall results. Because differences in the Mantel test results were small when only pteridophytes, only Melastomataceae, or both plant groups were used, it was considered unnecessary to repeat the multiple regression tests with pteridophytes only and Melastomataceae only.

The chronological clustering based on the satellite image data on the one hand and the three floristic data sets on the other were very similar (Fig. 2d–g). A break point was invariably found at 23 km, whichever similarity index and whichever alpha significance level was used. Additional statistically strongly supported break points were found at approximately 5 km, 8 km, at either 15 km or 17 km, and between 36 km and 38 km. The exact positions of the break points between 36 km and 38 km varied, indicating that the corresponding change was gradual rather than abrupt, but still distinct enough to be apparent in all data sets and at all alpha levels. Only one of the break points that were strongly supported by all three floristic data sets, namely that at about 21 km, was not apparent in the satellite image data.

The clustering hierarchy of the eight transect sections that had been obtained from the chronological clustering of the satellite image data (Fig. 2d, alpha significance level of 0.01) is shown in Fig. 3. The figure also shows for each clustering level the strong indicator species, defined here as those plant species whose indicator values were both 30% or higher and statistically significant (P < 0.05). Out of the 84 pteridophyte species that occurred in at least 5 of the 500-m sampling units, 61 (73%) had a statistically significant indicator value for at least one of the classification levels, and out of the 58 Melastomataceae species that occurred in at least 5 units, 49 (84%) did.


Figure 3. Hierarchical clustering of a 43-km long transect in Peruvian Amazonia on the basis of satellite image pixel values. The rightmost branches in the clustering correspond to the eight transect sections in Fig. 2(d) that are marked with similar shading. The strong indicator plant species are shown for the different clustering levels; the species names are displayed next to the cluster for which they are indicators. The leftmost occurrence of each species name is given in full, other occurrences are abbreviated. Indicator values are in parenthesis, and the highest indicator value for each species is in bold.

Download figure to PowerPoint

Different shadings are used in Figs 2(d) and 3 to indicate the four most spectrally distinct classes of sampling units. It was found that each of these four classes had at least one strong indicator species for each plant group, and most had several. The highest number of strong indicator species (30 pteridophytes and 25 Melastomataceae) was found at the two-class level, which contrasts the plateau-like area at km 23–38 with the rest of the transect (compare Figs 2 and 3). This separation is well defined in the chronological clustering based on both satellite-derived data (Fig. 2d) and floristic composition data (Fig. 2e–g). The four-class level, although not emerging as consistently in the different chronological clusterings, has almost as many indicator species (27 pteridophytes and 27 Melastomataceae).

The distribution of the significant indicator values is different for the two plant groups. Pteridophytes clearly have more indicator species for the richer-soil areas with gentle topography than for poorer-soil areas with hilly topography, whereas Melastomataceae indicator species are more evenly distributed between the two types of terrain. This pattern is consistent with the observation that species richness of pteridophytes is clearly at its lowest in the hilly area, while species richness of the Melastomataceae is more evenly distributed along the transect (Fig. 2b,c).

A comparison between the floristic ordination of the 86 sampling units (Fig. 4a) with the topographic pattern (Fig. 2a) indicates that the main floristic gradient corresponds with the terrain gradient. Sampling units from terrain with small hills and fine grained, cation-rich soils are found towards the left of the ordination, and those from high hills with coarser grained, cation-poorer soils are found towards the right. Even though the ordination is based only on floristic data, the spectral classes (see Figs 2d and 3) form distinct groups. Spectral class 4 is almost separated from the other three classes along axes 1 and 2, and even more so when axis 3 is taken into account. The other three classes segregate from each other mainly along axis 3 (Fig. 4b). The agreement is even more remarkable because the first step in the satellite-based classification was spatially constrained, while the floristic ordination did not involve any constraint.


Figure 4. Principal coordinates ordination of 500-m long units of a 43-km long transect in Peruvian Amazonia. The ordination is based on the species composition of pteridophytes and Melastomataceae. Allocation to spectral classes is based on satellite image pixel values, and the inset shows how sampling units belonging to each of the classes are referred to in the text (numbers), in this ordination diagram (symbols) and in Figs 2(d) and 3 (lines with shading).

Download figure to PowerPoint


  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References

random walk

The random walk model predicts that species turnover in space is gradual and that the species occurrence patterns of independent plant groups are not correlated either with each other or with external factors other than geographical distance. These predictions are not supported by our results. Species composition along the transect showed distinct turnover zones between more homogeneous stretches and the turnover zones were very similar for pteridophytes, Melastomataceae and the satellite image. Furthermore, the floristic patterns were highly correlated between the two plant groups and with satellite image patterns, even when the effect of geographical distance was taken into account.

The third prediction of the random walk model, that floristic similarity among sites decreases linearly with the logarithm of geographical distance, was partly supported by our results. The highly significant Mantel correlations between floristic and geographical distances are in agreement with this prediction, and the logarithmic distance model gave a clearly better fit than linear distance. However, some sampling units from distant parts of the transect were floristically more similar to each other than to intervening units (Fig. 4), which is contrary to the prediction. Partial Mantel tests suggested that such discrepancies may be explained by the effect of environmental patchiness: taking into account satellite image patterns reduced the correlation between geographical distances and floristic patterns substantially, although the correlations involving ln-transformed geographical distances remained statistically significant.

Condit (1996) has proposed that much of Amazonia may be ecologically homogeneous enough for the random walk model to operate at regional scales. In a more recent study, Condit et al. (2002) found that the random walk model explains the tree species distributions in data from Amazonian Peru and Ecuador fairly well at distances between 0.2 km to 50 km. The obvious contradiction with our results may stem from at least two different sources. First, Condit et al. (2002) did not test the floristic pattern they observed against the alternative hypothesis that edaphic or other environmental factors may influence species distributions. Environmental conditions usually present high spatial autocorrelation, so the presence of spatial autocorrelation alone is not sufficient to prove that environmental factors are not playing a role for species distributions (cf. Borcard et al. 1992). Second, it is possible that rain forest tree species distributions are not controlled by the same factors as the distributions of pteridophytes and Melastomataceae. Duque et al. (2002) reported that the floristic composition of canopy trees showed weaker correlation with environmental factors than did the floristic composition of understorey trees, which may indicate that large trees are less sensitive to environmental effects than smaller plants. However, the result may also be due to the small number of large trees present in each study plot, which would increase the effect of sampling error on the results.

Several studies have reported that patterns in the species composition of pteridophytes and Melastomataceae correlate − independently of geographical distance − with patterns in the species composition of trees in Peruvian rain forests (Tuomisto et al. 1995; Ruokolainen et al. 1997; Ruokolainen & Tuomisto 1998; Vormisto et al. 2000). Furthermore, in the present study the floristic patterns of both pteridophytes and the Melastomataceae were significantly correlated with the reflectance patterns in the satellite image, and the indicator value analyses showed that distributions of most species were highly linked to a reflectance based classification of the transect. Because both plant groups are mainly understorey plants that have little effect on the reflectance characteristics of the forest, such results are only possible if the forest canopy (consisting of trees, lianas and epiphytes) also shows similar patterns. Whether the reflectance patterns are caused by spatial pattern in the floristic composition of the forest canopy or in some structural or physiological properties is not known at present. More detailed studies on the degree of congruence among species distributions of different plant life forms are needed to clarify these questions.

The random walk model has been developed for homogeneous habitats (Hubbell 1997, 2001). The main issue is to define which are the appropriate scales in geographical, environmental, and temporal variation for the model to operate. In fact, the conceptually related carousel model (van der Maarel & Sykes 1993) maintains that random turnover in species composition is observed at small spatial scales over short time intervals, but over longer time intervals the cumulative species list for each microsite converges towards the habitat-specific species pool as the species rotate between microsites. The carousel model thus emphasizes recurrent patterns, and it has successfully (and, apparently, without major controversies) been applied to herbaceous plant communities in temperate areas (Palmer & Rusch 2001 and references cited therein). It may well be that such a carousel would become evident from the apparently random dynamics of tropical tree communities, if a large enough number of trees were observed for a long enough time period.


The regional homogeneity model (Pitman et al. 1999, 2001; Terborgh et al. 2002) predicts that species composition is essentially the same over wide areas and shows no abrupt turnover zones, and that the same (dominant) species are found everywhere. Our results do not support any of these predictions. The floristic similarity among our field samples ranged from 0.0 to more than 0.7 (Jaccard index), distinct turnover zones were found with both pteridophytes and the Melastomataceae, and the indicator value analyses showed that the occurrence of most species of both plant groups were significantly related to a satellite image based classification of the transect. Even though we did not record species abundance in the present study, it was obvious in the field that the floristically distinct units of the transect were characterized by different dominant species. This has also been found in earlier studies (Tuomisto & Poulsen 1996; Tuomisto et al. 1998).

A possible source of difference between our results and those of Pitman et al. (1999, 2001) and Terborgh et al. (2002) is the efficiency of sampling. With tree sampling, a constant problem is the low number of individuals recorded per species, which makes it difficult to judge whether differences among sampling units are due to real floristic differences or to low sampling intensity. The resultant loss of power in unravelling distribution patterns of individual tree species has been discussed, for example, by Clark et al. (1999). The advantage of using small plants for this kind of an inventory is that a high number of individuals is observed per species, so species are unlikely to be absent from a particular sample just because too few individuals are observed. In the present study, we recorded 221 species, and densities that we have measured elsewhere in the region suggest that these represent over 55 000 individuals of pteridophytes and 13 000 individuals of Melastomataceae. This contrasts strongly with the common situation in tree surveys, where the number of individuals recorded is usually much lower but the number of species is higher.

The conclusion that floristic differences exist within the transect is probably valid also for other plant groups, including trees, as discussed above. However, our more specific results, such as the range of floristic similarity values or the proportion of species that show non-random distribution patterns, are not directly transferable to other plant groups. This is because there is a tendency, at least among Amazonian trees and palms, that large statured species are more wide-spread both geographically and ecologically than small statured species (Ruokolainen & Vormisto 2000; Ruokolainen et al. 2002). Therefore, canopy trees might present a smaller proportion of species with significant indicator values for the satellite image patterns than did pteridophytes and Melastomataceae. On the other hand, canopy trees have a direct effect on radiance values in the satellite image, which may increase the proportion of species with significant indicator values.

edaphic patchiness

The edaphic patchiness model predicts that spatial variation in floristic composition reflects spatial patterns in the environmental conditions, that sharp species turnover occurs where the environmental conditions change, that different (dominant) species are found at sites with different environmental conditions, and that the distribution patterns of independent plant groups are correlated. All these predictions were consistent with our results.

On the basis of visual analysis of satellite images, Tuomisto et al. (1995) estimated that the average patch size in tierra firme forests of Peruvian Amazonia is slightly less than 5 km, and that a 30-km long random transect crosses, on average, four different kinds of patch. These results agree very well with our present results: on the basis of the pixel values in the satellite image, our 43-km long transect was divided into eight sections (at a significance level 0.01), which translates into an average patch size of slightly more than 5 km. Clustering of these, when interpreted together with an indicator species analysis, suggested the recognition of four different kinds of patch, even though the transect only covered tierra firme forest and excluded floodplains, swamps and other distinct vegetation types that were included in the earlier analysis.

Unlike Tuomisto et al. (1995), we now have ground truth data to test the ecological relevance of the satellite image patterns. When our transect was subdivided using floristic data of pteridophytes and the Melastomataceae, very similar patterns emerged both between the two plant groups and when these were compared to the patterns obtained from the satellite image data. In addition, a great majority of the plant species yielded statistically significant indicator values for the clusters based on satellite image data, confirming that the satellite image reflects ecological characteristics that are important for the distribution patterns of plants.

As different geological formations often show different denudation patterns, the differences in topography along our transect suggest that different geological formations yield the soil parent material in different parts of the transect. Soil samples confirmed that at least soil texture and cation content vary accordingly. Even though the soil data along our transect are too scarce to be analysed at the same degree of detail as the floristic and satellite image data, they are consistent with the hypothesis that the observed floristic patterns are caused by differences in soils. These results are in agreement with results obtained in Ecuadorian Amazonia at the same spatial scale but using scattered 500-m long transects with more detailed soil sampling (Tuomisto et al. 2002, 2003a).

The sampling unit size used in the satellite image analysis is rather large (500 × 400 m, or 20 ha), and therefore each unit is in itself a mosaic of different local conditions, of which the variability in topography is obvious from Fig. 2(a). Generally it can be expected that the more heterogeneous the sampling unit, the more difficult it will be to unravel correlations between species distribution patterns and environmental conditions (e.g. Palmer & Dixon 1990). In spite of this, all the analyses we conducted pointed to the result that species distribution patterns are correlated with patterns in satellite imagery and terrain characteristics. In spite of local variability, the mean conditions therefore seem sufficiently different for species to segregate at the landscape scale.


  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References

In an earlier study at a wider spatial scale, we found that the homogeneity hypothesis gained no support, and that both the random walk model and the ecological specialization model were needed in order to understand floristic patterns in western Amazonian tierra firme forests (Tuomisto et al. 2003b). Our present results show that the same is true at the landscape scale.

When considering to what extent the results of the present study can be generalized, it is important to know how representative the observed degree of ecological heterogeneity is. Two points indicate that our present study area was not exceptionally heterogeneous for western Amazonian tierra firme. First, the earlier satellite image analysis of Tuomisto et al. (1995) found that the area is one of the most homogeneous parts of Peruvian Amazonia. Second, the Mantel correlation coefficients we obtained between floristic distances and satellite-based distances in the present study were comparable to similar correlations obtained in Amazonian Ecuador in an area that has earlier been considered homogeneous (Tuomisto et al. 2003a).

The finding that non-obvious but distinct floristic and edaphic variation exists within tierra firme rain forests at the landscape scale has important practical implications. In ecological research, we should be cautious about extrapolating field results from one site to other sites under the assumption that tierra firme rain forests are all the same. In biodiversity conservation, a high degree of habitat heterogeneity implies an increased need for wide-scale information on species distribution and endemism patterns to better assess where the different habitats are, which species they harbour, and where conservation efforts should be concentrated.


  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References

We thank ACEER and INRENA for the permissions to work in the area, and Explorama Tours (especially Peter Jenson) for practical help in preparing the field work. The crew of men from nearby villages provided invaluable assistance during the making of the transect. The Landsat Pathfinder HTF project of the University of Maryland and NASA provided the satellite image. Soil analyses were done at the Agricultural Research Centre of Finland. The herbaria in Iquitos (AMAZ) and Turku (TUR) provided facilities, and Alan R. Smith and Robbin C. Moran helped with species identifications. Daniel Borcard and Pierre Legendre provided constructive comments on the manuscript and data analyses.


  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  • Ashton, P.S. (1976) Mixed Dipterocarp forest and its variation with habitat in the Malayan lowlands: a re-evaluation at Pasoh. Malaysian Forester, 39, 5672.
  • Austin, M.P., Ashton, P.S. & Greig-Smith, P. (1972) The application of quantitative methods to vegetation survey. III. A re-examination of rain forest data from Brunei. Journal of Ecology, 60, 305324.
  • Baillie, I.C., Ashton, P.S., Court, M.N., Anderson, J.A.R., Fitzpatrick, E.A. & Tinsley, J. (1987) Site characteristics and the distribution of tree species in Mixed Dipterocarp Forest on Tertiary sediments in central Sarawak, Malaysia. Journal of Tropical Ecology, 3, 201220.
  • Borcard, D., Legendre, P. & Drapeau, P. (1992) Partialling out the spatial component of ecological variation. Ecology, 73, 10451055.
  • Chave, J., Muller-Landau, H.C. & Levin, S.A. (2002) Comparing classical community models: theoretical consequences for patterns of diversity. American Naturalist, 159, 123.
  • Clark, D.B., Clark, D.A. & Read, J.M. (1998) Edaphic variation and the mesoscale distribution of tree species in a neotropical rain forest. Journal of Ecology, 86, 101112.
  • Clark, D.A., Clark, D.B., Sandoval, R.M. & Castro Vinicio, M.C. (1995) Edaphic and human effects on landscape-scale distributions of tropical rain forest palms. Ecology, 76, 25812594.
  • Clark, D.B., Palmer, M.W. & Clark, D.A. (1999) Edaphic factors and the landscape-scale distributions of tropical rain forest trees. Ecology, 80, 26622675.
  • Condit, R. (1996) Defining and mapping vegetation types in mega-diverse tropical forests. Trends in Ecology and Evolution, 11, 45.
  • Condit, R., Pitman, N., Leigh, E.G. Jr, Chave, J., Terborgh, J., Foster, R.B., Nuñez, V.P., Aguilar, S., Valencia, R., Villa, G., Muller-Landau, H.C., Losos, E. & Hubbell, S.P. (2002) Beta-diversity in tropical forest trees. Science, 295, 666669.
  • Dufrêne, M. & Legendre, P. (1997) Species assemblages and indicator species: the need for a flexible asymmetrical approach. Ecological Monographs, 67, 345366.
  • Duivenvoorden, J.F. (1995) Tree species composition and rain forest-environment relationships in the middle Caquetá area, Colombia, NW Amazonia. Vegetatio, 120, 91113.
  • Duivenvoorden, J.F. & Lips, J.M. (1998) Mesoscale patterns of tree species diversity in Colombian Amazonia. Forest Biodiversity in North, Central and South America, and the Caribbean: Research and Monitoring (eds Dallmeier, F. & Comiskey, J.A.), pp. 535549. Man and the Biosphere Series, Vol. 21. UNESCO, Paris and The Parthenon Publishing Group, New York, USA.
  • Duque, A., Sanchez, M., Cavelier, J. & Duivenvoorden, J.F. (2002) Different floristic patterns of woody understorey and canopy plants in Colombian Amazonia. Journal of Tropical Ecology, 18, 499525.
  • Foody, G.M. & Hill, R.A. (1996) Classification of tropical forest classes from Landsat TM data. International Journal of Remote Sensing, 17, 23532367.
  • Gentry, A.H. (1988) Changes in plant community diversity and floristic composition on environmental and geographical gradients. Annals of the Missouri Botanical Garden, 75, 134.
  • Hill, R.A. (1999) Image segmentation for humid tropical forest classification in Landsat TM data. International Journal of Remote Sensing, 20, 10391044.
  • Holmgren, P.K., Holmgren, N.H. & Barnett, L.C. (1990) Index Herbariorum. Part I: The Herbaria of the World. New York Botanical Garden, New York.
  • Hoorn, C. (1993) Geología del nororiente de la Amazonía peruana: la Formación Pebas. Amazonía Peruana. Vegetación Húmeda Tropical en el Llano Subandino (eds R.Kalliola, M.Puhakka and W. Danjoy), pp. 6985. PAUT and ONERN, Jyväskylä, Finland.
  • Hubbell, S.P. (2001) The Unified Neutral Theory of Biodiversity and Biogeography. Monographs in Population Biology 32. Princeton University Press, USA.
  • Hubbell, S.P. (1997) A unified theory of biogeography and relative species abundance and its application to tropical rain forests and coral reefs. Coral Reefs, 16 (Suppl.), S9S21.
  • Hubbell, S.P. & Foster, R.B. (1986) Biology, chance, and history and the structure of tropical rain forest tree communities. Community Ecology (eds J.Diamond and T.J.Case), pp. 314329. Harper & Row, New York, USA.
  • IFG. (1984) Mapa Planimétrico de Imágenes de Satélite 1:250 000 (Perú). Institute for Applied Geosciences, Neu Isenburg, Germany.
  • Kalliola, R., Puhakka, M., Salo, J., Tuomisto, H. & Ruokolainen, K. (1991) The dynamics, distribution and classification of swamp vegetation in Peruvian Amazonia. Annales Botanici Fennici, 28, 225239.
  • Kalliola, R., Ruokolainen, K., Tuomisto, H., Linna, A. & Mäki, S. (1998) Mapa Geoecológico de la zona de Iquitos y variación medioambiental. Geoecología y Desarrollo Amazónico: Estudio Integrado en la Zona de Iquitos, Perú (eds R.Kalliola and S. Flores Paitán), pp. 443457. Annales Universitatis Turkuensis Ser A II 114. University of Turku, Finland.
  • Legendre, P. (2000) Comparison of premutation methods for the partial correlation and partial Mantel tests. Journal of Statistical Computation and Simulation, 67, 3773.
  • Legendre, P., Dallot, S. & Legendre, L. (1985) Succession of species within a community: chronological clustering, with applications to marine and freshwater zooplankton. American Naturalist, 125, 257288.
  • Legendre, P., Lapointe, F. & Casgrain, P. (1994) Modeling brain evolution from behavior: a permutational approach. Evolution, 48, 14871499.
  • Legendre, P. & Legendre, L. (1998) Numerical Ecology. Second English Edition. Developments in Environmental Modelling, 20, 1853.
  • Lieberman, M., Lieberman, D., Hartshorn, G.S. & Peralta, R. (1985) Small-scale altitudinal variation in lowland wet tropical forest vegetation. Journal of Ecology, 73, 505516.
  • Van Der Maarel, E. & Sykes, M.T. (1993) Small-scale plant species turnover in a limestone grassland: the carousel model and some comments on the niche concept. Journal of Vegetation Science, 4, 179188.
  • Marengo, J.A. (1998) Climatología de la zona de Iquitos, Perú. Geoecología y Desarrollo Amazónico: Estudio Integrado en la Zona de Iquitos, Perú (eds R.Kalliola and S. Flores Paitán), pp. 3557. Annales Universitatis Turkuensis Ser A II 114. University of Turku, Finland.
  • Newbery, D.McC. & Proctor, J. (1984) Ecological studies in four contrasting lowland rain forests in Gunung Mulu national park, Sarawak. IV. Associations between tree distribution and soil factors. Journal of Ecology, 72, 475493.
  • Novo, E.M. & Shimabukuro, Y.E. (1997) Identification and mapping of the Amazon habitats using a mixing model. International Journal of Remote Sensing, 18, 663670.
  • Palmer, M.W. & Dixon, P.M. (1990) Small-scale environmental heterogeneity and the analysis of species distributions along gradients. Journal of Vegetation Science, 1, 5765.
  • Palmer, M.W. & Rusch, G.M. (2001) How fast is the carousel? Direct indices of species mobility with examples from an Oklahoma grassland. Journal of Vegetation Science, 12, 305318.
  • Pitman, N.C.A., Terborgh, J., Silman, M.R., Nuñez, V.P. (1999) Tree species distributions in an upper Amazonian forest. Ecology, 80, 26512661.
  • Pitman, N.C.A., Terborgh, J., Silman, M.R., Nuñez, V.P., Neill, D.A., Cerón, C.E., Palacios, W.A. & Aulestia, M. (2001) Dominance and distribution of tree species in upper Amazonian terra firme forests. Ecology, 82, 21012117.
  • Poore, M.E.D. (1968) Studies in Malaysian rain forest. I. The forest on Triassic sediments in Jengka forest reserve. Journal of Ecology, 56, 143196.
  • Poulsen, A.D. & Balslev, H. (1991) Abundance and cover of ground herbs in an Amazonian rain forest. Journal of Vegetation Science, 2, 315322.
  • Poulsen, A.D. & Tuomisto, H. (1996) Small-scale to continental distribution patterns of neotropical pteridophytes: the role of edaphic preferences. Pteridology in Perspective (eds J.M.Camus, M.Gibby, and R.J. Johns), pp. 551561. Royal Botanical Gardens, Kew, London.
  • Räsänen, M., Linna, A., Irion, G., Rebata Hernani, L., Vargas Huamas, R. & Wesselingh, F. (1998) Geología y geoformas de la zona de Iquitos. Geoecología y Desarrollo Amazónico: Estudio Integrado en la Zona de Iquitos, Perú (eds R.Kalliola and S. Flores Paitán), pp. 59137. Annales Universitatis Turkuensis Ser A II 114. University of Turku, Finland.
  • Van Reeuwijk, L.P. (1993) Procedures for Soil Analysis. Technical Paper 9, 4th edn. International Soil Reference and Information Centre, Wageningen, the Netherlands.
  • Ruokolainen, K., Linna, A. & Tuomisto, H. (1997) Use of Melastomataceae and pteridophytes for revealing phytogeographic patterns in Amazonian rain forests. Journal of Tropical Ecology, 13, 243256.
  • Ruokolainen, K. & Tuomisto, H. (1998) Vegetación natural de la zona de Iquitos. Geoecología y Desarrollo Amazónico: Estudio Integrado en la Zona de Iquitos, Perú (eds R.Kalliola and S. Flores Paitán), pp. 253365. Annales Universitatis Turkuensis Ser A II 114. University of Turku, Finland.
  • Ruokolainen, K., Tuomisto, H., Vormisto, J. & Pitman, N. (2002) Two biases in estimating range sizes of Amazonian plant species. Journal of Tropical Ecology, 18, 935942.
  • Ruokolainen, K. & Vormisto, J. (2000) The most widespread Amazonian palms tend to be tall and habitat generalists. Basic and Applied Ecology, 1, 97108.
  • Saatchi, S.S., Nelson, B., Podest, E. & Holt, J. (2000) Mapping land cover types in the Amazon basin using 1 km JERS-1 mosaic. International Journal of Remote Sensing, 21, 12011234.
  • Sabatier, D., Grimaldi, M., Prévost, M.-F., Guillaume, J., Godron, M., Dosso, M. & Curmi, P. (1997) The influence of soil cover organization on the floristic and structural heterogeneity of a Guianan rain forest. Plant Ecology, 131, 81108.
  • Svenning, J.-C. (1999) Microhabitat specialization in a species-rich palm community in Amazonian Ecuador. Journal of Ecology, 87, 5565.
  • Terborgh, J., Foster, R.B. & Nuñez, V., , P. (1996) Tropical tree communities: a test of the nonequilibrium hypothesis. Ecology, 77, 561567.
  • Terborgh, J., Pitman, N., Silman, M., Schichter, H. & Nuñez, V., , P. (2002) Maintenance of tree diversity in tropical forests. Seed Dispersal and Frugivory: Ecology, Evolution and Conservation (eds , D.J., Levey, W.R. Silva andM. Galetti), pp. 117. CABI Publishing, Wallingford, UK.
  • Tuomisto, H. (1998) What satellite imagery and large-scale field studies can tell about biodiversity patterns in Amazonian forests. Annals of the Missouri Botanical Garden, 85, 4862.
  • Tuomisto, H., Linna, A. & Kalliola, R. (1994) Use of digitally processed satellite images in studies of tropical rain forest vegetation. International Journal of Remote Sensing, 15, 15951610.
  • Tuomisto, H. & Poulsen, A.D. (1996) Influence of edaphic specialization on pteridophyte distribution in neotropical rain forests. Journal of Biogeography, 23, 283293.
  • Tuomisto, H. & Poulsen, A.D. (2000) Pteridophyte diversity and species composition in four Amazonian rain forests. Journal of Vegetation Science, 11, 383396.
  • Tuomisto, H., Poulsen, A.D. & Moran, R.C. (1998) Edaphic distribution of some species of the fern genus Adiantum in Western Amazonia. Biotropica, 30, 392399.
  • Tuomisto, H., Poulsen, A.D., Ruokolainen, K., Moran, R.C., Quintana, C., Celi, J. & Cañas, G. (2003a) Linking floristic patterns with soil heterogeneity and satellite imagery in Ecuadorian Amazonia. Ecological Applications, 13, 352371.
  • Tuomisto, H. & Ruokolainen, K. (2001) Variación de los bosques naturales en las áreas piloto a lo largo de transectos y en imágenes de satélite. Evaluación de Recursos Vegetales no Maderables en la Amazonía Noroccidental (eds J.Duivenvoorden, H.Balslev, J.Cavelier, C.Grandez, H.Tuomisto and R. Valencia), pp. 6396. IBED, Universiteit van Amsterdam, The Netherlands.
  • Tuomisto, H., Ruokolainen, K., Kalliola, R., Linna, A., Danjoy, W. & Rodriguez, Z. (1995) Dissecting Amazonian biodiversity. Science, 269, 6366.
  • Tuomisto, H., Ruokolainen, K., Poulsen, A.D., Moran, R.C., Quintana, C., Cañas, G. & Celi, J. (2002) Distribution and diversity of pteridophytes and Melastomataceae along edaphic gradients in Yasuni National Park, Ecuadorian Amazonia. Biotropica, 34, 516533.
  • Tuomisto, H., Ruokolainen, K. & Yli-Halla, M. (2003b) Dispersal, environment, and the floristic variation of western Amazonian forests. Science, 299, 241244.
  • Vormisto, J., Phillips, O.L., Ruokolainen, K., Tuomisto, H. & Vásquez, R. (2000) A comparison of fine-scale distribution patterns of four plant groups in an Amazonian rainforest. Ecography, 23, 349359.