Study sites and vegetation sampling

The first study site was in the Peruvian cloud forest ‘Bosque de Neblina de Cuyas’ (BNC), 4°35ʹS, 79°42ʹW. The elevations range from 2359 to 3012 m above sea level and are characterised by irregular surfaces and steep slopes. Accordingly, the forest stand is unevenly aged, and has a high rate of biodiversity and endemism [4]. This forest is one of the last original cloud forests on the Pacific slope of the Andes in Northern Peru. Although the zone has been identified as an Important Bird Area (IBA005), the forest has no other protection or conservation. This forest belongs to the local communities inhabiting the areas, named ‘Comunidad Campesina de Cuyas Cuchayo Ambasal’, who have declared it a private conservation area. I have a verbal agreement with them that allows me to work there; they have no written records yet. Nevertheless, the forest faces a serious threat because it is probable that mining activity will begin in the region, as it has in many areas of the Andes. I conducted an inventory in 2008 by establishing three different 1-ha square plots located in the old-growth stand. Physiography and environmental conditions were relatively homogenous across all three plots [7]. As the plots were similar, the patterns identified for each species in the different plots can be considered replicates of the same ecological processes. In each plot, all the free-standing woody plants taller than 1.3 m were mapped and the diameter at breast height (DBH), height, and species were recorded. A list of the characters for the species found, including their life form and shade tolerance, is provided in Supplementary material S1 Table, additional information on the plots and the species has been published previously [4]. BNC census data are given in Supplementary material S1 Data and more data are freely available from http://www.alicialedo.com. The BCI data is available upon request at http://ctfs.arnarb.harvard.edu/webatlas/datasets/bci/

The second study site was Barro Colorado Island (BCI), a seasonally dry tropical moist forest, 9°10′N, 79°51′W, in Panama [27, 28]. A 50-ha permanent plot was established in this well documented forest in 1982. Every 5 years, all free-standing trees in the plot ≥ 1 cm in diameter are re-measured in the plot. BCI is managed by the Smithsonian Tropical Research Institute (STRI), and it forms part of the Barro Colorado Nature Monument (BCNM). I used data from the 6th census on BCI, which was conducted in 2005. Due to the large size of the plot, diverse areas in relation to the topography can be observed, and six different habitats have been described within the plot: low plateau, high plateau, slope, swamp, streamside, and young forest [6].

Both study areas are neotropical forests, although there are some differences with regard to the ecosystems, environmental conditions, inventory, and plot size. BNC is a montane cloud forest, whereas BCI is a lowland seasonal forest. Therefore, BNC experiences a colder climate, but is less affected by seasonal changes in humidity than the BCI forest [7]. Nevertheless, the overall spatial pattern of the tree species is similar at both sites: the tree species in both BNC [4] and BCI [1] show clustered distributions, which are typical for most of the tropical forests analysed to date [2]. Hence, the ecological processes producing the spatial pattern of the different species, dispersal limitations [1], and niche preferences or habitat filtering [6] operate in both studied forests. However, as the plot size differed considerably between the sites, so did the heterogeneity within the studied areas. BNC plots were more homogeneous in terms of environmental variation than the BCI plots because a smaller area was examined, because of which different habitat types could not be differentiated in BNC as they have been in BCI [6]. Accordingly, BNC plots can be considered a single habitat under a habitat-partitioning definition that is similar to the BCI habitat definition. Hence, to avoid the effect of species-habitat associations in the BCI plot and, thus, be able to compare the plots in terms of the environmental variability of the two study areas, spatial analyses for pairs of species characteristics from each of the defined habitats were performed. Thus, the BCI species considered (S2 Table) were those that are associated with a particular habitat (p-value < 0.01), following the results of Smith et al [7]. A list of the characteristics for the species found, including life form and shade tolerance, is also provided in S2 Table. However, it is important to note that spatial variability and microhabitat associations also exist within BNC [7] and BCI habitats [29]. While the methods used in this study should enable the study plots to be compared in terms of environmental heterogeneity, the microhabitat effect may not have been totally removed. Indeed, there may be more environmental variation in BNC, because it is a mountain forest with a marked slope, which may lead to a micro-environmental gradient with elevation, absent in BCI. This suggests that BNC trees are under greater environmental stress.

Statistical analysis

A multivariate point pattern analysis approach was used to study the spatial association between species [30, 31]. The point pattern analysis technique describes the second-moment properties of the spatial pattern of a point layer as a function of the inter-point distance as the distance of observation increases. Hence, this approach allows the detection of the observed spatial pattern of a point layer at different scales. Currently, point pattern techniques are widely used in ecology to analyse the spatial structure of the elements that make up ecosystems.

To assess the pairwise associations between species when considering the individuals of both species together, this study employed the bivariate function [32], which is used to analyse the spatial association (attraction, repulsion, or spatial independence) between two types of points, which in this analysis were both woody plant species. (1) where d is the distance lag of the analysis, n_r and n_s are the number of trees in each class, and , respectively (A being the area of the plot), and ω_ιj is the second-order moment measure, , which incorporates the boundary-effect correction proposed by Ripley (1979).

The toroidal shift null model [33] was constructed to test the spatial independence between the species. In the toroidal shift model simulation, the position of the points of one class is left unchanged, while all the points of the other class are shifted by the same random vector, and the area is considered a torus. Deviations from the model would imply spatial attraction (if the empirical function is above the 95% quantile) or spatial repulsion (if the empirical function is below the 95% quantile) between the species included in the analysis. A schematic visualization of those processes can be found in Fig 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Conceptual framework detailing the potential spatial patterns between species that can be detected with the point pattern functions used in this study.

In every of the 5 different cases, the intertype function used to detect the spatial dependence is in a white box, the observed spatial dependence and the values in a grey box, the function that support that result is in a blue box. The potential underlying mechanisms that caused the observed pattern are indicated in green characters. Only the species comprising the pair of target species were represented in this scheme, named species A and B, while the other species were blind. Species A and B may be any pair of species, different in each case. Rosy upper box: light tolerance and growth form classification of the analysed species. The four different cases of inter cross guilds for species A and B are represented below.

https://doi.org/10.1371/journal.pone.0141387.g001

To assess the pairwise associations between species when considering their different developmental ages, the intertype mark correlation function, was used, which we proposed and tested previously [34] and is used for the first time in this analysis. This function assesses the expected covariance of the marks between pairs of points belonging to two different tree types within a given distance, which means that it is a pairwise and mark correlation function. The trunk diameter was considered as a surrogate of an individual’s developmental stage in this analysis. In tropical forests, seedling growth can be supressed for years until a high light opportunity presents itself [35], so smaller trees can be decades old, but, as they are waiting for an opportunity to grow rapidly, I considered them to be individuals in the sapling developmental stage. Besides, although the trunk diameter is not exactly correlated with age, it is reasonable to consider that larger trees of a given species (50–70 cm DBH) are older than the smaller conspecific trees (1–5 cm DBH). Hence, I introduced the diameters as marks, being a continuous variable, and the two types of points as two different species. It should be pointed out that, as observed in the function, the range of the two marks is normalised in the function, allowing the comparison of two species with different diameter ranges. Because I had considered the diameter as a measure of tree size and, thus, a surrogate of developmental stage, the results of the function would allow the identification of the spatial relationships between two species when considering younger individuals of one species versus older individuals of the other, and vice versa. (2) where m_i and m_j are the values of the mark at the points i and j, respectively, and s_r and s_s are the standard deviation of the mark of the points considered in class r and class s.

The random marking null model [34] was used to test the spatial independence of the diameters. The random marking null model tests independence in the mark distribution when the spatial pattern of both point classes is fixed and the mark is randomised. If the trees of two species with similar normalised values for trunk diameter coexist at a certain scale, the larger (or smaller) individuals of one species and the larger (or smaller) individuals of the other will appear to be spatially related when the empirical function is above the 95% quantile. In contrast, the function will be below the 95% quantile bounds if species having opposite diameter values appear spatially together, that is larger individuals of one species occur with smaller individuals of the other species. If the diameters of the two species are independently distributed, the empirical function would be within the quantile bounds, regardless of the spatial pattern. A schematic visualization of those processes can be found in Fig 1.

For BNC only, I calculated the overall and functions by standardising and replicating the Eq 1 and Eq 2 functions obtained in the tree plots. I used a standardisation method [36]. The standardized function ( for Eq 1 and for Eq 2) for each r replicate was standardized by the transformation: ; where is the target function (either Eqs 1 or 2), and , with and representing the upper and lower 95% quantile bounds, respectively, of the null model. I translated the 95% upper and lower quantile bounds ( and of the null model to between –1 and 1 for each distance d to make them comparable. Then, the overall standardized function was calculated for the three replicates: (3)

Type II errors (false positives) were minimised in the BNC approach because the results from three different spatially separated plots were used. Therefore, the casual associations in one plot were removed, and only coincident results were considered, making the results more robust.

In the spatial analysis, I included only species with more than 60 individuals in each of BNC plots and in the BCI plot. In BNC forest, a few species had fewer than 60 individuals in only one of the three plots. When this occurred, the results for this species in that plot were not included in the analysis and only the two plots with a sufficient number of individuals were included in the replicate analysis (Eq 3). Although this approach excluded rare species, these are seldom included in spatial pattern analysis, and ignoring them does not induce a notable bias [2]. Hence, 20 species were included in the BNC analysis, comprising 40% of the species found and 93% of the total woody plant individuals in the measured plots (S1 Table). Twenty-five species (associated with a habitat and with more than 60 individuals) were analysed in the BCI plot (S2 Table).

A total of 399 simulations were performed to build the quantile boundaries of the null models, and I designated 95% of the quantile boundaries of the null model as the area corresponding to an acceptance of the null hypothesis. Moreover, I calculated the quasi p-values of the analysis using a previously described method [37] and considered a significance level of 0.01 to mean acceptance. The distance considered in this analysis was 50 m. These values corresponded to half the length of the shorter dimension of the BNC plots, as previously recommended [30].

Both the and functions revealed all the scales at which deviations from the null hypothesis occur, which are considered to be an interval. Nevertheless, to synthesise the results, I considered a single distance at which a spatial interaction arose, which was also the point at which the spatial association was more significant, i.e. the point at which the difference between the empirical function and the limit band of acceptance (the upper and lower quartile, respectively) was greater, as weighed by the distance interval area of the null hypothesis acceptance.

As a large number of tests were carried out, 5% of the associations were expected to be false positives (type I errors), which equates to four associations in each of the two analyses (with and without including trunk diameter) in BNC and six in each of the two analysis for BCI, based on the number of statistical tests (20 and 25 species, respectively) that were conducted.

After the above analyses and based on the results of the pairwise associations, I examined whether pairwise associations were more common between species with similar life forms and shade tolerance. This may also indicate species arrangement and forest organization. The life forms and shade tolerance of species from BNC are in [4] and S1 Table. The life forms of species from BCI are described in [23] and presented in S2 Table. Each functional trait has been represented by a different symbol in Fig 1, which is used throughout the manuscript to represent that particular functional trait.

Pairwise associations and habitat associations

In order to check whether the found pairwise associations are due to a pairwise concordance in habitat associations, I compared the results obtained from examining the pairwise associations with the microhabitat associations found in [7] for BNC. If the spatially associated species were also associated to a similar habitat, the spatial association is probably due to a positive response to similar habitat conditions. If the species are not associated with the same microhabitats, then it can be assumed that the underlying process that causes the pairwise associations may be due to a different process. An analogous analysis could not be done in BCI. Species microhabitat association results do not exist and were not within the scope of this study.

Spatial associations between species in a tropical cloud forest

Fourteen species, out of the 20 species included in the analysis, showed some spatial association with other species, according to the analysis (Table 1). Twelve were positive and two were negative. The interaction distances ranged from 0.5 to 46 m, but most of them were at short distances of up to 10 m (Table 1, lower left). They represent 7% of the potential associations (a hypothetical scenario where all the species have spatial associations). For those pairs of species that exhibited spatial repulsion, it was noted primarily at a small scale, with maximum distances for the spatial association reaching 1 m. Most of the pairs showing spatial associations were between species that had similar life forms (Fig 2, Table 1) and, hence, coexisted in the same vertical layer of the stand. For example, Parathesis Ms and Eugenia Ms showed attraction at a small scale, and these species are the most frequent under-canopy shade-tolerant species in the lower layer of the forest. Persea Ms and Morus insignis are both canopy species, and also appeared spatially associated at a small scale (Table 1). Species with different shade tolerances generally showed spatial repulsion (Fig 2, Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Number of pairwise association between species with similar (dark grey) or different (light grey) dispersal modes, shade tolerance and life-forms found using the and analyses for positive and negative associations in BNC.

In , positive and negative associations indicate spatial attraction and repulsion, respectively. In the analysis, positive values indicate attraction among species at similar ages, and negative values indicate attraction among species at opposite ages.

https://doi.org/10.1371/journal.pone.0141387.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Matrix of results from BNC: (i) the lower and left half of the matrix indicates the distance (metres) at which a spatial association was found using the analysis.

A positive value indicates attraction, and a negative value indicates repulsion: (ii) the upper and right half of the matrix indicates the distance at which a spatial association was found using the analysis. If a positive value indicates attraction among species at similar life-stages, then a negative value indicates attraction among species at opposite life-stages. In every case, the distances correspond to the point at which the spatial association is more important. The numbers in italics are the number of spatial associations that each species exhibited, corresponding to the results plotted in that half of the matrix. The cells marked in bold indicate that that pair of species showed similar association to a habitat (Ledo et al 2013), if the relation between species is positive or showed attraction and repulsion to a habitat if the relation is negative. Ms refers to Morphospecies, from Ledo et al. 2012.

https://doi.org/10.1371/journal.pone.0141387.t001

The number of spatial associations found when diameter size was accounted for in the analysis was notably larger than in the previous analyses, as derived from the results (Table 1). In this second case, 86 interactions among the pairs of species were found (59 positive and 27 negative), with distances ranging from 0.5 to 47 m (Table 1, upper right). In this case, positive associations meant spatial attraction between species at a similar developmental stage and negative associations meant spatial attraction between species at different developmental stages. The number of observed associations represented 45% of the potential associations. In addition, associations were found between species in similar vertical layers and also between species with different life forms (Fig 2, Table 1). The under-canopy species had a greater number of associations than the other life forms (Table 1, S1 Table). The species with the greatest numbers of associations were Solanum Ms1 and Piper elongatum, both pioneer species (S1 Table). Once again, positive associations, in this case, meaning spatial attraction among species of similar ages, appeared to be more common between species with different shade tolerances and life forms (Fig 2, Table 1).

Spatial associations between species in a tropical lowland moist forest

Fourteen species, out of the 25 species included in the analysis, showed some spatial association with other species, according to the analysis (Table 2). The total number of spatial interactions in this second forest was nine and all of them were positive, with distances ranging from 7 to 40 m (Table 2). That represents 15% of the potential associations. No associations were found in the selected species in the high plateau and stream habitats. However, this may be due to the lower number of species included in this analysis, and hence we cannot make any ecological assumptions based on this result. In this analysis, it was observed that most of the spatial associations were between different growth form species, such as Cordia bicolor (midstorey) and Coussarea curvigemmia (understorey), or Ocotea oblonga (canopy) and Xylopia macrantha (midstorey). However, no clear pattern was observed for species shade tolerance (Table 2), which was mainly due to the lack of information on this characteristic (Fig 2, Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Results of the and analyses for the species considered in each habitat in the BCI plot.

S indicates the number of species analysed in the habitat, m indicates the distance (meters) at which a spatial association was found. In the analysis, if positive means attraction, then negative means repulsion. In the analysis, if positive means attraction among species at similar life-stages, then negative means attraction among species at opposite life-stages.

https://doi.org/10.1371/journal.pone.0141387.t002

Once again, the number of spatial associations found when diameter size was accounted for in the analysis was larger than in the previous analyses, although the difference was not as evident as in the cloud forest. Fourteen interactions among pairs of species were found by the analysis (10 positive and 4 negative), with distances ranging from 5 to 35 m (Table 2). That represents 23% of the potential associations. Most of the pairwise associations were positive, but some were negative, as the case of Alibertia edulis and Cassipourea elliptica (Table 2). Most pairs of associated species had different growth forms (Fig 2, Table 2), such as Coussarea curvigina (understorey) and Tetragilis panamensis (canopy). However, pairwise associations were also found for species with the same growth form, such as Protium panamense and Simarouba amara, both of which are canopy species (Table 2).

Pairwise associations, light tolerance, and growth form

The undercanopy and shade species showed a greater number of associations with other species than the canopy and gap species in both forests (Tables 1 and 2). In the montane BNC, the percentage of spatial associations regardless of pairs of species belonging to a similar shade-tolerance group was the same as the percentage of pairs of species belonging to a different group, but spatial repulsion occurred only for species with different growth forms. In the lowland BCI forest, the pattern for positive spatial associations was the opposite: species with positive associations belonged to different groups. As for growth form, species had positive spatial associations with species mainly belonging to the same group in BNC, but with species from different groups in BCI (Fig 2). When size was included in the analysis, pairs of associated species mainly belonged to different guilds for both light tolerance and growth in BNC. In BCI, this pattern was observed only for growth form (Fig 2).

Pairwise associations and habitat associations in BNC

Five out of the 14 spatial interactions derived from the analysis (36% of the total spatial associations) were also found in the habitat association analysis (Table 1). That means, for five of the 14 found spatial interactions the pairs of species either showed spatial attraction and were associated with a particular microhabitat or showed spatial repulsion where one species was associated to a habitat whereas the other avoided that habitat.

When plant size was considered in the analysis, 12 out of 86 interactions (14% of the total spatial associations) followed the aforementioned pattern, in which both had a positive association and exhibited either a positive or a negative association with a particular microhabitat (Table 1).

Spatial associations that could be due to microhabitat associations, because they were also consistent with this pattern, were found mainly between three species: Miconia media, Morus insignis, and Piper elongatum. Both M. media and P. elongatum species are pioneer, undercanopy species (S1 Table). The distribution of these species may be related to gaps. M. insignis is a species frequently associated with streams or wetter habitats, so it may be more sensitive to and dependent on particular habitat types [7].

Spatial pairwise species associations

A pattern of species associations that is inconsistent with a random distribution and supports species spatial associations, both positive (spatial attraction) and negative (spatial repulsion or desegregation), has been clearly shown in both study areas (Tables 1 and 2). The pairs of species that were positively spatially associated increased notably in both forest types when tree size was considered (Tables 1 and 2). This indicates that the spatial preferences and associations between species are not constant throughout the life of trees, and nor are species-habitat associations [23] and distance- and density-dependence mechanisms [3]. The positive spatial associations found may arise from a few underlying mechanisms. One of the most explored and recognized mechanisms is habitat associations [6]. We expect two species to be spatially related if they appear to prefer the same habitat. Indeed, in the current analyses, few of the observed positive pairwise spatial associations may be due to habitat associations, whether or not the tree size of the individuals is taken into account. Nevertheless, those associations were found for pioneer species in this particular case; therefore, this may be due either to a dependence on specific canopy disturbance conditions, or a response to a determinate amount of light [20, 37]. Nevertheless, the percentage of species associated with a habitat may be larger if we consider the ‘habitat’ definition in a wider context, i. e. including a particular mycorrhizal community, or microbiota existing in the soil. As that characteristic was not considered in the former habitat partitioning, we cannot support or discard it because no data are available; therefore, habitat associations may be more numerous than the ones we found. However, we cannot rule out another effect: that some species are in the same habitat due to a spatial pairwise positive association (facilitation or mutualism), which is not due to habitat associations. Further, a third situation could also be possible: a few species may be close together due to a combination of both effects—habitat associations and facilitative interactions—acting simultaneously.

As the effect of a similar response to a habitat cannot be assumed to be the sole mechanism underlying positive associations, we cannot discard the possibility that facilitation (both direct and indirect) between species may also exist in the forest. Two different cases can be discerned here: most of the pairwise spatial associations were found when the life stages of the species were similar, which indicated the existence of species-specific cohort groups. Alternatively, associations between mature individuals of one species and younger individuals of the other species may indicate that a nursery could exist in the forest. Those associations were found mainly among species with different shade tolerances and growth forms (Table 1, Fig 2). Hence, the observed patterns suggest that some undercanopy species appear mainly close to a specific canopy species, and vice versa. This is interesting because these species associations indicate that some successional species-directed group of species appears in the forest. This agrees with the successional niche hypothesis [25]. However, nursery effects were not observed in the studied BCI species. BNC, a montane cloud forest, is found on steep slopes, which are exposed to more variable temperature conditions, lower light levels, and high wind speeds in the dry season. The environmental conditions are not as benign as in lowland tropical forests. This suggests that BNC trees are under greater environmental stress. The postulated stress hypothesis [15, 19] states that facilitative plant-plant interactions are more important and more intense in high-stress environments. Nevertheless, despite finding a potential nursery effect, we cannot discard that this may occur only at an early stage but end with the small tree being suppressed by the larger tree in the future [37]. If so, we are observing facilitation at a particular point but not throughout the life stages. This, however, agrees with the idea that those spatial associations are not constant throughout life.

Another possible explanation (and perhaps the simplest) for the positive spatial interaction in the forest is that both species were dispersed into the space by different dispersers, at roughly similar times, such as when a gap was formed [38]. Alternately, the same dispersal agent, such as a particular bird or mammal, lives in a preferential area or habitat in the forest and so disperses both species in the same areas. Hence, the dispersal mechanism could also explain some of the observed pattern.

Negative pairwise spatial associations, which reflect repulsion between species, were never found in BCI (Table 2) and seldom in BNC (Table 1). Furthermore, there were very short distances (less than 1 m) between these associations, which is probably due to the physical impediment of trees growing close to each other [4]. This suggests that competitive exclusion between species [39] was not observed in the studied forests, which explains why tropical forests contain large numbers of species [8]. It also indicates that pairwise direct allelopathy [22] is not commonly present among tropical trees. However, it should be pointed out that repulsion among trees does exist in tropical forests due to competition [13, 14], but this competition for space, light, and nutrients may be on a small scale and be species-generalist, rather than species-specific. Direct experiments are needed to determine the mechanisms and causes responsible for the observed pattern of spatial associations.

Implications for community assembly

It is well known and acknowledged that both niche preferences and distance-dependence mechanisms are involved in species’ spatial patterns and community assembly [8, 40]. The importance of these processes varies among species: some species are more sensitive to habitat conditions [7, 40], others to density dependence [9, 10], disturbance dependence [28, 41], or the characteristics of their neighbours, as the results of this study showed. This indicated that not all species are sensitive to or affected by a particular factor to the same degree, but differ in their responses to those ecological mechanisms. Consequently, different species appear to be sensitive in various degrees to different mechanisms; for species with stronger habitat preferences, habitat association may be the main mechanism determining the final distribution of the species, and rare species are more strongly affected by density dependence than common species [9, 10]. Specific spatial associations with a particular paired species may also play a notable role in the distribution and development of some species, as shown in this study. Spatial associations appear frequently between species with different functional traits (Fig 2), indicating that functional traits may be important and are predictors of forest organization, agreeing with the previously suggested idea [17]. Furthermore, the aforementioned mechanisms not only act together, but in a hierarchical way. A species may first need suitable environmental and stand conditions, which can either be a determinate forest structure or a response to a disturbance. Under those conditions, density dependence mechanisms would appear. Moreover, if a companion species is present, it may facilitate the development of the target species at that point, which may otherwise fail.

Facilitation among species has increased in importance in evolutionary theory [24]. Possibly, it is time to presume, or at least consider, the idea of a network of facilitative trees and incorporate it into the ecology of tropical tree coexistence. Most of the analysed species displayed more than one pairwise association (Tables 1 and 2). This finding suggests the existence of a mutualist network among species in tropical community forests, as described for arid ecosystems [18]. It is also necessary to consider the developmental stage of the individuals in community assembly. This study shows that age-independent analysis of species associations, the most common analyses done to date, underestimates interspecific plant spatial associations.