1. One of the earliest hypotheses to explain high tropical forest diversity proposes that species are differentially specialized in their germination or growth and survival to particular habitats.
2. I examined evidence for habitat associations in seedling density and demography in 9 years of seedling dynamics data from Yasuní National Park, Ecuador, a lowland rain forest where previous studies have demonstrated habitat preferences among adult trees. I included 136 species or morphospecies from multiple annual seedling cohorts with known age of recruitment.
3. Approximately 90% of the species examined demonstrated negative or positive associations with one or more topographic habitats in their recruitment, growth and/or mortality at some point in the study, and approximately 60% of species had significant associations in at least half of the census periods studied. The survival of newly recruited seedlings varied among seedlings in response to topographic gradients, indicating the potential for species to partition habitat niches at a young stage.
4. There was significant inter-annual variation in seedling habitat associations, indicating the characteristics of the topographic niche important to seedling performance change through time. This variability alone can contribute to the maintenance of species diversity through storage effects. While associations may also be weak or ephemeral, the seedling dynamics for many species supported the possibility that associations seen in adult populations develop through differential mortality across habitats as seedlings.
5.Synthesis. That species’ seedlings perform differently among topographic habitats and that these differences are detectable very early on in a plant’s life indicate the potential for the abiotic environment to mediate or exaggerate the roles of other mechanisms in influencing the composition of the understorey seedling assemblage.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Nevertheless, a number of authors have revitalized arguments in favour of the importance of niche specialization (Silvertown 2004; John et al. 2007) or community assembly via environmental filtering of species with traits favourable to survival in a particular habitat (Kraft, Valencia & Ackerly 2008; Kraft & Ackerly 2010). There have been several recent examples with strong evidence of habitat associations in tropical tree species across topographic gradients (Webb & Peart 2000; Harms et al. 2001; Valencia et al. 2004; Gunatilleke et al. 2006; Queenborough et al. 2007), degrees of topographic heterogeneity (Potts et al. 2004), soil types (Cannon & Leighton 2004), or gradients of soil minerals and nutrients (John et al. 2007) in large-scale plots covering between 4.5 and 52 hectares of tropical forest. These associations were typically assessed using static measures (e.g. relative species abundance at one time) and were common among species, but were only examined in adults and saplings, generally > 1 cm DBH. Although habitat associations were widespread among species in these studies, the authors also concluded that habitat associations were not sufficient to represent the primary mechanism maintaining high levels of diversity in tropical forests. Many species were generalists or showed neutral associations with habitats (but see Gunatilleke et al. 2006), and species with significant associations also demonstrated significant distributional overlap with other species.
If species disperse everywhere yet exhibit distinct patterns of habitat association (i.e. niches), plant species must have differential success across environmental gradients in seedling establishment, growth and survival that leads to changes in abundance in different habitats. Such an effect may be difficult to detect in the field, as species do not have universal dispersal to reach all sites, site characteristics may vary through time, and the majority of seedlings tend to occur near the parent, likely within a similar habitat (Nathan & Muller-Landau 2000). Thus, the spatial patterns of species’ seedling abundance caused by limited dispersal can be difficult to differentiate from spatial patterns caused by differential performance in response to environmental variation. In addition, the environmental characteristics that adults or saplings respond to may occur at quite different scales from those relevant to seedlings because of the strong asymmetric competition for resources imposed on seedlings by their larger neighbours (Wright 2002). The survival or growth of small seedlings may be more influenced by microtopography or the local light environment over small areas (square centimetres or metres) than by features measured at the scales relevant to adults (perhaps tens of square metres). Nevertheless, for the patterns to be evident in adult assemblages, there must be some relationship between the characteristics shown to correlate to adult associations and to seedling performance through time, despite the mismatch in scales.
Few studies have considered habitat specialization during early life stages. Webb & Peart (2000) examined habitat associations in seedlings ≥ 5 cm tall using relative abundance indices. In comparison with habitat associations demonstrated by saplings and adults, they found that fewer species demonstrated associations as seedlings. Comita, Condit & Hubbell (2007) examined distributions of established seedlings, ≥ 20 cm tall, and compared their habitat associations to patterns found in trees ≥ 1 cm DBH in Panama, where the prevalence of habitat associations had already been established (Harms et al. 2001). Using relative abundance as a measure of habitat preference, only one-third of the species that showed associations as adults also demonstrated them as seedlings. The results from both studies are consistent with a scenario where adult habitat associations develop over years of differential growth and survival across habitats. In such a case, the static patterns of abundance in younger stages may not reflect species’ habitat associations, while tracking individuals through time across habitats would reveal differential survival that eventually causes the association patterns seen in older life stages. Indeed, patterns of seedling mortality tracked across wet and dry seasons in Panama suggest that associations may develop because of differences in water availability across habitats and drought sensitivity across species (Comita & Engelbrecht 2009).
Here, I present an analysis of 9 years of seedling dynamics data from a 50-hectare forest dynamics plot in Yasuní National Park, Ecuador, that examines the evidence for habitat specialization at the seedling stage by tracking changes in seedling abundance and performance across habitats and through time. I examined recruitment, growth and survival of over 18 000 seedlings of known age from 136 woody species to ask: (i) Do seedlings at Yasuní demonstrate significant associations with topographic habitats or performance biases across habitats in growth or survival? (ii) Do species vary in their response to topography to indicate niche partitioning among species? The long-term and community-level data set permits an examination of these questions across eight annual seedling cohorts from the time of seedling establishment through survival and growth over multiple years for many species of trees, shrubs and lianas.
Yasuní is an ideal site to examine the role of habitat associations in maintaining forest diversity because it is extremely species-rich and because there is ample evidence at older life stages that tree species display differential abundances across habitats: the composition of saplings and adults has been shown to vary among topographic habitats (Valencia et al. 2004; Queenborough et al. 2007), and 40% of species’ distributions are significantly correlated with the distribution of soil nutrients (John et al. 2007). Valencia et al. (2004) determined that approximately one-quarter of Yasuní’s > 1100 tree species were habitat specialists as adults or saplings, one-quarter were generalists, and the remainder differed in abundance among habitats, but without strong enough patterns to permit categorization of habitat preferences. Evidence from the functional traits of species and phylogenetic relationships among tree species at Yasuní also suggests a strong role for environmental filtering of species assemblages (Kraft, Valencia & Ackerly 2008; Kraft & Ackerly 2010). Here, I investigate whether patterns exhibited by adults develop from early patterns of differential growth and survival across habitats by seedlings.
Materials and methods
This research was conducted inside the Yasuní Forest Dynamics Plot (FDP), a 50-ha plot that is part of a network of large-scale, long-term forest dynamics plots coordinated by the Center for Tropical Forest Science (CTFS) at the Smithsonian Tropical Research Institute (STRI). The Yasuní FDP is located in the northern part of Yasuní National Park in eastern Ecuador (0°41′ S, 76°24′ W). Within the plot, all stems ≥ 1 cm DBH have been mapped, identified and measured for girth and have been censused every 5 years, starting in 1995.
Yasuní’s evergreen lowland wet forest lies within the upper Amazon basin, one of the most species-rich regions in the world (Gentry 1992). There are approximately 1100 tree species in the 50-ha Yasuní FDP (Valencia et al. 2004), and levels of up to 300 tree species per hectare have been observed in the area (Valencia, Balslev & Paz y Miño 1994). Yasuní has an aseasonal climate, receiving a mean annual rainfall of 3081 mm, with constant temperature throughout the year (Losos & Leigh 2004). The plot lies at 230 m above sea level, and there is a 33.5 m difference between the plot’s highest and lowest points. Losos & Leigh (2004) describe the site and the Yasuní FDP in greater detail.
Seedling Census Design
In 2002, I established 600, 1-m2 seedling census plots within the Yasuní FDP (Metz, Sousa & Valencia 2010) in association with a network of 200 seed traps that had been previously established as part of other ongoing phenological studies at Yasuní (Persson 2006). Together, a seed trap and its adjacent seedling plots comprise a seedling census station. The stations are arrayed systematically every 13.5 m along the trails, on alternating sides of the trail and at random distances of 4–10 m into the forest from the trails (See Fig. S1a).
In annual June–July censuses of the plots from 2002 to 2010, I mapped, identified and marked every seedling recruited into the plots and measured the height of all marked individuals. The 2002 baseline census included all individuals < 1 cm DBH that were already established in the plots and thus are of an unknown age. I restricted the following analyses, however, to include only seedlings that were recruited into the study plots over the eight census intervals following the 2002 baseline census and thus are of known age. I refer to these hereafter as ‘seedlings’, although species may differ greatly in stature or the length of time cotyledons are retained. ‘Recruits’ are seedlings that enter the annual seedling census for the first time in particular year. Data from the annual censuses permit estimation of species-specific and age-specific rates of recruitment, growth and survival.
Each 20 × 20 m quadrat in the Yasuní FDP has been assigned a mean elevation (m), slope from the horizontal (degrees) and an index of convexity (see detailed descriptions in Harms et al. 2001 and Valencia et al. 2004). Valencia et al. (2004) used these to assign habitat categories to the western 25 ha of the Yasuni FDP, and I extended these categories to the eastern 25 ha of the study area, as the network of seedling census stations traverses the entire 50-ha area. Valencia et al. (2004) initially examined five topographic habitats, but condensed these to three categories when the species composition in two pairs of habitats showed as much within-habitat similarity as between-habitat similarity, making the pairs indistinguishable. Following the conventions of their study, I assigned each 20 × 20 m quadrat to one of the three condensed topographic habitats: (i) ridge, combining ridge-top (elevation ≥ 227.2 m, slope < 12.8° and convexity > 0) and high-slope (elevation ≥ 227.2 m, slope ≥ 12.8° and convexity > 0); (ii) slope, combining high gully (elevation ≥ 227.2 m, slope ≥ 12.8° and convexity < 0) and low-slope (elevation < 227.2 m and slope ≥ 12.8°); and (iii) valley (elevation < 227.2 m and slope < 12.8°). While their topographic categorizations result in few habitat types, the topography in Yasuní and other forests is significantly correlated with a number of soil nutrients and resources that affect plant distributions (John et al. 2007), so that these simple topographical categories may stand as proxies for many different environmental niches. In addition to the topographical categories, there is a former oil company helicopter landing circle in the northwestern portion of the plot, dominated by the pioneer Cecropia sciadophylla, which is classified as secondary forest. None of the seedling census stations was located in this area, and quadrats with that classification were not used in these analyses.
Seedling census stations were categorized according to the habitat category of the 20 × 20 m quadrat in which they were located. With these classification criteria, approximately 40%, 25% and 33% of the quadrats in the Yasuní FDP are assigned to the valley, mid-slope and upper ridge habitats, respectively (Fig. S1). The distribution of the seedling census stations samples these habitats in representative proportions (χ2 = 12, d.f. = 9, P = 0.2133).
I restricted the analyses presented here to seedlings of known age, or only those seedlings that recruited after the baseline census in 2002, grouped into eight annual cohorts. Only a portion of the > 600 species or morphospecies identified as seedling recruits in this study have sample sizes large enough to allow rigorous statistical analyses. I restricted the analyses to the 136 species or morphospecies that recruited new seedlings into at least 10 different census stations over the course of the study. These included trees, shrubs and lianas from 94 genera and 44 families (Table S1). The most species-rich group was the legumes (Fabaceae), with 24 species included in the analyses, 12 of which were in the genus Inga. The palms (Arecaceae) were also abundant with eight species, including one of the most abundant trees in the Yasuní FDP, Iriartea deltoidea. Spatial and temporal heterogeneity in recruitment was largely responsible for limiting the number of species that met our sample size criterion.
The tests described later focus on two separate age classes of seedlings. The torus permutation tests examine general seedling patterns by including both young-of-the-year seedling recruits and older seedlings ≤ 50 cm height that have survived one or more years following recruitment. The generalized linear mixed effects survival model focuses only on newly recruited seedlings to standardize comparisons among seedlings of the same age.
Tests of Habitat Associations Using Torus Permutations
I conducted torus permutational tests of habitat associations or performance bias separately for each species in each census year or interval. Not all species were abundant in every census, so analyses were restricted to species that had seedlings occurring in at least five census stations during the focal census (for analyses of abundance) or at the start of the census interval (for analyses of growth or survival). Although mortality for newly recruited seedlings is higher than for established seedlings, some individuals survive to the next census interval, and the number of stations at which they occur can increase through recruitment. In addition, the 2005 census had far higher recruitment than was observed in other years (Metz et al. 2008), and the number of species meeting the sample size requirements also increased. Thus, the number of species analysed tended to increase each year.
For each species that fit the minimum sample size requirements in each census, I calculated the abundance, growth rate and survival rates at each census station and averaged station-level measures across stations within a habitat. There were two measures of abundance: seedling density (m−2) and relative abundance, or the number of the focal species’ seedlings divided by the total number of seedlings at the station. To assess whether a species’ performance varied among habitats, potentially accounting for its differential distribution, I used three measures of seedling success, standardized to annual rates: absolute growth in height (cm; measured as height of apical meristem from soil surface), relative growth (the difference in the logarithm of the height measurements) and survival (measured as the proportion of seedlings present at a station in one census that survive to the following census). Annual growth rates were averaged across all seedlings of the focal species to obtain a station-level average.
Apparent differences in seedling abundance among habitats may be a simple coincidence of the spatial structure of the topography and spatial patterns of seedling distribution and not a result of species’ preferences because both the topography and the distribution patterns of seedlings are spatially autocorrelated. The initial distributions of seedlings may continue to influence seedling growth or survival if, for example, related individuals are dispersed near each other and perform more similarly than seedlings from other parents. I used torus translations of the topography, as described in Harms et al. (2001), to decouple the spatial coincidence of the topography and overlying seedling distributions, while preserving the spatial structure of each. In this method, the habitat map was shifted in 20-m steps to the north or east while retaining the same spatial structure of the true topography. As the habitat map ‘fell off’ the end of the FDP boundaries, the boundary wrapped around and connected to the other side, as in a torus (Fig. S1b). Each of the 1250 possible translations of the map was also mirrored and/or inverted (Fig. S1c, d, e), leading to 5000 permutations of the map, one of which was the true habitat map. The seedling census network was overlaid on each map, and measures of seedling abundance or performance in each habitat calculated on every translated map created a null distribution that I compared to the values observed on the original ‘true’ map. If the observed values were in the extremes of the null distribution, the species was considered to have a significant association with that habitat. Examining habitat associations in this way is a two-tailed test of ‘local’ significance (sensuCannon & Leighton 2004) that determines whether seedling density, growth or survival in a habitat on the true map was more extreme than 97.5% of the seedling performance values measured in that habitat on all 5000 maps. A significant positive or negative habitat association or performance bias indicated that a species was significantly more/less abundant or grew significantly faster/slower or survived significantly better/worse in a habitat than would be expected from a chance spatial alignment of the seedling distributions with the topography of the plot.
Partitioning Topographic Niche Axes
I used a generalized linear mixed effects model (GLMM) to examine the relationship between topography in the Yasuní FDP and the survival of newly recruited seedlings, and to understand the potential for species to partition topographic niche axes as seedlings. The analysis included all recruit cohorts of seedlings from the 136 focal species. There are seven cohorts of known-age seedlings in the data set, recruiting in the 2003–09 censuses, for which performance can be tracked until 2010. Here, I present the results of 3-year survival for the five cohorts that have been in the study for this period. Similar models for 1-year performance (seven cohorts) or 5-year performance (three cohorts) had qualitatively similar results and are not presented here.
Survival was modelled using the logit transformation of the survival probability for individual seedlings and a binomial error structure. The independent predictor variables were continuous topographic variables (mean elevation, slope and convexity, as described above) for the 20 × 20 m quadrat in which the focal seedling’s census plot was located, and the initial height (cm) of the newly recruited seedling, from the soil surface to the apical meristem. Each of these variables was centred by subtracting the mean prior to analysis. I included species identity in the model as a random factor and allowed species to vary in their response to each topographic variable (the slope or intercept of the relationship between survival and, for example, mean elevation). This allowed species to have positive or negative relationships with topography (indicated by direction and strength of the parameter estimate) or to vary in their baseline mortality rates (the intercept of the relationship). Large variation among species would indicate partitioning of topographic niches. To account for the spatial non-independence of seedlings located within the same plot, and within plots at the same census station, I included the random effect of plot nested within station. To account for differences in overall survival across cohorts due to annual variation in climate or other factors, I also included the cohort identity for each recruit. This was included as a fixed factor in the model because the low number of replicate cohorts precluded decent estimation of this effect as a random factor.
I confirmed that the model structure including plot and census station location accounted for any potential spatial autocorrelation not related to topography (e.g. common light environments or biotic factors at a plot) by examining correlograms of the model’s Pearson residuals and a Moran’s I test for spatial autocorrelation. No residual spatial autocorrelation was detected.
Measures of abundance, growth and survival varied among habitats and years for all species combined (Fig. 1). The patterns were similar whether examining seedlings of all known ages combined or only the young-of-the-year recruits (Fig. S2), although the latter examination included fewer species. On average, the density of each species was lower than one individual per m2 and ranged from 0.5 ± 0.07 m−2 (mean ± SE) in ridge stations in 2003 to 1.5 ± 0.7 m−2 in valley stations in 2005. In 2003–05, the mean density of a species was highest in the valleys but in other years was higher in the slope or ridge habitats (Fig. 1a). As more seedlings recruited into the study, the mean relative abundance of any one species decreased, especially in the 2005 census which had higher recruitment overall (Metz et al. 2008).
Overall growth rates were quite variable among intervals, and the ranking among habitats changed often for both absolute and relative growth rates (Fig. 1b). Absolute growth rates ranged from 0.8 ± 0.5 cm y−1 in ridge stations over 2003–04 to 3.0 ± 1.1 cm y−1 (mean ± SE) in slope stations over the 2005–06 interval.
Annual survival rates ranged from 56.8% ± 2.5 (mean ± SE) in valley stations over the 2007–08 interval to 76.7% ± 3.4 in ridge stations over the 2003–04 interval. In general, the rates differed little among habitats or years, although survival tended to be lower in the valley overall (Fig. 1c).
Tests of Habitat Associations Using Torus Permutations
The vast majority of species analysed did show significant habitat associations (Table 1). At least one-third and up to three quarters of the species had a significant positive or negative association measured by their relative or absolute abundance with at least one habitat type in any given year (Table 1), and many species had similar associations over multiple years (Table S2). There were more significant associations measured by density than by relative abundance in every year. Far fewer species, 5.6–22.7%, had differential success among habitats measured by their absolute or relative growth rates over each census interval (Table 1), but when success was measured by rates of survival, approximately half of the species had a significant association with at least one habitat (Table 1). Generally, species associations were not evident in each census they were tested, and, although most associations were consistent within a habitat, there were some species that switched between positive and negative associations in different years (Table S2). As the duration of the study increased, and more annual seedling cohorts could be included, there were more species with sufficient sample sizes for analysis and more species that showed at least one significant association (Fig. 2). The proportion of the total species demonstrating at least one significant positive or negative association levelled off at approximately 90% (for measures of density or survival) once five or more annual cohorts were included (Fig. 2); 60% of species had significant associations in four or more of the tested census periods.
Table 1. Percentage of species demonstrating significant habitat associations in torus permutation tests. The number of species tested in each census or over each interval and the percentages of those species that showed a significant positive or negative association with at least one habitat type as measured by a) relative abundance of the species in a habitat (%) or density (seedlings per m2); b) relative growth rates (log(cm) year−1) or absolute growth rates (cm year−1); and c) survival rates (proportion surviving year−1). Species may have an association with more than one habitat type. See Table S2 for detailed summary of species’ responses
(b) Growth rates
(c) Survival rates
Partitioning Topographic Niche Axes
Topographic descriptors (measured by slope, convexity and elevation) were not significant predictors of overall recruit survival because there was great variation among individual species in their response to the topography (Table 2, Fig. 3). The large variation among species in the intercept of the survival relationship indicated differences in the baseline mortality rate, and the variation among species in the slope parameter estimates (both the direction and strength) for the relationship with topography demonstrated the potential for partitioning of these niche axes. In particular, the response to the topographic slope was quite variable (Table 2, Fig. 3). Many species’ survival appeared insensitive to changes in the topographic slope, while others survived better or worse as slopes steepened. Species that had a particularly sensitive response to slope and tended to survive better on steeper slopes also tended to be found across a wider range of slopes (including the steepest) than were species that survived better in flatter areas and tended to be found over a more narrow range of slopes. There was also significant variation among cohorts in overall survival, indicating the importance of inter-annual variation in the response to topography.
Table 2. Survival of newly recruited seedlings over 3 years. Results of a generalized linear mixed effects model with binomial errors where survival for 3 years following recruitment was predicted by initial seedling height and the continuous topographic variables of slope, convexity and mean elevation of the 20 × 20 m quadrats where seedling plots were located. Random effects to account for spatial non-independence included the seedling plot nested within the census station and species identification. The slope and intercept of the relationship between survival and topography were permitted to vary among species. The recruit cohort was included as a fixed effect to account for overall differences in survival across years. Seedling height and the topographic variables were each centred prior to analysis. See Fig. 3 for illustration of variation among species in response to the slope of the area
Model AIC 6045.2.
Examples of Species’ Patterns
Both sets of analyses demonstrated significant variation among species and years in the relationship between seedling dynamics and topography. Some species demonstrated qualitatively similar patterns in abundance and performance year to year, while other species were much more variable (Table S2). For example, Cedrelinga cateniformis tended to have a negative association with or performance biases in the valley and a positive association with slope. Other species, like Pourouma minor, Matisia malacocalyx or Talisia novogranatensis, had positive associations in some years and negative associations in other years in the same habitat type. Associations detected by patterns of abundance often matched performance biases detected in growth or survival rates (e.g. Brownea grandiceps in the valley, or Pouroma tomentosa on slopes and ridges). In contrast, survival rates in some species were higher than expected by chance in habitats where the species also tended to be more rare (i.e. a positive association measured by survival rates, but a negative association measured by abundance), such as Inga umbractica or Tapirira guianensis in the slope habitat. The result for T. guianensis was confirmed in the GLMM analysis of survival of seedling recruits where this species survived better with increasing slope, but was not found on the steepest of slopes (Fig. 2). The results of all torus permutation tests are available in Supplementary Table S2.
Differences in Performance Across Habitats
The results from this multi-year study demonstrate that young seedlings of most species show differential abundance and performance across topographic habitats. Abundance, growth rates and survival rates for the seedling assemblage as a whole varied across the ridge, slope and valley habitats, but these patterns often differed among years. Topography is significantly correlated with many soil nutrients at Yasuní and elsewhere and also stands as a proxy for many soil characteristics that are important resources for young plants, such as water availability and drainage, and soil bulk density (Silver et al. 1994; John et al. 2007). The availability of different resources may interact to create an environment that changes from year to year.
The majority of species showed non-random patterns of associations with one or more habitat, such that abundances or survival rates in one habitat were significantly higher or lower relative to other habitats than would be expected by a chance relationship between seedling performance and habitat characteristics. In addition, early seedling survival varied significantly among species in response to variation in topographic variables, suggesting that species may partition topographic niches. That these local habitat associations are evident in the first few years of a tree’s existence, indeed even in young-of-the-year recruits, indicates a strong potential for environmental characteristics to play an important role as a filter for species distributions through increased survival in some habitats and decreased survival in others. If individuals of some species accumulate preferentially in certain habitats, the bank of shade-tolerant seedlings in understorey, or the ‘advanced regeneration’, will differ among habitats. There is generally high survival of understorey seedlings and saplings existing at the time of gap formation, so that canopy turnover through gap regeneration may strongly depend on advanced regeneration rather than new recruitment into gaps (Uhl et al. 1988; Fraver, Brokaw & Smith 1998; Brokaw & Busing 2000). Habitat effects that influence the composition of seedlings in the understorey, such as the differential survival among habitats demonstrated here, could therefore play a role in determining the composition of the canopy, especially as the effects of these processes accumulate over the many years required for gap turnover.
Tracking Individuals across Years
I hypothesized that evidence of habitat associations would be stronger in seedlings when measuring performance biases in growth or survival because these rates would reflect the environmental filtering of species in habitats that are more or less suitable to a species’ traits, while patterns of seedling abundance would reflect broader dispersal to many habitats. Contrary to my expectations, I found that strong habitat associations were least prevalent when measured by growth rates and found as frequently in survival rates as in the measures of abundance. Both survival rates and measure of abundance provided a strong signal of differential performance across habitats for a large proportion of species. Once established, rain forest seedlings spend many years as advanced regeneration in the understorey (Connell & Green 2000). They may need to survive long enough to take advantage of several cycles of canopy openings and closings before reaching the overstorey (Grubb 1977; Delissio et al. 2002). These results show that for many species, seedlings survive differently across habitats and thus may have more opportunities to recruit into the overstorey in some habitats than in others.
In contrast to survival rates, growth rates provided much less evidence for habitat associations in seedlings. While differences in growth should also be important in creating patterns of habitat associations, seedling growth is difficult to measure without error, and seedling growth rates are notoriously variable, with much noise hindering the detection of any biological signal (Baraloto & Goldberg 2004). Seedlings suffer mechanical damage from falling debris or passing animals that may be random with respect to the topographic habitat. Further, herbivory and browsing alone are enough to make negative growth rates very common in seedlings. Previous work at Yasuní and other forests showed that assemblage-wide seedling growth rates fit a bi-exponential distribution because of the symmetric distribution of negative and positive growth rates and the long tails in each (Metz et al. 2008). Seedlings were most often recruiting into and persisting in the very shaded microsites of the forest understorey (and rarely into light gaps), so growth may not be expected to provide a very strong signal. The confounding influences of a shaded understorey and damage from biotic or other physical factors lead to a noisy mixture of positive, negative and stagnant growth rates that may swamp signals of any positive or negative effects of the topographic environment on seedling growth.
Comparison of Seedling and Adult Patterns of Habitat Association
Given the habitat associations exhibited by adults at Yasuní (Valencia et al. 2004; John et al. 2007; Queenborough et al. 2007), I expected seedling performance to provide evidence for the filtering effect of habitat through differential growth and survival across habitats. Although Valencia et al. (2004) did not directly assess species-specific habitat preferences, the habitat associations I found are similar to some of the patterns in adult dominance and distributions at Yasuní in their study. For example, Rinorea lindeniana, Yasuní’s second most common tree ≥ 1 cm DBH, is the first and third most dominant species in the ridge and slope habitats, respectively, but is less abundant in the valley habitat (Valencia et al. 2004). Similarly, R. lindeniana seedlings were significantly less abundant in the valley habitat in most years, measured by density and/or relative abundance, and tended towards lower growth rates and survival rates at valley census stations (Table S2). Two of the most common species at Yasuní, Iriartea deltoidea and Eschweilera coriacea, are dominant in all habitats (Valencia et al. 2004), and similarly, seedlings of these species did not generally have differential performance among habitats. While the patterns of seedling abundance and performance were suggestive of adult associations for many species, there were also several species where seedling patterns did not immediately appear to reflect adult patterns.
Queenborough et al. (2007) used torus translations to test habitat associations for adults and saplings in the Myristicaceae family, and the seedling habitat associations for six of those species were also studied here, permitting a direct comparison of patterns (Table S2). Iryanthera hostmannii adults did not have significant associations with topographic habitats, and accordingly, the seedlings tended to perform equally among habitat. Otoba glycycarpa seedlings survived significantly better in slope and valley habitats, but tended to be less common in these habitats, although the adults had significant positive associations there (Queenborough et al. 2007). Virola elongata, V. ‘microfuzzy’ and V. pavonis had no significant associations as adults (Queenborough et al. 2007) and, as seedlings, varied between positive and negative associations in all habitats (Table S2). Virola duckei seedlings tended to be significantly less abundant, but have significantly higher survival rates, in slopes and valleys (Table S2) although the adults were significantly negatively associated with valleys and positively associated with slopes (Queenborough et al. 2007). For these six Myristicaceae species, the patterns of differential seedling abundance and performance among habitats observed here support the hypothesis that adult associations could develop from differential mortality at the earliest life-history stages and result in the patterns of association observed by Queenborough et al. (2007).
Comita, Condit & Hubbell (2007) showed that the habitat associations of adult trees might be reinforced through local dispersal, simply due to the greater numbers of reproductive adults present in the preferred habitat, and this increased local dispersal could overwhelm other negative effects that adults are hypothesized to have on nearby juveniles (Janzen 1970; Connell 1971; Wright 2002). Local associations in the first or even first few years of a plant’s life span may simply be the result of this localized recruitment, so that one could argue that the torus translation test cannot distinguish between a species that disperses everywhere but only establishes in the preferred habitat and a species that only disperses to the preferred habitat because that is where the adults are located. Although this study only follows individuals once they have germinated and cannot speak to processes occurring between the time of seed dispersal and the annual seedling census period, every species analysed here was found in all three of the habitats types and across a gradient of each of the continuous topography measures. Because associations were equally prevalent when measured by abundance and survival rates (but not growth), these associations reflect species filtering by the environment and not just the legacy of dispersal patterns. However, the focus of the Valencia et al. (2004) study on comparing species composition among habitats does not permit more than a brief qualitative comparison between the two sets of results. Additionally, Valencia et al. (2004) found many species to be difficult to characterize as specialists or generalists although their distributions clearly differed among habitats. Because trees are so long-lived, it will be important to examine the trends in tree mortality over longer time frames than the one-time picture of abundance distributions they studied to better understand whether these species have habitat preferences.
Implications of Habitat Associations for Species Diversity
Clearly, the environment can have a strong influence on young seedling performance, which could indicate a role for niche differentiation in the maintenance of diversity. Upwards of 90% of species had abundance distributions or patterns of survival exhibiting positive or negative associations with one or more habitats during the study (Fig. 2). The cumulative proportion of species exhibiting significant associations or performance biases did not reach a plateau until five or more annual cohorts of seedling recruits were included in the study, stressing the necessity of long-term, community-wide monitoring to understand the importance and prevalence of habitat associations for a number of species. Each additional year of observation increased both the number of species studied and the number of species for which habitat associations were detected (Fig. 2).
Tracking seedling performance over a number of years also revealed that habitat associations were, for many species, ephemeral or even variable across years (Table 2, Table S2). This inter-annual variation sets the stage for coexistence to be maintained through storage effects and the lottery model where recruitment success varies with fluctuations in the environment, but long-lived adult trees survive through periods of low recruitment and will again experience conditions favourable to recruitment success (Warner & Chesson 1985). In this way, time represents another axis along which species might vary in their response to the topographic habitat.
Many components of the niche that are correlated with topography may vary through time, leading plant species to respond differently to the same habitat in different years. Although topography has been shown to be a useful surrogate for many correlated resource axes (Hall et al. 2004; John et al. 2007), changes in climatic or other environmental conditions among years may affect soil resources in variable ways (Sollins 1998), leading to differential responses of seedlings to the same topographic habitat in different years. In addition, populations of herbivores or pathogens might fluctuate in response to inter-annual variation in climate, or in response to changing characteristics of the abiotic environment or host populations. This means that biotic interactions with natural enemies could differ among years within the topographic habitat because of habitat characteristics (Fine, Mesones & Coley 2004) or species abundance (Comita et al. 2010; Mangan et al. 2010; Metz, Sousa & Valencia 2010).
This study illustrates the widespread occurrence of habitat associations at the seedling stage for large numbers of species in a very diverse forest. The differential dynamics across topographic habitats observed here contrast with expectations of neutral models where species have equivalent fitness and stochastic fluctuations in abundance as a result of limited dispersal (Hubbell 2001). Whether the effects of these associations are strong enough to have a lasting imprint on the composition of the adult forest requires further study. The seedling dynamics of many species were consistent with patterns of adult associations (Valencia et al. 2004; Queenborough et al. 2007), but dynamics of other species fluctuated or were inconsistent so as to suggest a weak role in determining forest composition. Further research is needed to determine whether absent or weak associations are due to a scale mismatch whereby seedlings respond to environmental variation at much smaller scales than the 20 × 20 m quadrats studied here. The generality of adult habitat associations detected at this scale should also be tested at larger spatial scales beyond the 50-ha plot. It is possible that the weak effects of seedling habitat associations accumulate over repeated cycles of seedling recruitment and survival in the shaded understorey, so that the composition of individuals filling canopy gaps is strongly influenced by the topographic niche. Because associations fluctuated across years, however, it is also possible that the impact on forest composition of seedling responses to the topographic environment is reversed or overwhelmed by other processes at later life stages. Nevertheless, seedling habitat associations are stabilizing mechanisms (sensuChesson 2000) that may make important contributions to the maintenance of species diversity in tropical forests.
I thank the people and government of Ecuador for protecting their forests and making them available for study. R. Valencia graciously provided the topographic information for the 50-ha plot, and I am grateful to him and all the many people involved in the establishment and maintenance of the Yasuní 50-ha forest plot. The Yasuní FDP is managed by the Pontifical Catholic University of Ecuador and has been generously supported by the government of Ecuador (Donaciones del Impuesto a la Renta), STRI, the US National Science Foundation (NSF), the Andrew W. Mellon Foundation, and the University of Aarhus of Denmark. An NSF Graduate Research Fellowship supported me during the establishment of this project. W. Sousa, D. Ackerly, J. Battles, R. Condit, C. D’Antonio, P. Fine, K. Harms, M. Rees and one anonymous reviewer provided very helpful comments on the manuscript. I thank D. Armitage and M. Daugherty for discussions about GLMM analyses. Completion of the study would not have been possible without funding from several generous sources including CTFS and NSF (Dissertation Improvement Grant DEB-0407956, LTREB Grant EF-0614525) and the University of California at Berkeley. For assistance with logistics and data collection, I thank N. Garwood, R. Valencia, S. J. Wright, E. Zambrano, M. Zambrano, J. Suarez, F. Hopkins, C. Hayden, A. Hartley, L. Zambrano, G. Grefa, R. Grefa and P. Alvia.