Seasonal drought limits tree species across the Neotropics

A central challenge for ecologists and biogeographers is to understand how climate controls large-scale patterns of diversity and species composition. Climate-related gradients in diversity observed by some of the earliest tropical biogeographers, including the global latitudinal diversity gradient itself (von Humboldt 1808, Wallace 1878), are often attributed to the physiological limitations of taxa imposed by climate conditions (Dobzhansky 1950). This idea is expressed in the ‘physiological tolerance hypothesis’ (Janzen 1967, Currie et al. 2004), which posits that species richness varies according to the tolerances of individual species to different climatic conditions. Thus, species able to withstand extreme conditions are expected to be widely distributed over climatic gradients, while intolerant species would be constrained to less physiologically challenging locations and have narrower geographical ranges. An implicit assumption of this hypothesis is that species' realized niches tend to reflect their fundamental niches, and a key implication of the hypothesis is that past, present, and future distributions of species will tend to track changes in climate (Boucher-Lalonde et al. 2014).

Within the tropics tree diversity varies considerably, possibly as a consequence of variation in water supply (ter Steege et al. 2003). Water-stress is indeed one of the most important physiological challenges for tropical tree species (Engelbrecht et al. 2007, Brenes-Arguedas et al. 2011), and precipitation gradients correlate with patterns of species richness at macroecological scales (Clinebell et al. 1995, ter Steege et al. 2003). In particular, tree communities in wetter tropical forests tend to have a greater number of species than in drier forests (Gentry 1988, Clinebell et al. 1995, ter Steege et al. 2003). If this pattern were driven by variation among species in the degree of physiological tolerance to dry conditions, then we would predict that all tropical tree species could occur in wet areas whilst communities at the dry extremes would be made up of a less diverse, drought-tolerant subset. Thus, we would expect a nested pattern of species' occurrences over precipitation gradients, characterised by widespread dry-tolerant species and small-ranged species restricted to wet environments. In this paper we refer to this scenario as the dry tolerance hypothesis (Fig. 1a).

Alternatively, nestedness may not be the predominant pattern for tropical tree metacommunities over precipitation gradients. Multiple studies have documented substantial turnover in floristic composition over precipitation gradients in tropical forests (Pitman et al. 2002, Engelbrecht et al. 2007, Quesada et al. 2012, Condit et al. 2013). This pattern could be driven by a trade-off between shade-tolerance and drought-tolerance (Markesteijn et al. 2011, Brenes-Arguedas et al. 2013). Whilst drought-tolerant species tend to have a higher capacity for water conductance and CO₂ assimilation under water-limiting conditions, they grow more slowly in the scarce understory light of wet forests where shade-tolerant species have a competitive advantage (Brenes-Arguedas et al. 2011, 2013, Gaviria and Engelbrecht 2015). Drought-tolerant species are also apparently more vulnerable to pest damage in moist areas (Baltzer and Davies 2012, Spear et al. 2015). Thus, in less physiologically stressful environments, tropical tree species' occurrences could be limited by stronger biotic interactions, both with competitors and natural enemies (MacArthur 1972, Normand et al. 2009). In a scenario in which both wet and dry limitations to species distributions are equally important, we would expect progressive turnover of species' identities along precipitation gradients (cf. Fig. 1b), rather than the nested pattern described above.

Both nested and turnover patterns have to some extent been documented in the tropics. A nested pattern has been detected in the Thai-Malay peninsula where widespread species, occurring across both seasonal and aseasonal regions, are more resistant to drought than species restricted to aseasonal areas (Baltzer et al. 2008). Across the Isthmus of Panama, Engelbrecht et al. (2007) found a direct influence of drought sensitivity on species' distributions, whilst light requirements did not significantly limit where species occur, which is consistent with the mechanisms underlying a nested pattern of species distributions. Also in Panama, another experimental study found that pest pressure was similar for species regardless of their distribution along a precipitation gradient (Brenes-Arguedas et al. 2009), indicating that the distributions of taxa that occur in drier forests may not be constrained by pest pressure. However, recent data from the same area show that drought-tolerant species are more likely to die than drought-intolerant taxa when attacked by herbivores or pathogens (Spear et al. 2015). Furthermore, when comparing two sites, an aseasonal (Yasuní; ca 3200 mm yr^–1 rainfall) and seasonal (Manu; ca 2300 mm yr^–1) forest in lowland western Amazonia, Pitman et al. (2002) reported that a similar proportion of species were unique to each site (Yasuní, 300 exclusive species out of 1017; Manu, 200 out of 693). The presence of a similar and large proportion of species restricted to each site is consistent with species distributions showing a pattern of turnover among sites. While there is thus evidence of both nestedness and turnover in tropical tree species distributions, a comprehensive investigation at large scale is lacking.

There are various approaches to estimate the tolerance of taxa to water-stress. For example, experimental studies of drought imposed on trees provide the clearest indicator of sensitivity to water-stress and provide insight into the ecophysiological mechanisms involved. Yet in the tropics, these are inevitably constrained to a minor proportion of tropical diversity, limited by tiny sample sizes (Nepstad et al. 2007, da Costa et al. 2010) and practical challenges of achieving any spatial replication and of integrating effects across multiple life stages (Brenes-Arguedas et al. 2013). By contrast, observational approaches, which consist of mapping species' distributions across precipitation gradients, could potentially indicate the sensitivity of thousands of species to dry or wet conditions (Slatyer et al. 2013). Fixed-area inventories of local communities from many locations, offer a particular advantage for this kind of study as they avoid the bias towards more charismatic or accessible taxa that affects ad hoc plant collection records (Nelson et al. 1990, Sastre and Lobo 2009). Inventory-based attempts to classify tropical tree taxa by their affiliations to precipitation regimes have already advanced the understanding of species precipitation niches (Butt et al. 2008, Fauset et al. 2012, Condit et al. 2013), but have been fairly limited in terms of spatial scale, number of sample sites and taxa. In this paper we apply this inventory-based approach to investigate the macroecological patterns of trees across the world's most species-rich tropical forests, those of the western Neotropics, an area of 3.5 million km² that encompasses Central America and western South America. Because species richness in this region is so high, meaning that individual species' identifications are often challenging, we also explore whether analyses at the genus – or family – level offers a practical alternative for assessing the impacts of water-stress on floristic composition.

We selected the western Neotropics as our study area for two reasons. First, there is substantial variability in climate at small spatial scales relative to that of the entire region, meaning that associations between precipitation and floristic composition are less likely to be the result of dispersal limitation and potential concomitant spatial autocorrelation in species' distributions. The Andean Cordilleras block atmospheric moisture flow locally, maintaining some areas with very low precipitation levels, whilst enhancing orographic rainfall in adjacent localities (Lenters and Cook 1995). As a result, there are wetter patches surrounded by drier areas across the region, such as the wet zones in central Bolivia and in south east Peru (Fig. 2). The inverse is also observed, such as the patches of drier forests south of Tarapoto in central Peru. There is also a general tendency for precipitation to decline away from the equator in both northward and southward directions (Fig. 2). Secondly, the western Neotropics is a cohesive phylogeographic unit. Western Amazonian forests are floristically more similar to forests in Central America than to those in the eastern Amazon, despite the greater distances involved and the presence of the world's second highest mountain range dividing Central America from southern Peru (Gentry 1990). This floristic similarity between the western Amazon and Central American forests is thought to be because: 1) the Andes are young (∼ 25 Ma) so represent a recent phytogeographic barrier (Gentry 1982, 1990), and 2) the soils of moist forests in western Amazonia and Central America are similar, being young, relatively fertile, and often poorly structured, largely as a consequence of the Andean uplift and associated Central American orogeny (Gentry 1982, Quesada et al. 2010).

Here, we use a unique, extensive forest plot dataset to investigate how precipitation influences the distribution of tree taxa, at different taxonomic levels, across the western Neotropics. Using 531 tree plots that include 2570 species, we examine the climatic macroecology of the region's tropical trees. Specifically, we 1) test the dry tolerance hypothesis, which posits that tolerance to dry extremes explains taxa geographic ranges within closed-canopy forests (Fig. 1a); and 2) quantify the affiliations of taxa to precipitation using available data, in order to assess individual taxon-climate sensitivities and predict how tropical trees may respond to potential future climatic changes.

Methods

Analyses

Precipitation and diversity

If water supply broadly limits species' distributions, then community-level diversity should also be controlled by precipitation regime. However, variation in local diversity is nevertheless expected as a consequence of other factors (ter Steege et al. 2003). For example, even under wet precipitation regimes, local edaphic conditions such as extremely porous soils could lead to water stress and lower diversity. Therefore, we fitted a quantile regression (Koenker and Bassett 1978), describing the role of precipitation in controlling the upper bound of diversity. Diversity was quantified using Fisher's α because this metric is relatively insensitive to variable stem numbers among plots. In addition, to assess whether the correlation between diversity and precipitation is robust to the potential influence of spatial autocorrelation we applied a Partial Mantel test (Fortin and Payette 2002), computing the relationship between the Euclidian distances of diversity and precipitation, whilst controlling for the effect of geographic distances. Lastly, we also used Kendal's τ non-parametric correlation coefficient to assess the relationship between diversity and precipitation. We restricted all diversity analyses to the 116 1-ha plots that had at least 80% of trees identified to species level.

Metacommunity structure

We used the approach of Leibold and Mikkelson (2002) to test whether the distribution of taxa along the precipitation gradient follows a turnover or nested pattern. Our analysis was performed by first sorting the plots within the community matrix by their precipitation regimes. Then we assessed turnover by counting the number of times a taxon replaces another between two climatologically adjacent sites and comparing this value to the average number of replacements found when randomly sorting the matrix 1000 times. More replacements than expected by chance indicate a turnover structure, whilst fewer imply that the metacommunity follows a nested pattern (Presley et al. 2010) as predicted by the dry tolerance hypothesis. This analysis was conducted applying the function ‘Turnover' from the R package ‘metacom’ (Dallas 2014).

Precipitation and taxa distribution

To explore the influence of precipitation on taxa distributions firstly, we simply plotted taxa precipitation ranges, i.e. the range of precipitation conditions in which each taxon occurs, to visually inspect the variation of precipitation ranges among taxa. According to the dry tolerance hypothesis, for each taxon the precipitation range size should be positively associated with the driest condition at which it is found, i.e. the more tolerant to dry conditions the taxon is, the larger its climatic span should be. However, the predicted pattern could also arise artefactually if taxa that occur under extreme regimes have on average bigger ranges regardless of whether they are associated to dry or wet conditions. We therefore, secondly, used Kendall's τ coefficient of correlation to explore analytically the relationship between taxon precipitation range and both the driest and wettest CWD values at which each taxon occurs. If the dry tolerance hypothesis holds we expect precipitation range size to be negatively correlated with the driest precipitation condition where each taxon occurs and not correlated with wettest precipitation where each taxon is found.

Thirdly, we compared taxa discovery curves, which represent the cumulative percentage of taxa from the whole metacommunity that occur in each plot when following opposite environmental sampling directions, i.e. from wet to dry and from dry to wet. The dry tolerance hypothesis predicts that wet to dry discovery curves should be steeper initially than dry to wet curves, as wet areas are expected to have more narrow-ranged taxa.

Finally, we examined the loss of taxa from extremely wet and from extremely dry plots over the precipitation gradient. We tested whether tree taxa found at the driest conditions within our sample can tolerate a larger range of precipitation conditions than taxa in the wettest plots. We thus generated taxa loss curves to describe the decay of taxa along the precipitation gradient within the 10% driest plots and the 10% wettest plots.

We compared discovery and loss curves in different directions of the precipitation gradient (i.e. from wet to dry and from dry to wet) against each other and against null models of no influence of precipitation on taxa discovery or loss. These null models represented the mean and confidence intervals from 1000 taxa discovery and loss curves produced by randomly shuffling the precipitation values attributed to each plot. Taxa recorded in 10 plots or fewer are likely to be under-sampled within the metacommunity and were excluded from the analyses regarding metacommunity structure and taxa distribution.

Taxa precipitation affiliation

To describe the preferred precipitation conditions for each taxon we generated an index of precipitation affiliation, or precipitation centre of gravity (PCG). We adopted a similar approach to that used to estimate the elevation centre of gravity by Chen et al. (2009) (see also Feeley et al. 2011), which consisted of calculating the mean of precipitation of locations where each taxon occurs in, weighted by the taxon's relative abundance in each community (Eq. 1).

(1)

where: n = number of plots; P = precipitation; Ra = relative abundance based on number of individuals.

The resulting taxon-level PCG values are in units of millimetres per year, the same scale as the precipitation variables: CWD or MAP. We tested the null hypothesis of no influence of precipitation on the distribution of each taxon by calculating the probability of an observed PCG value being higher than a PCG generated by randomly shuffling the precipitation records among the communities, following Manly (1997) (Supplementary material Appendix 4). We also generated an alternative estimator of precipitation affiliation for each taxon by correlating its plot-specific relative abundance and precipitation values using Kendall's τ coefficient of correlation (following Butt et al. 2008). Here, a negative correlation indicates affiliation to dry conditions, whilst a positive correlation indicates affiliation to wet conditions (Supplementary material Appendix 6).

PCG values were calculated for each taxon recorded in at least three localities (1818 species, 544 genera and 104 families), and Kendall's τ values were calculated for each taxon recorded in at least 20 localities (525 species, 327 genera and 78 families). We also calculated the proportions of significantly dry- and wet-affiliated taxa. To verify that these proportions were not merely a consequence of the number of taxa assessed, we compared our observed proportions to 999 proportions calculated from random metacommunity structures where taxa abundances were shuffled among plots (Supplementary material Appendix 5).

Each analysis was repeated at family, genus and species levels. All analyses were performed for CWD, and precipitation affiliations were also calculated for MAP. Analyses were carried out in R ver. 3.1.1 (R Core Team).

Results

In the western Neotropics, diversity was negatively related to water-stress at all taxonomic levels, being strongly limited by more extreme negative values of maximum climatological water deficit (CWD) (Fig. 3). This result remained after accounting for possible spatial autocorrelation (Partial Mantel test significant at α = 0.05 for all taxonomic levels: r = 0.31 for species; r = 0.38 for genera; r = 0.37 for families). The large increase in diversity towards the wettest areas was most evident at the species level (around 200-fold), but was also strong at genus (ca 70-fold) and family levels (ca 16-fold) (Fig. 3).

For all our analyses of taxa distributions it was evident that they follow a nested pattern along the water-deficit gradient, as predicted by the dry tolerance hypothesis. Thus, firstly, when investigating metacommunity structure, among any given pair of sites, the number of times a taxon replaced another was significantly lower than expected by chance at all taxonomic levels (Table 1). Secondly, compared to all taxa, those able to tolerate the dry extremes were clearly distributed over a wider range of precipitation regimes (Fig. 4a–c). This was confirmed by precipitation ranges being very strongly and negatively correlated to the driest condition where each taxon occurs (Kendall's τ = –0.93 for species, –0.96 for genera and –0.99 for families, one-tailed p-values < 0.001) and not correlated to the wettest condition of occurrence (Kendall's τ = 0.01 for species, 0.05 for genera and – 0.01 for families, p-values > 0.05).

Thirdly, nested patterns were evident in most taxa discovery curves, with the floristic composition of dry plots being a subset of wet plots (Fig. 4d–f). At species and genus levels, the wet–dry cumulative discovery curves were steeper than the dry–wet curves, indicating more taxa restricted to wet conditions. However, this distinction in the shape of the discovery curves between the directions of the precipitation gradient (wet–dry vs dry–wet) was much less evident at the family level (Fig. 4f). Finally, the loss curve analysis also showed that plots at the wet extremes of the precipitation gradient have many more taxa restricted to wet conditions than expected by chance (Fig. 4g–i). Extreme dry plots also had a much greater proportion of species with wide precipitation ranges than the wettest plots, with at least 80% of their species persisting until all but the very wettest forests are reached (Fig. 4g – red curve). Again, these patterns were most clearly evident for species and genera.

For the 1818 species, 544 genera and 104 families assessed across the western Neotropics, we found a large proportion of taxa with significant values for rainfall affiliation (Table 2aa, Supplementary material Appendix 9, Table A9.1, A9.2 and A9.3). Affiliations to wet conditions were substantially more common than affiliations to dry conditions at all taxonomic levels (Table 1b) (Supplementary material Appendix 5). Anacardiaceae and Rutaceae are examples of the 10 most dry-affiliated families registered in 10 or more localities and Lecythidaceae, Myrsinaceae and Solanaceae are amongst the most wet affiliated families (see Supplementary material Appendix 7, Table A7.1 and A7.2 for the most wet and dry affiliated taxa). Lastly, the observed patterns persisted when repeating the analyses excluding those species possibly affiliated to locally enhanced water supply (Supplementary material Appendix 8).

Discussion

Our results demonstrate the influence of precipitation gradients on the patterns of diversity and composition for families, genera and species of Neotropical trees. We confirm that community diversity is much higher in wet than in drier forests, being as much as 200-fold greater at the species level (Fig. 3). Additionally, our analyses indicate that the diversity decline towards more seasonal forests is a consequence of increasingly drier conditions limiting species distributions. To our knowledge this is the first time that the influence of precipitation affiliation has been quantified at the level of individual Amazon tree species.

Water-stress during the dry season, represented here by the climatological water-deficit (CWD), limits tree species distributions across the western Neotropics (Fig. 4). In areas with a very negative CWD, forest composition is a subset of those communities that do not suffer water-stress (Fig. 4). These findings are consistent with results from studies at much smaller scales (Engelbrecht et al. 2007, Baltzer et al. 2008). The physiological challenges in dry areas require species to have specific characteristics in order to recruit and persist. For example, certain species have the capacity to maintain turgor pressure and living tissues under more negative water potentials at the seedling stage, which allow them to obtain water from dry soils (Baltzer et al. 2008, Brenes-Arguedas et al. 2013). At the wet extreme of the gradient, more favourable conditions may allow a wider range of functional strategies to coexist (Spasojevic et al. 2014). Consistent with this, most taxa in our data set occur in the wet areas, with only a small proportion restricted to dry conditions (Fig. 4). Furthermore, our results indicate that other factors such as pests and pathogens (Spear et al. 2015) or tolerance to shaded environments (Brenes-Arguedas et al. 2013), are much less important in determining the distribution of taxa. In some cases these may restrict the abundance of dry affiliated taxa but generally appear not to limit their occurrence. Geomorphology and dispersal limitation can impact species' distributions, and these drivers likely account for some of the unexplained variation in the relationship between diversity and precipitation shown here (Higgins et al. 2011, Dexter et al. 2012). The scarcity of plots from the very wettest forests (Supplementary material Appendix 3, Fig. A3.2) may also have limited our ability to fully document patterns of species turnover. Nevertheless, our analysis shows that more than 90% of the species occurring in the driest 10% of the neotropical forest samples are also registered in at least one forest with zero mean annual CWD (Fig. 4g). It could be argued that such widespread taxa may not necessarily tolerate dry conditions, but instead be sustained by locally enhanced water supply due to particular conditions such as the presence of streams. However, our results were robust even after excluding taxa potentially affiliated to such local water availability (Supplementary material Appendix 8). Thus, our findings, together with those from Asian and Central American tropical forests (Baltzer et al. 2008, Brenes-Arguedas et al. 2009), suggest that the limitation of most tree species' distributions by water-stress may represent a general macroecological rule across the tropics. This has obvious parallels to the well-known pattern for temperate forest tree species, for which frost tolerance substantially governs species' geographical ranges (Pither 2003, Morin and Lechowicz 2013).

Affiliations to specific precipitation regimes are strongest at the species level, but climate sensitivity can still be clearly detected with genus-level analyses (Fig. 4d–i). The stronger relationship between species and precipitation when compared to other taxonomic levels could be a consequence of a relatively stronger influence of climate on recent diversification. In particular, massive changes in precipitation regimes took place in the Neogene and Quaternary due to Andean uplift and glacial cycles (Hoorn et al. 2010). During this period, global fluctuations in climate and atmospheric CO₂ concentrations, which affect water-use efficiency (Brienen et al. 2011), are thought to have influenced speciation (cf. Richardson et al. 2001, Erkens et al. 2007, although see Hoorn et al. 2010). Climate sensitivity was also clearly evident at the genus level (Fig. 4), which has relevant practical implications for tropical community and ecosystem ecology. Because of the challenges of achieving sufficient sample size and accurate identification in hyperdiverse tropical forests (Martinez and Phillips 2000), ecosystem process and community ecological studies in this ecosystem often rely on the simplifying assumption that the genus-level represents a sufficiently functionally-coherent unit to address the question at hand (Harley et al. 2004, Laurance et al. 2004, Butt et al. 2014). Our results suggests that analysis at the genus-level could be used to assess, for instance, the impacts of climate change on diversity, but that nevertheless such impacts would be underestimated without a species-level analysis.

In addition to the physiological tolerance to dry conditions, other, underlying geographical and evolutionary processes could conceivably drive the patterns we observe in this study. These are, notably, 1) a greater extent of wet areas (Terborgh 1973, Fine 2001), 2) greater stability of wet areas through time leading to lower extinction rates (Klopfer 1959, Jansson 2003, Jablonski et al. 2006), and 3) faster rates of speciation in wet forests (Rohde 1992, Allen et al. 2002, Jablonski et al. 2006). The first alternative (Rosenzweig 1992) requires that species–area relationships govern the climate-diversity associations that we find. Within our region, the areas that do not suffer water-stress (i.e. CWD = 0) are where the great majority of the species (90%) can be found (Fig. 4), yet they occupy a relatively small area (25% of the western Neotropics and 31% of plots). Thus, the area hypothesis appears unlikely to be driving the precipitation–diversity relationship.

The other two alternative hypotheses could more plausibly be contributing to the patterns observed here. Climate stability is indeed associated with diversity throughout the Neotropics (Morueta-Holme et al. 2013). In contrast with most of the Amazon basin, the lowland forests close to the Andes and in Central America apparently had relatively stable climates, with only moderate changes during the Quaternary/Neogene (Hoorn et al. 2010), which could have reduced extinction rates (Klopfer 1959, Jablonski et al. 2006). The diversity gradient may also be a consequence of more diverse areas having higher diversification rates (Rohde 1992, Jansson 2003, Jablonski et al. 2006). While both lower extinction rates and higher speciation rates in wet forest might contribute to explaining the climate-diversity gradient, their influence does not invalidate the idea that wet-affiliated species are drought-intolerant. Indeed, the mechanisms that might have favoured lower extinction rates in wetter forests are related to the inability of many taxa to survive environmental fluctuations such as droughts. Experiments showing that seedlings of species from wet tropical environments have higher mortality under water-stress than dry-distributed taxa (Engelbrecht et al. 2007, Baltzer et al. 2008, Poorter and Markesteijn 2008) indicate that water stress can have direct impacts on species survival and distribution. As ever, untangling ecological and historical explanations of patterns of diversity is difficult with data solely on species distributions (Ricklefs 2004).

Supplementary material (Appendix ECOG-01904 at < www.ecography.org/appendix/ecog-01904 >). Appendix 1–9.