Local and regional palm (Arecaceae) species richness patterns and their cross-scale determinants in the western Amazon


  • Thea Kristiansen,

    1. Ecoinformatics and Biodiversity Group, Department of Biological Sciences, Aarhus University, Build. 1540, Ny Munkegade 114–116, DK-8000 Aarhus C, Denmark
    Search for more papers by this author
  • Jens-Christian Svenning,

    1. Ecoinformatics and Biodiversity Group, Department of Biological Sciences, Aarhus University, Build. 1540, Ny Munkegade 114–116, DK-8000 Aarhus C, Denmark
    Search for more papers by this author
  • Dennis Pedersen,

    1. Ecoinformatics and Biodiversity Group, Department of Biological Sciences, Aarhus University, Build. 1540, Ny Munkegade 114–116, DK-8000 Aarhus C, Denmark
    Search for more papers by this author
  • Wolf L. Eiserhardt,

    1. Ecoinformatics and Biodiversity Group, Department of Biological Sciences, Aarhus University, Build. 1540, Ny Munkegade 114–116, DK-8000 Aarhus C, Denmark
    Search for more papers by this author
  • César Grández,

    1. Facultad de Ciencias Biológicas, Universidad Nacional de la Amazonía Peruana, Iquitos, Peru
    Search for more papers by this author
  • Henrik Balslev

    Corresponding author
    1. Ecoinformatics and Biodiversity Group, Department of Biological Sciences, Aarhus University, Build. 1540, Ny Munkegade 114–116, DK-8000 Aarhus C, Denmark
      Correspondence author. E-mail: henrik.balslev@biology.au.dk
    Search for more papers by this author

Correspondence author. E-mail: henrik.balslev@biology.au.dk


1. Local and regional patterns of plant species richness in tropical rain forests, as well as their possible drivers, remain largely unexplored. The main hypotheses for local species richness (alpha diversity) are (i) local environmental determinism with species-saturated communities, and (ii) regional control, in which the immigration of species from the regional species pool (gamma diversity) determines how many species coexist locally. The species pool hypothesis suggests a combined influence of local and regional drivers on alpha diversity. Differences in gamma diversity may arise from divergent environmental conditions or biogeographic histories.

2. We investigated the cross-scale determinants of palm alpha and gamma diversity across the western Amazon using a large field-based data set: a census of all palm individuals in 312 transects, totalling 98 species. We used regression-based variation partitioning to understand how habitat, topography and region influence alpha diversity, and correlations to assess the importance of the present environment (climate, soil, regional topography) and history (long-term habitat stability) for average regional alpha diversity and gamma diversity, including the link between these two diversity measures (species pool effect).

3. Variation in alpha diversity was primarily explained by region (36%) and habitat (18%), whereas the effect of topography was negligible (1%). Within habitats, region was even more important (up to 69% explained variation). Within regions, habitat and topography covaried and had a variable but an important influence. The pure effect of topography remained of minor importance (up to 13%).

4. Average regional alpha diversity was related to gamma diversity, precipitation seasonality and possibly long-term habitat stability. Gamma diversity was related to long-term habitat stability, and possibly current climate.

5.Synthesis. Gamma diversity strongly influenced alpha diversity, although a clear influence of local environment was also evident, notably habitat type, with a minor, more geographically variable effect of small-scale topography. Apart from gamma diversity, the factor most strongly related to regional alpha diversity was precipitation seasonality, while gamma diversity itself was strongly linked to long-term habitat stability. These results imply that plant species richness is contingent on both contemporary and historical factors with a strong link between local species richness and the regional species pool.


Plant species richness varies greatly across the surface of the Earth. The highest richness is found in the tropics (e.g. Kreft & Jetz 2007), but much less is known about how richness varies within hyper-diverse rain forests (Pitman et al. 2002; Tuomisto, Ruokolainen & Yli-Halla 2003; Myster 2009). For example, why do adjacent sites sometimes differ widely in diversity? What determines larger scale richness gradients? These questions can only be answered by taking the scale dependency of species diversity and its controlling factors into account (Whittaker 1960; Willis & Whittaker 2002; Ricklefs 2004, 2008). Based on Whittaker (1960), we can partition species diversity into the species richness of local sites (alpha diversity), the turnover in species composition among sites (beta diversity) and the species richness of larger regions (gamma diversity). However, larger regions may also be characterized by their average alpha diversity, or regional alpha diversity. Although the exact scale limits are somewhat arbitrary, this framework is useful for investigating the patterns and drivers of species diversity (Ricklefs 2008).

The aim of the present study was to investigate possible drivers of species richness across scales in a hyper-diverse tropical rain forest area. We addressed a number of hypotheses at different spatiotemporal scales within a common framework, distinguishing among three levels of species richness: local alpha diversity (locality-specific alpha diversity), regional alpha diversity (average alpha diversity across a given region) and gamma diversity (Fig. 1).

Figure 1.

 Hypothesized causal pathways for factors influencing species richness across scales at three levels of diversity. Factors with variation on the smallest scale are shown to the left with progressively larger scale variation to the right. The scale of the factors matches the scale of the hypothesized underlying process, except for climate and soil fertility, for which the influence on regional alpha diversity is hypothesized to act via local mechanisms. Arrows indicate the hypothesized direction of effects. Branched arrows indicate that the effects can be either positive or negative; otherwise all effects are hypothesized to be positive. Vertical, dashed lines indicate one major and two minor divisions in scale. See the text for detailed mechanism explanations.

For local alpha diversity, two overarching, yet conflicting, points of view about the determinants of richness have been debated, namely local determinism and regional control. Local determinism entails niche assembly and the saturation of local communities, with the local environment and biotic interactions controlling species richness (e.g. Wright 1983; Tilman 2004). Among the many proposed mechanisms, habitat type and local topographic heterogeneity are often considered important due to environmental species sorting and more heterogeneous areas providing larger niche space, thereby promoting local species coexistence (Svenning 2001; Phillips et al. 2003) (Fig. 1). Environmental factors that vary on regional scales (climate, soil fertility, environmental heterogeneity) are also hypothesized to influence local species coexistence, but because they act similarly on all local communities within a region, they statistically contribute to the estimated regional effects. Regional control entails dispersal assembly in which species richness is ultimately driven by regional-scale speciation, extinction and dispersal dynamics. Local communities are not saturated with species, and their richness is determined by proportional sampling from the regional species pool (e.g. Ricklefs 1987; Harrison & Cornell 2008), as would be the case under neutral community assembly (Hubbell 2001). These two viewpoints represent extremes on a gradient from complete local control to complete regional control. The species pool hypothesis (Zobel 1997) posits an intermediate situation in which large-scale processes determine which species are available for local coexistence and interact with small-scale processes, such as the local environment and biotic interactions, to determine which species actually coexist.

For regional alpha diversity, the potential drivers include factors with a uniform influence on local processes and migration rates within regions, as well as species pool effects. The potential drivers of gamma diversity include factors with a uniform influence on local processes within regions and, independently hereof, factors influencing large-scale speciation, extinction and dispersal dynamics (cf. Harrison & Cornell 2008). Some of the primary proposed regional factors are climate, long-term habitat stability, soil conditions and large-scale topography. Gradients in the current climate are often used to explain large-scale species richness patterns (e.g. Kreft & Jetz 2007) with a number of proposed underlying mechanisms (Currie et al. 2004), (Fig. 1): (i) favourable climate conditions and fertile soils could increase productivity, which could support more individuals and, thus, more species, leading to a positive effect on both regional alpha diversity and gamma diversity (cf. Currie & Francis 1999; Currie et al. 2004). Higher productivity could also increase forest dynamics, thereby promoting local coexistence (Phillips et al. 1994), leading to a positive effect on both regional alpha diversity and gamma diversity; (ii) species sorting according to physiological tolerance could have a positive effect on both regional alpha diversity and gamma diversity in areas with favourable climate conditions for many taxa (Currie et al. 2004); (iii) higher speciation rates in warm and wet climates (Currie et al. 2004) could have a positive effect on gamma diversity, but it is not likely to have a direct effect on regional alpha diversity. Areas with more fertile soils could also have higher speciation rates due to increased forest dynamics and shorter generation times (Stropp et al. 2009; Hoorn et al. 2010), leading to a positive effect on gamma diversity. In addition to the current climate, long-term habitat stability (Fig. 1) may have a positive effect on gamma diversity because more stable regions would have suffered fewer climate-driven extinctions during past climate change episodes, such as the Quaternary glaciations, and therefore may harbour more species today (Araújo et al. 2008; Stropp et al. 2009). Stability may also have left a mark on present-day regional species composition with fewer small-ranged species in unstable areas as a result of decreased rates of relictual survival and speciation (Dynesius & Jansson 2000; Jansson 2003; Svenning & Skov 2007). Also, in tropical forests, unstable areas may have relatively more canopy species as a result of the high drought sensitivity of understorey species, which is due to their less developed and shallower root systems (Wright 1992). Regional environmental heterogeneity is thought to increase regional alpha diversity due to mass effects (Shmida & Wilson 1985) and to increase gamma diversity due to a larger niche space in more heterogeneous areas. However, in more heterogeneous areas, dispersal limitation may also act to decrease regional alpha diversity while increasing gamma diversity due to increased beta diversity (Nekola & White 1999). Regional environmental heterogeneity can be heterogeneity in soils (Tuomisto et al. 2002) or topography (Kerr & Packer 1997). Despite the long-standing interest in diversity patterns on various scales, the questions of how local and regional processes interact to determine local and regional species richness remain open (cf. Ricklefs 2008; White & Hurlbert 2010).

Here, we focus on palm species in the lowland western Amazonian rain forest, where palms constitute an abundant and diverse component of the vegetation (Kahn & de Granville 1992; Henderson 1995, 2002). Species distributions across the region and their present and past determinants lack exploration (Tuomisto, Ruokolainen & Yli-Halla 2003; Hopkins 2007; Myster 2009; Hoorn et al. 2010), although a number of studies of plant beta diversity suggest that both the local environment and dispersal determine species composition in Amazonian plant assemblages (Condit et al. 2002; Tuomisto, Ruokolainen & Yli-Halla 2003; Vormisto et al. 2004; Normand et al. 2006). The relative importance of local and regional control of plant species richness in the region has been studied even less, although a few studies have suggested that both local and regional control is important (Bjorholm et al. 2005, 2006; Stropp et al. 2009). On a small scale, palm species distributions within the region are related to habitat type (Kahn & Mejia 1990; Kahn & de Granville 1992; Svenning 1999, 2001; Vormisto, Tuomisto & Oksanen 2004; Vormisto et al. 2004). The main habitat division for plants in general is between non-inundated forest (terra firme) and inundated forest, with the highest richness recorded in non-inundated areas (Junk 1989; Myster 2009). Local topographic heterogeneity is a well-documented structuring force for the local distribution of palm species (Kahn & Mejia 1990; Svenning 1999, 2001; Vormisto, Tuomisto & Oksanen 2004; Vormisto et al. 2004) and other woody plants (Valencia et al. 2004), for the most part probably reflecting edaphic or hydrological niche specialization (Vormisto, Tuomisto & Oksanen 2004). Several plant and palm species have also shown preferential distributions along soil nutrient gradients (Kahn & de Granville 1992; Svenning 2001; Tuomisto et al. 2002; Phillips et al. 2003). On the scale of the entire Amazon, areas of high richness have been associated with high annual precipitation (Clinebell et al. 1995) and low precipitation seasonality (Clinebell et al. 1995; ter Steege et al. 2003). The same relationships influence patterns of palm species richness across the Americas (Bjorholm et al. 2005, 2006). Only a few studies have explicitly addressed the relative importance of the factors that drive local and regional richness (Harrison & Cornell 2008); some studies have advocated the strong importance of local factors (Carneiro, Fernandes & De Souza 2005), whereas others have suggested the simultaneous importance of local and regional factors (Hortal et al. 2008; White & Hurlbert 2010).

In the present study, we analysed a novel, field-based multi-scale data set from western Amazonian palm communities (312 transects of 5 × 500 m distributed across eight regions with a total of 98 species and 340,883 individuals). We expected higher local alpha diversity in non-inundated areas and areas with more heterogeneous local topography, and differences in local alpha diversity among habitats and regions. We expected higher regional alpha diversity in areas with a larger regional species pool (gamma diversity). We expected higher regional alpha diversity and higher gamma diversity in regions with favourable climate conditions for palms (warm and wet), stable long-term habitat conditions, high soil fertility, high soil heterogeneity, high topographic heterogeneity, but with possibly counteracting effects of heterogeneity on regional alpha diversity. More specifically, we examined: (i) how much of the variation in local alpha diversity could be explained by habitat, local topographic heterogeneity and region, (ii) whether regional alpha diversity correlates with gamma diversity or current or past environmental conditions, and (iii) whether gamma diversity correlates with current or past environmental conditions.

Materials and methods

Data collection

We collected data on palm species distribution and abundance across the western Amazon in Colombia, Ecuador, Peru and Bolivia, from 4°N to 14.4°S (Fig. 2), during 1995–2009 using a standardized transect methodology. The Guaviare data set from Colombia was collected by Rodrigo Bernal in collaboration with H. Balslev and D. Pedersen. A transect was 5 × 500 m (0.25 ha) and divided into 100 subunits of 5 × 5 m. All palm individuals within a transect were recorded and identified. Whenever possible, transects were placed in tall, old-growth rain forest with the entire transect being situated within a homogeneous forest type. Topographic maps or satellite images and Google Earth (http://earth.google.com/) were used for positioning. A total of 323 transects (80.75 ha) with a total of 365, 563 palm individuals were inventoried. In the coming years, data from ongoing fieldwork will be added (http://www.fp7-palms.org/). Individual records are available on the GBIF portal (http://www.gbif.org/), and the data base serves as an analytical tool and repository for facilitating data sharing (Peacock et al. 2007) and secure data storage. Some of the data have been subject to analyses of beta diversity, species distributions and range size patterns within one or, in one instance, two regions (Vormisto et al. 2004; Boll et al. 2005; Normand et al. 2006; Balslev et al. 2009; Kristiansen et al. 2009).

Figure 2.

 Map of the study area. Left: locations of the 312 transects and eight regions studied in the western Amazon. Right: detail of the three regions in north-eastern Peru. Symbols indicate which region the transect was in: + = Guaviare; × = Yasuni; • = Iquitos; ○ = Pastaza; □ = Amazon; Δ = Contamana; ◊ = Ucayali; and bsl00066 = Madidi.

In each transect subunit (5 × 5 m), all individuals of palms were identified to the species level and assigned to one of four developmental stages: (i) seedlings, (ii) juveniles, (iii) subadults, and (iv) adults, although the present analysis was based on the total number of individuals. Voucher specimens were collected in each transect and subsequently deposited in herbaria [COL, QCA, AMAZ, USM, LPB, NY, TUR, AAU, acronyms following Thiers (continuously updated)] in Colombia, Ecuador, Peru, Bolivia, USA, Finland and Denmark (2216 specimens in total). The data base is continuously updated with relevant taxonomic revisions. The nomenclature followed Henderson (1995) except where more recent taxonomic treatments were available: Bactris (Henderson 2000), Geonoma (Borchsenius, Balslev & Svenning 2001), Astrocaryum (Kahn 2008) and Attalea (Zona 2002). All individuals were determined to the lowest possible taxonomic level, i.e. subspecies or variety in Bactris, Desmoncus, Geonoma and Wendlandiella, but they were grouped into their respective species for the present analyses (Table S1 in Supporting Information). Taxonomic uniformity was ensured via herbarium specimens and digital photographs (AAU Herbarium palm collection, http://herb42.bio.au.dk/aau_herb/default.php), and the same specialist, H. Balslev, completed all fieldwork and identification except for 32 transects where T. Kristiansen identified the palms in expeditions in which H. Balslev participated.

For the present analysis, we excluded seven transects in disturbed forests, four transects in areas with transitions between homogeneous forest types, and 4072 records (1.2%) that were not identified to the species level. Thus, the data set consisted of 312 transects (78 ha) and 340,883 individuals.

Study area and analytical design

The Amazon is one of the world’s most species-rich areas (Gentry 1988a; Kreft & Jetz 2007) and has a complex geological and climatic history (Sombroek 2000; Anhuf et al. 2006; Hoorn et al. 2010). The highest plant and palm species richness is found in the central and western areas (Gentry 1988b; Kahn & de Granville 1992; Bjorholm et al. 2005; Stropp et al. 2009) with local vascular plant species richness reaching >900 species ha−1 (Balslev et al. 1998). On the largest spatial scale, the western Amazon is characterized by a humid tropical climate with decreasing precipitation and increasing seasonality with increasing distance from the equator and towards east (Silman 2007). Soils in the western areas have developed on Cenozoic sediments from the Andean orogeny, whereas soils in the east (Guyana shield) developed on Proterozoic crystalline bedrock (Sombroek 2000). Palms are important in Amazonian forests, in terms of the number of species, number of individuals (Kahn & de Granville 1992; Henderson 1995, 2002) and life-form variation (Dransfield et al. 2008); therefore, they are well suited for investigating general determinants of tropical plant species richness.

Within this larger area, the 312 transects were sampled from four habitat types with 10–180 transects per habitat: non-inundated, inundated, white-sand and pre-montane hills (Table 1), located in eight regions, with 28–77 transects per region (Table 2, Fig. 2, Fig. S1 in Supporting Information). We defined the regions on the basis of geomorphology and location rather than spatial distance to ensure homogeneous past and present environmental conditions, so that the species constituting the regional species pool share these conditions. This approach means that some regions in well-surveyed areas were close to each other. The sharpest boundaries were around the Iquitos region with a distinct border between the Iquitos arch and the alluvial Pastaza fan (clearly visible in satellite images and in the field) and along the Amazon River (Fig. S1).

Table 1.   The four habitat types sampled across eight regions in the western Amazon, with a description, number of transects sampled and a summary of the local topography of transects within habitats
HabitatDescriptionNo. transectsLocal topography (m)
Mean ± SD [min.–max.]
  1. *White-sand was not included in within-habitat variation partitioning (Fig. 3b) due to the small sample size.

  2. †White-sand areas can be inundated or waterlogged, but those included in the present study were not.

Non-inundatedNever inundated. Clay soils. Variable terrain, ranging from areas with steep hills to flat terraces. All regions1804.53 ± 3.37 [0.22–19.70]
InundatedHigh water-table, areas either inundated from rivers or permanently waterlogged. All regions except Madidi981.26 ± 0.81 [0.00–4.60]
White-sand*White-sand soils. Very nutrient poor. Not inundated†. Only Iquitos and Amazon102.95 ± 2.17 [1.03–8.50]
Pre-montane hillsHigh-elevation pre-montane areas, either Andean foreland (Ucayali, Madidi) or inland sandstone massif (Contamana)2416.83 ± 11.6 [1.28–42.00]
Table 2.   The eight regions sampled in the western Amazon, with (a) a description and sampling information and (b) a presentation of regional factors for each region. For definitions, see Materials and methods
RegionDescription*Location†N–S extent (km) area sampled (km2)Min. and max. inter-transect distance (km)No. transect
TotalNon-inundatedInundatedWhite-sandPre-montane hills
  1. *Number in parentheses is the proportion of transects sampled in the Harmonized World Soil Database (FAO/IIASA/ISRIC/ISS-CAS/JRC 2008) major soil group.

  2. †Latitude of the northernmost and southernmost transects within a region.

  3. ‡Number of small-ranged species and canopy species is given in parentheses.

  4. §Altitude of transect midpoints. Source: 3 arc s (c. 90 m) spatial resolution SRTM data set version 4.1 (Jarvis et al. 2008).

GuaviarePart of the Guyana shield. Sandstone, crystalline bedrock. Gleysol (0.46), ferralsol (0.29), arenosol (0.25)4°0′26′′ N520.61281711
3°32′12′′ N2127159     
YasuniClose to the equator. Same tectonic plate as Iquitos, less variable soil. Acrisol (0.73), gleysol (0.27)0°39′23′′ S40.30302010
0°41′45′′ S2613     
IquitosSame tectonic plate as Yasuni, more variable soil. The Iquitos arch, eroding areas. Acrisol (0.77), gleysol (0.21), cambisol (0.03)3°53′32′′ S1840.537752169
4°33′15′′ S30093317     
PastazaThe Pastaza fan, sinking tectonic plate, alluvial-volcanic sediment, predominantly Holocene origin. Gleysol (1)3°23′1′′ S1520.25472027
4°45′19′′ S25523285     
AmazonTerraces, west of the Amazon River. Acrisol (0.79), gleysol (0.21)3°37′58′′ S1540.34282341
5°1′25′′ S3582179     
ContamanaSame tectonic plate as Ucayali. Sandy soil on the hills. Acrisol (0.38), gleysol (0.38), cambisol (0.24)6°49′30′′ S1090.244220166
7°48′38′′ S3687114     
UcayaliSame tectonic plate as Contamana. Terraces, red clay on the hills. Gleysol (0.63), cambisol (0.31), leptosol (0.06)8°28′56′′ S2580.653513148
10°48′57′′ S9512269     
MadidiHigh elevation, Andean foreland. Cambisol (0.60), leptosol (0.40)13°46′53′′ S710.13251510
14°25′16′′ S62199     
RegionAnnual precipitation (mm)Precipitation seasonality (%)Min. temperature coldest month (°C)Long-term habitat stabilityProportion small-ranged species‡Proportion canopy species‡Soil fertility (all habitats/inundated excl.)Soil heterogeneityRegional topography (m) [min.–max§]
Guaviare29884721.8Unstable0 (0)0.35 (9)2/1725 [104–182]
Yasuni31241619.7Stable0.24 (9)0.34 (13)5/5414 [217–280]
Iquitos28731420.6Stable0.21 (16)0.27 (21)1/2217 [102–169]
Pastaza22891819.9Stable0.17 (8)0.31 (15)6/6116 [110–182]
Amazon25822520.7Stable0.11 (6)0.29 (16)1/3315 [104–153]
Contamana17063019.5Stable0.09 (4)0.36 (16)4/3655 [134–174]
Ucayali18234319.1Stable0.08 (4)0.28 (14)6/6890 [163–625]
Madidi18735117.0Unstable0.05 (1)0.50 (11)3/4576 [241–548]

Local species richness was defined as the number of species found in a transect (local alpha diversity). Previous analyses of a subset of 57 transects showed that, on average, 88% of the total number of species encountered in a given transect were found after the first 250 m, and that a transect length of 500 m thus provides a good estimate of the local species pool (Kristiansen et al. 2009). Regional richness was represented in two ways: (i) mean richness per transect within regions (regional alpha diversity), and (ii) total regional richness, i.e. all species found within a region (gamma diversity). These represent two different aspects of regional diversity, with different hypothesized mechanisms (cf. Fig. 1). We computed species accumulation curves to assess the completeness of regional sampling and to illustrate the progressive accumulation of species. Accumulation curves were calculated as the mean number of species found per number of transects sampled (± SD) in 999 runs with a randomized order of transects (Fig. S2). The accumulation curves indicated thorough sampling of all regions, particularly Guaviare, Yasuni and Madidi. We also used map-derived estimates of regional richness to assess our field-based estimates of gamma diversity, and we found high accordance between the two (Spearman’s rank correlation coefficient rs = 0.905, = 0.002, Fig. S3), although our estimates were lower than the range map estimates, except for Iquitos and near identical estimates for Ucayali. Map-derived regional richness was defined as the average richness of those 1 × 1° grid cells that overlapped with the transect locations in a given region, with the richness for each grid cell derived from overlaid range maps (Henderson, Galeano & Bernal 1995) digitized by Bjorholm et al. (2005). Still, we note that gamma diversity can be biased by the variation in sampling extent and intensity among regions.

Determinants of local alpha diversity

Our predictors for local alpha diversity were habitat type, local topographic heterogeneity and region. We distinguished four broad habitat types: non-inundated terra firme forest, inundated forest (floodplain and swamp), white-sand areas and pre-montane hills (Table 1). These are main habitat types that are generally recognized in the western Amazon (Kahn & de Granville 1992; Myster 2009). Local topographic heterogeneity represents environmental heterogeneity at the smallest scale, within habitats, and it was quantified as the standard deviation in subunit altitude above the lowest point within a transect (Hofer et al. 2008). For brevity, we hereafter refer to local topographic heterogeneity simply as local topography (Table 1). Many studies have identified topography as the main factor in structuring palm distributions in the western Amazon, although they have used many different measures of this variable (e.g. Kahn & Mejia 1990; Svenning 1999, 2001; Vormisto, Tuomisto & Oksanen 2004; Vormisto et al. 2004).

Determinants of regional alpha and gamma diversity

We investigated the following potential drivers of regional alpha diversity and gamma diversity: current climate, long-term habitat stability, soil fertility, soil heterogeneity and regional topographic heterogeneity (Table 2). For regional alpha diversity, we also assessed the role of gamma diversity (cf. Fig. 1). Current climate was represented by annual precipitation (mm), minimum temperature of the coldest month (°C) and precipitation seasonality (coefficient of variation of monthly precipitation) (Hijmans et al. 2005) at the centroid of each region’s transect locations. In terms of long-term habitat stability, two regions (Guaviare and Madidi) were classified as unstable, and all other regions were classified as stable, reflecting that marginal areas of the Amazon forest biome experienced a decline in precipitation during glacial periods that resulted in periodic displacement of the rain forest by drier, more open vegetation, such as savannas. In contrast, the central parts of the western Amazon remained wet and covered by tropical rain forest, even during glacial periods (Martin et al. 1997; Mayle et al. 2004; Anhuf et al. 2006; Stropp et al. 2009). With regard to soil fertility, base saturation (BS, %), cation exchange capacity (CEC, cmol kg−1), total exchangeable bases (TEB, cmol kg−1), pH and sand fraction (% wt.) were calculated as the mean of the transect midpoint values for each region using the Harmonized World Soil Database (HWSD) (FAO/IIASA/ISRIC/ISS-CAS/JRC 2008). Regions were ranked according to each variable (low BS, CEC, TEB, pH and high sand content to high BS, CEC, TEB, pH and low sand content) and the ranks were averaged to obtain a single measure of soil fertility, with higher values corresponding to richer soil. Soil heterogeneity was calculated in the same way as soil fertility, but using the standard deviations of the transect midpoint values for each region instead of the means. Higher values corresponded to more heterogeneous soil. For descriptive purposes, we also extracted the HWSD major soil groups (FAO-90) for each transect midpoint. Regional topographic heterogeneity (hereafter referred to as regional topography) was represented by the standard deviation of the absolute altitude of transect midpoints within each region, with values extracted from the 3 arc s (c. 90 m) spatial resolution SRTM version 4.1 data set (Jarvis et al. 2008).

We created three additional variables to gain more information about the possible influence of long-term habitat stability and soil fertility on species richness. For long-term habitat stability, we calculated the proportion of small-ranged species and canopy species within regions. Small-ranged species are hypothesized to be more extinction-prone during large-scale environmental changes, such as glaciations (Jansson 2003; Svenning & Skov 2007). Small-ranged species were defined as the 25% of species with the smallest range size (Gaston 1994), and range size was defined as the number of 1 × 1° grid cells occupied across the Americas based on the Henderson, Galeano & Bernal (1995) range maps (Bjorholm et al. 2005). For 12 species not included in Henderson, Galeano & Bernal (1995), range sizes were estimated by H. Balslev based on the present data set and field experience. Canopy species are hypothesized to have deeper roots, and thus to be less drought-sensitive compared with understorey species (Wright 1992). Canopy species were defined as species that are tall enough to reach the canopy (stem height at least 15 m). For soil fertility, we excluded transects from inundated habitats because these generally have higher nutrient levels than other habitats, and we wanted to remove the potential noise introduced by some regions having more inundated transects than other regions.

Statistical analysis

We used partial regression analysis to partition the total amount of variation in local alpha diversity into fractions representing the pure effects of region, habitat, local topography and their shared effects (Legendre & Legendre 1998; Hortal et al. 2008). This partitioning was performed by subtracting the explained variation from full models including all explanatory variables and individual models including each explanatory variable. Models were built with ordinary least squares regression, and we used adjusted R2 (R2adj) to account for the different degrees of freedom per predictor variable. Region and habitat were categorical variables and entered into the analyses as multi-state nominal variables. Local topography was a quantitative variable and log-transformed to meet the assumptions of parametric statistics. We conducted four analyses: (i) variation in local alpha diversity partitioned between region and habitat, (ii) variation in local alpha diversity partitioned among region, habitat and topography, (iii) variation in local alpha diversity within habitats partitioned between region and topography, and (iv) variation in local alpha diversity within regions partitioned between habitat and topography. The subtractions for models with two explanatory variables were: R2pvar1 = R2tot − R2pvar2, R2pvar2 = R2totR2pvar1, R2shared = (R2var1 + R2var2) − R2tot and = 1 − R2tot (Legendre & Legendre 1998). R2tot is the total variation explained by the full model; R2var1 and R2var2 are the variation explained by the individual models; R2pvar1 and R2pvar2 are the variation explained purely by individual explanatory variables; R2shared is the variation jointly explained by two variables, i.e. their shared effect; and U is the unexplained variation. Fractions of variation for the model with three explanatory variables were calculated using the same statistical framework, but partitioning the variation into eight different fractions (see Appendix S1 in Supporting Information). Negative fractions were not interpreted as explained variation because they reflect the conflicting effect of correlated explanatory variables (Legendre & Legendre 1998). White-sand was not included in the within-habitat variation partitioning due to the small sample size. We note that the data sets from the different analyses differed in extent. Non-inundated habitats spanned the extent of the full data set, whereas inundated habitats, pre-montane hills and transect groups within regions had smaller spans (Table 2a).

The relationship between regional alpha diversity and gamma diversity was assessed using Spearman’s rank correlation for both all transects and only transects from non-inundated habitats (to account for the uneven distribution of habitats among regions). Relationships between regional richness measures and current climate variables, soil fertility, soil heterogeneity and regional topography were similarly assessed. We note that the low number of regions prohibited more complex analyses. Differences in richness, the proportion of small-ranged species and the proportion of canopy species between stable and unstable regions were interpreted visually because the small sample size precluded statistical evaluation. Independent data sets are one way to gain more information about such small sample sizes.

Analyses were carried out using JMP 7.0 (SAS Institute Inc., Cary, NC, USA) for variation partitioning and correlations, R 2.7.0 (http://www.r-project.org/) for species accumulation curves, and ArcGIS Desktop 9.3 (ESRI Inc., Redlands, CA, USA) for geographic data extraction and the computation of region area, centroids and inter-transect distances.


Western amazonian palm species richness

We found 98 palm species in the 312 transects (Table S1). Local alpha diversity ranged from 2 to 30 species per transect, with an overall mean of 16 species per transect. Within habitats, the highest species richness was found in non-inundated areas, in terms of both maximum and mean local alpha diversity. However, there was considerable variation, and all habitats had transects with low species numbers (Fig. 3a). Regional alpha diversity showed large variation among regions, ranging from 10.4 to 20.8 mean species per transect. Yasuni, Iquitos and Amazon had the highest regional alpha diversity, and Guaviare, Contamana and Madidi had the lowest regional alpha diversity. Also, in Guaviare, Contamana and Madidi, local alpha diversity was always less than 20 species per transect (Fig. 3b). Gamma diversity was clearly highest in Iquitos, with a total of 78 species, whereas Guaviare and Madidi had notably lower richness than the other regions, with only 26 and 22 species (Fig. 3c).

Figure 3.

 Palm species richness estimated from 312 transects (tr) sampled across four habitat types and eight regions in the western Amazon, showing (a) local alpha diversity within habitats, (b) regional alpha diversity and (c) gamma diversity (total number of species per region). (a) and (b): filled symbols = mean number of species per transect within habitats and regions; open symbols = minimum and maximum number of species per transect; lines = standard deviations. (c): filled symbols = total number of species. Habitat types were non-inundated, inundated, white-sand and pre-montane hills. Regions are ordered from north to south: GU = Guaviare; YA = Yasuni; IQ = Iquitos; PA = Pastaza; AM = Amazon; CO = Contamana; UC = Ucayali; MA = Madidi.

Local alpha diversity

Region and habitat alone explained >70% of the variation in local alpha diversity. Region had a slightly larger effect than habitat, and there was almost no overlap between the two factors (Fig. 4a). Region, habitat and local topography together explained >71% of the variation in local alpha diversity (Fig. 5). The pure effect of local topography was negligible, and habitat and local topography covaried, as seen from their shared effect and the decreased pure effect of habitat (18.5% vs. 30.9% in the two-variable model). Region and local topography showed very little covariance. Region had a large, pure effect on local alpha diversity also in the three-variable model.

Figure 4.

 Variation in local alpha diversity (number of species per transect) partitioned between two explanatory variables, showing the pure, shared and unexplained fractions as percentage of total variation. (a) The effect of region and habitat (n = 312). (b) The effect of region and local topography within habitats (from the top: = 180, n = 98, = 24). The white-sand habitat was not included due to the small sample size. (c) The effect of habitat and local topography within regions (from the top: = 28, = 30, = 77, = 47, = 28, = 42, = 35, = 25). Negative fractions are shown in parentheses. All full models, except Madidi (= 0.375), were significant (< 0.01). All individual models were significant (< 0.05) except for pre-montane hills (model with local topography, = 0.590), Pastaza (model with local topography, = 0.088) and Madidi (all models, > 0.336). preg = pure region; phab = pure habitat; ptop = pure local topography; shared = shared effect; = unexplained.

Figure 5.

 Venn diagram showing variation in local alpha diversity (number of species per transect) of 312 transects partitioned between the three explanatory variables: habitat, region and local topography. The different parts show the pure, shared and unexplained fractions as a percentage of total variation (see Appendix S1, the size of the areas in the diagram do not correspond to fractions of variation). = unexplained fraction. All models were significant (< 0.001).

Within habitats, region alone explained nearly 70% of the variation in local alpha diversity in non-inundated areas and was also important in inundated areas (58%). The smaller effect of region in pre-montane hills (40%) is also noteworthy because this habitat type was found in only three regions. Local topography had almost no influence on local alpha diversity within habitats (Fig. 4b).

Within regions, we found large differences in both the total amount of explained variation and in the pure and shared effects of habitat and local topography on local alpha diversity (Fig. 4c). The pure effect of habitat was largest in Ucayali, Amazon and Iquitos, whereas little to no pure effect of habitat was found in Yasuni and Madidi. There was a large shared effect between habitat and local topography in all regions, except Pastaza, Ucayali and Madidi. The pure effect of local topography on local alpha diversity was non-negligible in Pastaza, Guaviare and Yasuni, despite an overall low importance local topography.

Regional alpha diversity

Regional alpha diversity was positively correlated with gamma diversity, both when all habitats were included (rs = 0.71, Fig. 6) and even more so when only transects in non-inundated areas were included (rs = 0.88). Precipitation seasonality had a negative relationship with regional alpha diversity (rs = −0.74, Fig. 7c). For long-term habitat stability, the two regions that were considered unstable (Madidi and Guaviare) had low regional alpha diversity similar to that of the stable region with the lowest value (Contamana; Fig. 7e). None of the remaining factors was related to regional alpha diversity: annual precipitation (rs = 0.33, = 0.42, Fig. 7a), minimum temperature of the coldest month (rs = 0.24, = 0.57, Fig. S4a), soil fertility (rs = −0.27, = 0.53, Fig. S5a), soil fertility when inundated habitats were excluded (rs = −0.04, = 0.93), soil heterogeneity (rs = −0.70, = 0.06, Fig. S5c) and regional topography (rs = −0.64, = 0.09, Fig. S5e).

Figure 6.

 Relationships between regional alpha diversity (mean number of species per transect) and gamma diversity (total number of species per region) for eight regions sampled in the western Amazon with (a) all habitats included and (b) only non-inundated habitats included.

Figure 7.

 Scatterplots of regional alpha diversity (mean number of species per transect) and gamma diversity (total number of species per region) in the eight regions in the western Amazon showing how richness varies in relation to (a, b) annual precipitation, (c, d) precipitation seasonality (coefficient of variation of monthly precipitation) and (e, f) long-term habitat stability (regions classified as either stable or unstable according to the persistence of tropical rain forest vegetation during glacial periods; regions are shown ranked by richness). Differences in richness between stable and unstable regions were not evaluated statistically due to the small sample size. Black symbols = stable regions; grey symbols = unstable regions.

Gamma diversity

Gamma diversity was lower in the two regions that were considered unstable (Madidi and Guaviare) than in any other regions (Fig. 7f). The remaining factors were not significantly related to gamma diversity: precipitation seasonality (rs = −0.64, = 0.09, Fig. 7d), annual precipitation (rs = −0.05, = 0.91, Fig. 7b), minimum temperature of the coldest month (rs = 0.31, = 0.46, Fig. S4a), soil fertility (rs = −0.22, = 0.61, Fig. S5b), soil fertility when inundated habitats were excluded (rs = 0.03, = 0.94), soil heterogeneity (rs = −0.41, = 0.32, Fig. S5d) and regional topography (rs = −0.21, = 0.61, Fig. S5f). In addition, with regard to species composition, the unstable regions had small proportions of small-ranged species (Fig. S4c), and one region, Madidi, had a large proportion of canopy species (Fig. S4d). Unstable regions were also those with the greatest present-day precipitation seasonality (Fig. 7c,d).


We found that palm species richness within the western Amazonian lowland rain forest exhibited strong spatial patterns, on both local and regional scales. With a large spatial extent, local alpha diversity was highly and independently influenced by habitat type and region, whereas the influence of local topography was generally small. With smaller spatial extents, within regions, both habitat type and local topography influenced local alpha diversity, and there were large fractions of unexplained variation. Regional alpha diversity was strongly correlated with gamma diversity, indicating an important species pool effect, and was also linked to an aspect of current climate, namely precipitation seasonality. Long-term habitat stability seemed to influence both regional alpha diversity and gamma diversity, with a stronger link to gamma diversity.

Western Amazonian palm species richness

The Amazon region has been estimated to harbour a total of 151 palm species (Henderson 1995); thus, our 98 species from the western part make up nearly two-thirds of all Amazonian palm species. Our estimates of gamma diversity corresponded well to map-derived estimates (Bjorholm et al. 2005). The lower absolute values of our richness estimates might be due to the nature of hand-drawn range maps, which have a tendency to overestimate richness, and the variation in sampling intensity among regions. The most thoroughly sampled region, Iquitos, had more species than the corresponding range map estimate, whereas the other regions had fewer species. The accumulation curves indicated good sampling of all regions.

Determinants of local species richness

The high importance of habitat in explaining variation in local alpha diversity, with large spatial extent and in most regions (Figs 4 and 5), corroborates previous studies showing that Amazonian terra firme plant communities are richer than floodplain plant communities (Balslev et al. 1987; Kahn & de Granville 1992; ter Steege et al. 2000; Myster 2009). The variation explained by habitat can be considered a signal of species sorting, largely reflecting that only few species tolerate flooded and waterlogged conditions (Junk 1989; Myster 2009). Generally, the importance of habitat is also consistent with previous findings of relatively strong habitat specificity for palms and other plant species in the western Amazon (e.g. Kahn & de Granville 1992; Ruokolainen, Linna & Tuomisto 1997; Svenning 1999; Phillips et al. 2003; Fine, Mesones & Coley 2004; Vormisto, Tuomisto & Oksanen 2004).

Habitat and local topography covaried, especially within regions, resulting in an inability to uniquely attribute that part of the variation in local alpha diversity to either of the two variables (Legendre & Legendre 1998). A part of the high richness in non-inundated areas could thus be due to a more heterogeneous topography providing a larger niche space (Tuomisto et al. 2002; Phillips et al. 2003). Inundated and white-sand areas are generally flat, whereas non-inundated areas and especially pre-montane hills are much more heterogeneous (Table 1). Still, local alpha diversity in pre-montane hills was comparable to inundated and white-sand areas (Fig. 3a). As these hills had maximum altitudes of 625 m, their relatively low richness is not likely to be the result of an altitudinal effect on temperature and remains unexplained. Despite differences in the mean local topography among habitats, we also found large overlaps (Table 1), and the two variables represent different aspects of the local environment.

The pure effect of topography on local alpha diversity was negligible with large extents (Figs 4b and 5), but it had some importance within individual regions, particularly in Pastaza. In Pastaza, topography presumably reflects small-scale differences in flooding regime resulting from the overall flat topography and swampy nature of that region (backswamps and elevated floodplain) and, thus, particularly strong topographic species sorting within habitats (cf. Junk 1989; Normand et al. 2006). In addition, this finding illustrates that the actual ecological importance of topography (an indirect environmental factor sensuGuisan & Zimmermann 2000) may depend on the general environmental setting of a given area (Kerr & Packer 1997; Tuomisto et al. 2002; Vormisto, Tuomisto & Oksanen 2004; Hofer et al. 2008). In the present study, analysing local topography across a large spatial extent may entail dissimilar ecological significance among regions, although the basic effect of increasing niche space with increasing heterogeneity should remain constant. Given that topography is a complex variable that has been measured in several different ways, we do not draw strong conclusions from the overall low importance of topography shown here compared with the well-documented importance of topography in shaping palm species’ local distributions in the Amazon (Svenning 1999, 2001; Vormisto, Tuomisto & Oksanen 2004).

The effect of region on local alpha diversity shows that transects within a given region have similar species richness and that richness varies among western Amazonian regions. The effect is quite strong, with up to 69% of variation explained when accounting for the effect of habitat (Fig. 4b). Regional differences could reflect current environmental differences (regional factors influencing regional alpha diversity in Fig. 1) or divergent biogeographic histories influencing local richness via species pool effects (horizontal arrow from gamma diversity to regional alpha diversity in Fig. 1). The effect of region decreased with decreasing spatial extent (Fig. 4b), probably reflecting the lesser importance of underlying drivers at smaller scales (cf. Willis & Whittaker 2002). The unexplained fraction of variation was higher at smaller spatial extent, within regions and in pre-montane hills. This finding may indicate that additional factors not considered in this study were important for the variation in local alpha diversity in some of the study regions. Interpretations should be made while bearing in mind that the sampling was not balanced. For example, habitat had a very small effect in Madidi, but this region also lacked transects in inundated areas.

Determinants of regional species richness

The strongest influence on regional alpha diversity was gamma diversity. The positive relationship between regional alpha diversity and gamma diversity reflects regional enrichment (White & Hurlbert 2010), in accordance with the species pool hypothesis (Zobel 1997): the more species in the regional species pool, the more species coexist on a local scale. Still, local species richness varied considerably within regions, and even the richest regions had transects with low numbers of species (Table 2, Fig. S1). Hence, although we found clear evidence that regional drivers strongly influence local diversity, our findings also provide evidence for the simultaneous importance of smaller-scale factors. This is in accordance with previous diversity analyses on large spatial scales within (Stropp et al. 2009) and outside the Amazon (Harrison et al. 2006; White & Hurlbert 2010). Similarly, Loreau (2000) argued that, even though communities may be assembled through niche-based environmental control, local species richness is still likely to depend on regional species richness through changes in the degree of niche packing.

Two potential drivers emerged as plausible correlates of regional alpha diversity and gamma diversity, namely precipitation seasonality and long-term habitat stability. Precipitation seasonality was negatively related to regional alpha diversity, which is consistent with the importance of seasonal drought as a key factor controlling diversity in the tropics (Wright 1992; Engelbrecht et al. 2007), and the expected importance of water-related variables for palm distributions (Condit, Hubbell & Foster 1996; Bjorholm et al. 2006). For gamma diversity, precipitation seasonality neared significance and the region with the highest richness (Iquitos) was characterized not only by low seasonality, but also by high annual precipitation and a high minimum temperature of the coldest month compared to the region with the lowest richness (Madidi; Fig. 7, Fig. S4). Thus, our findings are not inconsistent with some influence of current climate also at this scale, in line with previous continental-scale studies (Bjorholm et al. 2005; Kreft & Jetz 2007). Still long-term habitat stability seemed to influence present-day regional richness patterns with a particularly strong pattern for gamma diversity; the two climatically unstable regions, Madidi and Guaviare, had lower gamma diversity than any other regions. In the Amazon, Quaternary climate changes have caused shifts in the precipitation regimes. Concurrent shifts in vegetation have occurred, with areas of evergreen forest being replaced by semi-deciduous, dry forests and savannas at the margins and in the eastern part, whereas the central western Amazon should still have received sufficient rainfall to allow the persistence of evergreen forest (Mayle et al. 2004; Anhuf et al. 2006; Stropp et al. 2009). These historical dynamics could be particularly important for palms because they are associated with high moisture levels (Bjorholm et al. 2005, 2006), and because most palms are understorey species and, thus, drought-sensitive (Wright 1992). Similar to our findings, Stropp et al. (2009) concluded that high regional tree diversity in the Amazon is associated with Quaternary climate stability. The importance of stability was supported by the smaller proportions of small-ranged species in unstable regions, as recent studies of other organism groups and regions have found small-ranged species to be concentrated in regions that were relatively unaffected by the Pleistocene ice ages (Jansson 2003; Svenning & Skov 2007; Araújo et al. 2008). We also predicted a greater proportion of canopy species in palm communities in more unstable regions due to their presumed generally greater tolerance of drought; however, this was the case in only one of the unstable regions (Madidi), perhaps reflecting a greater importance of current climate for this aspect of palm community composition.

We speculate that the somewhat contrasting effects of precipitation seasonality and long-term habitat stability reflect the hypothesized differences in causal pathways (cf. Fig. 1): both current climate and stability-driven historical dynamics may influence regional alpha diversity and gamma diversity, but these drivers should not be equally important for different levels of diversity. Precipitation seasonality is hypothesized to influence richness via local mechanisms acting within local communities, hence the stronger link to regional alpha diversity. Stability is hypothesized to directly influence gamma diversity (with indirect species pool effects on regional alpha diversity), hence the stronger link to gamma diversity. The small sample size did not allow us to clearly differentiate which factors are more important at what scale, but we note that the non-random distributions of species among regions point to the importance of large-scale dynamics. Some species occurred in all eight regions, whereas other species occurred predominantly in certain regions, and all regions had at least one exclusive species (e.g. Itaya amicorum in Iquitos; Table S1).

Soil fertility was unrelated to regional alpha diversity and gamma diversity, and we found no support for an influence on species richness via productivity or long-term variation in speciation conditions (cf. Phillips et al. 1994; Currie & Francis 1999; Stropp et al. 2009; Hoorn et al. 2010) (Fig. 1). This finding contrasts a previous study by Bjorholm et al. (2006), which showed that palm species richness across the Americas is positively related to soil fertility. Also, high Neotropical palm species richness has been related to speciation rates (Svenning et al. 2008). However, the small sample size at the regional level argues for caution. The species-poor Guaviare region is part of the Guiana shield, where infertile soils have been proposed to cause low plant speciation rates as a result of less dynamic forests (Stropp et al. 2009). Thus, Guaviare illustrates the problem of separating covarying explanatory variables, especially with small sample sizes. The low richness in this region could be the result of high precipitation seasonality, long-term habitat instability and/or a low soil-induced speciation rate.

Similarly, regional environmental heterogeneity, estimated by soil heterogeneity and regional topography, was unrelated to regional alpha diversity and gamma diversity. Therefore, none of the hypothesized mechanisms was supported (arrows indicating mass effects, niche space and dispersal limitation in Fig. 1). We note that soil heterogeneity on a finer grain than represented by the HWSD soil data may play a role in regional richness differences, as many species are non-randomly distributed along soil nutrient gradients (Kahn & de Granville 1992; Svenning 2001; Tuomisto et al. 2002; Phillips et al. 2003). Notably, Iquitos had the highest gamma diversity, partly because of species specialized on white-sand soils (Euterpe catinga, Mauritiella aculeata and Mauritia carana). Richness has generally been proposed to increase with increasing topographic heterogeneity when topographic heterogeneity relates to the inclusion of large-scale habitats, such as mountainous areas and lowlands (e.g. Nekola & White 1999; Rahbek & Graves 2001). This aspect was not assessed in the present study because we only sampled lowlands and a few relatively low-lying pre-montane areas.


Harrison & Cornell (2008) argued that there is a particular need for better data at scales of 101–106 km2 to progress from the scarcity of studies directly addressing the relative influence of local and regional determinants of species richness. The sample size and extent of the data set in the present study made it possible to address questions regarding local and regional drivers at exactly these scales. Local alpha diversity in western Amazonian palm communities was determined not only by local habitat and topography but also by strong regional effects. Although the inherently small sample size at the regional level precluded strong inference at this scale, regional alpha diversity, and thus the regional effects on local species richness, was strongly linked to gamma diversity in accordance with the species pool hypothesis. Apart from gamma diversity, the factor most strongly related to regional alpha diversity was precipitation seasonality, while gamma diversity itself was strongly linked to long-term habitat stability. This influence of large-scale historical dynamics reinforces an increasing emphasis on integrated studies of species richness drivers across spatial and temporal scales (Hoorn et al. 2010). Thus, processes acting on both small and large scales need to be taken into account to understand how the extraordinary Amazonian biodiversity has been generated and distributed (cf. Ricklefs 2008; Svenning et al. 2008).


We gratefully acknowledge permission from Rodrigo Bernal (Instituto de Ciencias Naturales, Universidad Nacional de Colombia) to use data from Rio Guaviare, which were collected with permission from the Corporación para el Desarrollo Sostenible del Norte y el Oriente Amazónico (CDA). We thank the two anonymous referees whose comments greatly improved the manuscript. Our palm research is funded by the Danish Natural Science Research Council with a grant to H.B. (272-06-0476), a WWF/Novozymes Research Grant to H.B., and the European Community FP7 programme (grant agreement 212631). T.K.’s PhD study is supported by the Faculty of Natural Sciences at Aarhus University.