Seediscussions,stats,andauthorprofilesforthispublicationat:https://www.researchgate.net/publication/293488522

MontanebiasinlowlandAmazonianpeatlands

ARTICLEinPALAEOGEOGRAPHYPALAEOCLIMATOLOGYPALAEOECOLOGY·JANUARY2015

ImpactFactor:2.34

4AUTHORS,INCLUDING:

EthanHouseholder

BotanicalResearchInstituteofTexas

9PUBLICATIONS57CITATIONS

SEEPROFILE

F.Wittmann

MaxPlanckInstituteforChemistry

90PUBLICATIONS1,150CITATIONS

SEEPROFILE

MathiasTobler

SanDiegoZoo

29PUBLICATIONS875CITATIONS

SEEPROFILE

Allin-textreferencesunderlinedinbluearelinkedtopublicationsonResearchGate,

lettingyouaccessandreadthemimmediately.

Availablefrom:EthanHouseholder

Retrievedon:25February2016

Palaeogeography, Palaeoclimatology, Palaeoecology 423 (2015) 138–148

Contents lists available at ScienceDirect

Palaeogeography, Palaeoclimatology, Palaeoecology

j ourna l homepage: www.e lsev ie r .com/ locate /pa laeo

Montane bias in lowland Amazonian peatlands: Plant assembly onheterogeneous landscapes and potential significance topalynological inference

J.E. Householder a,b,⁎, F. Wittmann c, M.W. Tobler d, J.P. Janovec d,e

a Instituto Nacional de Pesquisas da Amazônia, Av. André Araújo 2936, Manaus, AM, Brazilb Botanical Research Institute of Texas, 1700 University Drive, Fort Worth, TX, USAc Max Plank Institute for Chemistry, Otto Hahn Weg 1, 55128 Mainz, Germanyd San Diego Zoo Global Institute for Conservation Research, 15600 Pan Pasqual Valley Road, Escondido, CA, USAe Facultad de Ciencias Ambientales, Universidad Científica del Sur, Cantuarias 385 and Antigua Panamericana Sur km 19, Lima, Peru

⁎ Corresponding author at: Instituto Nacional de PesqAraújo 2936, Manaus, AM, Brazil. Tel.: +55 92 3643 3377

E-mail address: jehouseholder@gmail.com (J.E. House

http://dx.doi.org/10.1016/j.palaeo.2015.01.0290031-0182/© 2015 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 31 March 2014Received in revised form 9 January 2015Accepted 26 January 2015Available online 4 February 2015

Keywords:PeatlandWetlandClimate historyAndesAmazonGentryMontane

Past temperature changes in tropical mountain regions are commonly inferred from vertical elevational shifts ofmontane indicator taxa in the palynological record. However temperature is one of several abiotic factors drivingthe low-elevational limits of species and many montane taxa can occur in warmer lowlands by tracking appro-priate habitat types, especially highly flooded wetlands. In this paper we explore ways in which lowland habitatheterogeneity might introduce error into paleo-temperature reconstructions, based on field data of seven mod-ern peatland vegetation communities in the southern Peruvian Amazon (~200 masl). Peat-rich substrates arecommon edaphic transitions in pollen cores and provide detailed records of past vegetation change.The data show that indicators ofmodernpeatlands include generawithmontane aswell as lowlanddistributions,while indicators of surrounding forests on mineral substrates have predominantly lowland distributions. Basedon family-level analyses we find that modern peatland vegetation communities have taxonomic compositionsappearing to be 389m to 1557m (mean=1050±391m) above their actual elevations due to a high abundanceand number of families with high elevation optima.We interpret the relatively higher prevalence of montane elements in modern peatlands as habitat tracking of aconservedmontane niche on heterogeneous lowland landscapes.We suggest that both highmoisture availabilityand stressful edaphic conditions of peatland habitat may explain the montane bias observed. To the extent thatfossilization provides a better record of past vegetation that occurred proximate to the site of deposition, we sug-gest that habitat tracking of montane elements may introduce a cool bias in lowland paleo-temperature recon-structions based on pollen proxies.

© 2015 Elsevier B.V. All rights reserved.

1. Introduction

Climatic interpretation of fossil pollen assemblages along elevationalgradients plays a prominent role in our understanding of neotropical cli-mate change, especially in paleo-temperature reconstruction (Farreraet al., 1999;Weng et al., 2004).Many taxa tend to reach their highest di-versity and abundance along a limited elevational range as they trackappropriate conditions along mountain slopes. Terrestrial pollen re-cords document the vertical migration of stenothermic taxa in responseto past climate change, so knowledge of adiabatic lapse rates can beused to convert elevational shifts to temperature change. Thismethod is standard practice and has been especially important in

uisas da Amazônia, Av. André.holder).

quantifying past temperature depression from the down slope migra-tion of neotropical montane elements during Pleistocene glacial periods(Bush et al., 1990; Colinvaux et al., 1996; Groot et al, 2011).

The contemporary distribution patterns of higher-level taxa(e.g., families and genera) along mountain slopes are expected to bethe result of niche differences across space, however we know littleabout specific ecological factors underlying these distribution patterns(Körner, 2007; Wiens, 2011). While temperature is often assumed toplay a prominent role, this single variable is likely only one of a varietyof abiotic and biotic factors expected to drive the lower elevationalrange limits of montane taxa (Bush et al., 2004; Grubb, 1971; Körner,2007; Wiens, 2011). For example, on modern landscapes a number ofmontane indicator taxa with characteristic peaks in diversity andabundance in high elevation regions are also represented in nearbyAmazonian lowlands by one to few species, including Hedyosmum(Chloranthaceae), Panopsis (Proteaceae), Podocarpus (Podocarpaceae),

139J.E. Householder et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 423 (2015) 138–148

Salix (Salicaceae), Ternstroemia (Pentaphylaceae), Ilex (Aquifoliaceae),and others (Colinvaux et al., 2000; Marchant et al., 2002; Van derHammen and Hooghiemstra, 2000). While these montane taxa are ex-ceedingly rare in most lowland forest types, they can become abundantin stressful edaphic conditions, and especially in highly floodedwetlandhabitats (Gentry, 1982; Marchant et al., 2002; Oliveira-Filho and Ratter,1995; Steyermark, 1979). Apparent trackingofmanymontane elementsto particular lowland habitats suggests that the abiotic and biotic condi-tions inwhich they can successfully disperse and regenerate on lowlandlandscapes may be constrained by a highly conserved niche (Ackerly,2004; Harrison and Grace, 2007; Pillon et al., 2010; Wiens, 2011;Wiens and Donoghue, 2004; Wiens et. al., 2010).

The ecological and evolutionary processes that determine howmon-tane elements are assembled on modern heterogeneous lowland land-scapes may be relevant to palynological inference, and especiallypaleo-temperature inference, in at least two ways. First, the quality ofpaleo-temperature interpretation is dependent upon an adequate un-derstanding of the drivers of species low-elevational range limits,which are often poorly known (Körner, 2007). To the extent that mon-tane elements tend to occupy lowland sites that are most suitable totheir conserved phenotypes (Ackerly, 2004; Wiens, 2011), increasedunderstanding of the ecological conditions in which they occur in thelowlands can reveal insight into factors — other than temperature —

thatmight influence distribution patterns ofmontane elements throughtime. Second, habitat tracking of an ancestral niche on heterogeneouslowland landscapes can strongly alter the taxonomic composition andbiogeographic signatures of local vegetation communities during com-munity assembly, independent of regional climate change (Kubitzki,1989; Prance, 1979; Pyke et al., 2001; Wittmann et al., 2013). Because

Fig. 1. The 250-km stretch of themain drainage tributary of the southern Peruvian Amazonwh(fromwest to east; Table 3) are COLO, CICRA, HUIT 1, HUIT2, LAGA, MERC, and BOLI. The inset mthe peatland study area. Black and gray triangles represent Gentry sites at elevations greater thincluded in the supporting analysis (see Section 4.1.1).

vegetation directly occupying fluvial or depositional sites (i.e., wetlandvegetation) is known to contribute disproportionately to pollen sums,the ecological sorting processes operating at local scales have the poten-tial to influence the biogeographic signature of fossil assemblages and,in turn, the climate interpretations made from them (Berrio et al.,2002; Burnham et al. 2001; Marchant et al, 2009). Indeed, in somecases, alternative interpretations of past climate change seem to havebeen fueled by disagreement in the importance of climatic versus localecological processes in explaining compositional shifts in fossil pollenassemblages through time (Colinvaux et al., 2000; Punyasena et al,2011; Van der Hammen and Hooghiemstra, 2000).

In this study we develop a biogeographic framework to examine themodern taxonomic composition of peatland habitat in the southernPeruvian Amazon. Peat-rich substrates are commonplace in pollen recordsand have provided detailed accounts of past vegetation (e.g. Colinvauxet al., 1996; Liu and Colinvaux, 1985; Roucoux et. al. 2013). Makinguse of a publically available network of woody vegetation inventoriesestablished across a wide elevational gradient we examine the degreeto which the taxonomic compositions of woody peatland communitiesreflect that which would be expected given their actual elevation.

2. Methods

2.1. Study region

The peatland system considered in this paper is located within thesubsiding Beni-Mamore foreland basin below 250 masl in the Depart-ment of Madre de Dios in the Amazon region of southern Peru (Fig. 1).The main drainage tributary is the Madre de Dios River, an actively

ere focal peatlands (black) are located. Inventoried peatlands are outlined. Peatland namesap shows the location of the 76 plots obtained from the Gentry transect data in relation toan and less than 1000masl, respectively. The shaded area corresponds to the Chocó region

140 J.E. Householder et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 423 (2015) 138–148

meandering white-water river entrenched within alluvial deposits cor-responding to the late Neogene (Campbell et al., 2006). Non-floodedterra firme as well as seasonally flooded forests dominate the landscape.Satellite imagery combined with ground truthing revealed more than250 peatlands occupying a combined area of N290 km2 (Householderet al., 2012; Janovec et al., 2013) and distributed along a 250-km stretchof themodern meander belt of the Madre de Dios River from the Ande-an foothills eastward to the Bolivian border. Individual peatlands rangefrom b10 to 3500 ha. Mean peat thickness of individual peatlands rangefrom 0.85 to 3.04 m, however maximum thickness of up to 9 m wereregistered (Householder et al., 2012). The peatlands are replenishedcontinuously by ground water, with water level fluctuations typicallyless than 1 m in height (Goulding and Smith, 2007; Householder et al.,2012). The climate is humid tropical with annual rainfall varying be-tween approximately 1200 and 3300 mm and a strong dry seasonbetween August and October (Foster et al., 1994).

2.2. Vegetation sampling

Peatland vegetationwas sampled in seven of the largest permanent-ly saturated peatlands distributed along a 250-km east–west gradient,each accessed through 1–4 km transects that in most cases traversedthe entire peatland. Each transect was oriented to cross individualpeatlands in the direction parallelwith themain axis of habitat variationas judged by changes in pixel values exhibited by multispectral LandsatTM imagery configured to bands 3, 4, and 5. Spectral differences conve-niently reflect changes and transitions in vegetation structure rangingfrom mixed palm swamp forest to open boggy grasslands with stuntedwoody plants (Householder et al., 2012; Janovec et al., 2013).

Quantitative vegetation inventory was carried out in nested plotsdistributed at 100-m intervals along each transect. A total of 148 plotswere installed in the seven peatlands studied, with 18–30 plotssampled in each. Trees and shrubs were sampled in nested 10 × 10-mand 5 × 5-m square plots respectively, where trees are defined aswoody plants with DBH ≥ 10 cm and shrubs broadly defined aswoody plants with DBH b 10 cm and N1 m in height. To calculate abun-dance we counted the number of stems of each species in each nestedplot. Voucher specimens were collected for all species and depositedin the herbaria of the Botanical Research Institute of Texas (BRIT), theLa Molina University Forestry Herbarium (MOL), and the San MarcosMuseumof Natural History (USM). Allmetadata and annotation history,as well as associated images from the field and herbarium are openlyavailable through the Digital Herbarium of the Andes–Amazon AtriumBiodiversity Information System (http://atrium.andesamazon.org/index.php) under the project entitled “Aguajal Project”.

Data of vegetation on non-peat substrates was obtained from an ex-tensive vegetation plot network compiled by Alwyn Gentry and main-tained by the Missouri Botanical Garden (Phillips and Miller, 2002).The Gentry transect data are unique because they are based on a stan-dard sampling method used across an extensive geographic area and adiversity of habitats. Over the last two decades, the data have beenused in a number of plant macroecological studies (e.g., Enquistand Niklas, 2001; Enquist et al., 2002; Punyasena, 2008). Gentry in-ventory data were collected in 0.1-ha plots consisting of 10 contigu-ous 2 × 50-m linear subplots where all woody individuals ≥2.5 cmDBH were recorded, along with the occasional individuals with

Table 1Summary of the primary methodological differences between the Gentry plot data used to mo

Methodology Gentry

Minimum inclusion criteria DBH b 2.5 cmPlot size 100 m2

Plot arrangement Continuous lines oriented in random dirPlot shape LinearPlots per site 10Total sites 76

DBH down to 1 cm. Sequential subplots were placed end to end andoriented in random directions. We extracted data only from samplingsites located in the Andes–Amazon region (Bolivia, Colombia, Ecuador,Peru, and Venezuela) with elevations ranging from 10 to 3200 masl. Atotal of 76 sites suited our criteria, with 31 located above 1000 m, and38 lowland sites below 500 m (Appendix 1). Lowland inventoriesincluded wetland (n = 8) and upland (n = 30) habitats. Sites on themargins of our region of interest corresponding to Bolivian savannaand dry forest sites were excluded as seasonal drought is expected todrive floristic differences there. Sites on the Pacific slope were also ex-cluded. The accuracy of species names provided in the dataset wasassessed by cross-referencing voucher specimen numbers provided inthe original dataset with current determinations of collections storedat the Missouri Botanical Garden and accessible through the Tropicosdatabase (http://www.tropicos.org).

Plant nomenclature of both Gentry and peatland plot data was stan-dardized to Tropicos and analyzed at the taxonomic rank of family. Forthe peatland data, tree and shrub subplots were merged by calculatingthe sum of family abundances in each peatland. A matrix of relativeabundances was calculated initially, but we were concerned with po-tential biases in abundance estimates caused by a nested plot designand differences in minimum size criteria in the peatland and Gentryinventories (see Table 1 for a summary of important methodologicaldifferences). To account for this, presence/absence matrices wereconstructed for comparison.

2.3. Development of the biogeographic framework

Using the Gentry transect data, we modeled the biogeographicpattern of individual families along the elevation gradient using two pa-rameters, their elevation optima (abundance weighted mean eleva-tions), and their weighted standard deviation of elevation (a measureof elevational range). Previous assessments of the Gentry transect datahave demonstrated strong taxonomic structure along the elevation gra-dient, presumably driven by strong abiotic gradients that sort speciesaccording to highly conserved ecological niches (Gentry, 1982;Enquist et al, 2002; Punyasena et al., 2008). We thus assume that sucha straightforward assessment of family-level elevational distributionprovides a coarse metric by which to compare biogeographic patternsof taxonomic groups, as well as provide a useful predictor of differencesin the ecological response of species (Ackerly, 2003;Harrison andGrace,2007; Punyasena et al., 2008).

2.4. Community analysis

We used community-weighted averages of family elevation optimato quantify biogeographic tendencies of peatland vegetation assemblies.Thus, peatlands that possess an abundance of taxa with high elevationoptima will have higher community weighted averages. This approachprovides a practical and intuitive metric, expressed as an expectedelevation, by which to examine community structure within a biogeo-graphic framework. We used the difference between expected eleva-tions based on community weighted means and observed siteelevations to evaluate the degree to which the taxonomic compositionsof peatlands differed fromwhatwould be expected given their observedelevations.

del distribution patterns of families along the elevation gradient and the peatland plots.

Peatland

Height b 1 m100 m2 (trees) + 25 m2 (shrubs)

ections Discontinuous plots along main environmental gradientSquare18–307

Table 2Comparison of community-weighted average regression models of expected and actualelevations of the Gentry transect data. Rootmean square error (RMSE) and explained var-iance (r2) are presented for relative abundance or presence/absence matrices using in-verse (I) or classical (C) deshrinking (D/S) andwith orwithout tolerance downweighting.

Matrix Down weight D/S RMSE r2

Relative abundance No I 228.1 0.95C 234.5 0.95

Yes I 244.6 0.94C 252.5 0.94

Presence/absence No I 254.3 0.93C 263.2 0.93

Yes I 277.8 0.92C 289.5 0.92

Fig. 2. Scatter plot of actual and expected elevations for 76 training sites (open circles) ob-tained from theGentry transect data and seven peatland test sites (solid circles). Expectedelevations are based on abundance-weighted averages of family-level elevational optimaafter deshrinking (see Section 2.4). Scatter plots based on family presence/absence datashow qualitatively similar patterns. Based on a cross-validation procedure used to obtainp values, peatlands appear to occur at significantly higher elevationsdue to a greater abun-dance and number of families with higher elevation optima (Table 3).

141J.E. Householder et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 423 (2015) 138–148

A specially designed cross-validation procedure was developed totest the specific hypothesis that individual peatlands possess a taxo-nomic composition statistically different from what would be expectedgiven their actual elevation. Specifically, theGentry datawere repeated-ly partitioned into validation and training sets, where individual valida-tion sites were sub-sampled with probability of 0.1 drawn from arandom binomial distribution. Newly derived family elevation optimadeveloped with the remaining training sites were used to calculatecommunity-weighted averages of validation sites. A probability distri-bution of calculated differences between observed and expected eleva-tions was generated through 1000 iterations and used to calculate pvalues.

One inherent issue of theweighted averaging approach is a potentialcontraction in the range of calculated expected elevations as a result ofdouble averaging (of both families and communities). To account this,we examined two (inverse and classical) deshrinking correction factors,as suggested by ter Braak and van Dam (1989). We also examined thepotential undue influence of familieswith patchy or verywide elevationranges by alternatively down-weighting families by their weightedstandard deviation. These variants of community weighted elevationoptima were modeled as a function of observed site elevations usingstandard linear regression. A cross-validation procedurewas used to es-timate regression model errors of both simple and down-weightedcommunity means by iteratively removing 10 random sites from thetraining set. A randomization procedure (van der Voet, 1994) wasused to assess the difference in root mean square errors and chose thecorrection factor that best reproduced observed site elevations. Modelconstruction and evaluation was performed in R version 3.0.2 (R CoreTeam, 2013) using the package rioja (Juggins, 2012).

2.5. Analysis of indicator taxa

An additional analysis examined the degree of habitat specificity ofgenera in peat vs. mineral substrates in the lowlands. Informationabout forest types and substrates of Gentry transects were extractedfrom voucher specimens. All Gentry transects within a 400-km radiusof the peatlands and below 700 m elevation were included. The final,merged data included 14 sites, seven from each of the Gentry andpeatland datasets. Indicator values, measured as the product of the rel-ative frequency and relative average abundance in peat vs. non-peatsubstrate groups, were calculated for each genus using the R packagelabdsv (Roberts, 2010) and following the methods of Dufrene andLegendre (1997). Elevational distribution ranges of 20 genera with thehighest significant indicator values for each substrate class wereassessed by comparing herbarium collection records in the online data-base of the Global Biodiversity Information Facility (GBIF; http://www.gbif.org). Records were filtered to include only South American speci-mens deposited in the Missouri Botanical Garden, and those with obvi-ous georeference errors were excluded.

3. Results

We found that after applying a deshrinking correction factor,community-weighted averages of family elevation optima, hereafter re-ferred to as expected elevations, did well to reproduce observed site el-evations of the Gentry transect (Table 2). Down-weighting by familyelevational ranges tended to decrease explained variance and increasethe root mean square error, suggesting that even widely distributedtaxa are likely to have clear abundance patterns along the elevationalgradient. Models based on family abundances tended to outperformthose based on presence–absence data. Inverse deshrinking resultedin slightly lower root mean square errors and a slightly higher propor-tion of explained variance, independent of whether the abundance orpresence/absence matrix was used, so this deshrinking correctionmethod was used throughout (Table 2).

Differences between expected and observed elevations ranged from389 to 1557m (mean± sd=1050m±391) for the abundancematrixand 313 to 906 m (mean ± sd = 693 m ± 240) for the presence/absence matrix (Fig. 2 and Table 3). The cross-validation procedureshowed that in most cases these differences were significantly higher(Table 3). Details about all families, including their elevational optimaand number of stems in peatlands, are detailed in Appendix 2.

In the 14 lowland sites a total of 371woody genera were found oneither or both organic peat and mineral substrate classes. A total of272 (73%) genera were restricted to a single substrate type. A totalof 80 genera occurred in peatlands, 61% of which did not occur onmineral substrates. Likewise, 291 genera occurred on mineral sub-strates, 77% of which did not occur on peat substrates. According tothe indicator analysis, 20% of the 371 genera were significant indica-tors of one of the two substrate types. A total of 30% (24) of the gen-era present on peat were significant indicators of peat (p b 0.01)while this proportion dropped to 18% (52) for genera occurring onmin-eral substrates. Over 118,000 unique herbarium collection recordswereaccessed through the GBIF website to explore the elevational distribu-tion of all indicator genera. Indicator genera of forests on mineral sub-strates are mostly restricted to low elevations. In contrast, indicatorgenera of peat substrates exhibit montane, lowland, and widespreadelevational distributions (Fig. 3).

Table 3Differences between expected and actual elevations of each peatland where quantitativeinventories were performed. Expected elevations are based on abundance-weighted andpresence–absence averages of family-level elevation optima after deshrinking (seeSection 2.4). Expected elevations exceed actual elevation in all cases. Significance is basedon an iterative cross-validation procedure (see Section 2.4).

Peatland Matrix

Abundance Presence/absence

Prediction error (m) Significance Prediction error (m) Significance

BOLI 389 0.06 313 0.09CICRA 1557 b0.001 868 b0.009COLO 761 b0.001 517 0.031HUIT1 930 b0.001 512 0.053HUIT2 1301 b0.001 831 0.012LAGA 1290 b0.001 903 b0.001MERC 1124 b0.001 906 0.003

142 J.E. Householder et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 423 (2015) 138–148

4. Discussion

The biogeographic distributions of plants are closely linked to physio-logical responses to climate, which are often highly conserved within lin-eages. The high explanatory power of the regression model tested in thisstudy confirms this to the extent that family-level composition ofmodernvegetation varied predictably along the elevational gradient. However,due to increased frequency and abundance of families with preferencefor high elevation habitats modern lowland Amazonian peatland assem-blies demonstrated expected elevations of 313 to 1557mabove actual el-evations, significantly higher relative to non-peat habitat types. Theconsistently high elevations predicted for peatlands based on both abun-dance and presence/absence data demonstrate that rather than rare ex-ceptions, montane elements are abundant and taxonomically diverse inlowland peats, relative to surrounding forests.

The results are further supported by a local comparative analysis ofpeatland composition and the surrounding forests,whichdemonstratedthat floras on peat substrates are a mixture of both montane and low-land indicator taxa. In addition to the mid-high elevation forest generaidentified by the indicator analysis (Cestrum, Cybianthus, Graffenriedia,Hedyosmum, Ilex, Myrcia and Nectandra), other montane elements arealso common in peatlands, such as Begonia (Begoniaceae), Centropogon(Campanulaceae), Coccosypselum (Rubiaceae), Cyathea (Cyatheaceae),Isoetes (Isoetaceae), Myrsine (Primulaceae), Polypodium(Polypodiaceae) and Talauma (Magnoliaceae). However they wereeither not within the top 20 indicators, occurred at reduced frequency,or did not meet minimum size criteria. Nonetheless, the data showthat montane taxa represent an important component of peatlandvegetation, a pattern we refer to as “montane bias”.

4.1. Palynological relevance

Interpretation of down slope migration apparent from the palyno-logical record must ultimately be framed within our understanding ofthe drivers of species low-elevational range limits. While these remainpoorly understood for most species they are widely expected to includea combination of factors that are abiotic (e.g., temperature and mois-ture) and biotic (e.g., competition, predation, parasitism, dispersal)(Körner, 2007;Wiens, 2011). During paleo-temperature reconstructionit is often assumed that temperature change outweighs other potential-ly important drivers, even though alternative hypotheses can often notbe fully rejected (Bush et al., 2004; Punyasena et al., 2011; Van derHammen and Hooghiemstra, 2000). We propose that examination ofthe conditions in which montane taxa occupy modern lowland land-scapes can reveal insight into the factors — other than temperature —

that might influence their low-elevation limits. Our arguments arebased on the premise that biogeographic patterns result from the ten-dency for species to retain their ancestral niche that, in turn will deter-mine the ability of species to expand and colonize in changing

environments (Ackerly, 2004; Harrison and Grace, 2007; Wiens, 2011;Wiens and Graham, 2005). While it is plausible that individual specieswithin higher taxa can behave idiosyncratically, rendering familymem-bership a poor predictor of ecological behavior (Slivertown et al., 2001)our results suggest otherwise. The tendency for montane taxa to preferparticular lowland habitat types seems best interpreted as habitat track-ing of an ancestral, montane niche on a heterogeneous lowland land-scape. Below we suggest that high moisture availability and stressfuledaphic conditions characterizing peatlands may be important factorscontributing to the observed montane bias.

4.1.1. Moisture availabilityThe montane bias in modern lowland peatlands, characterized by

year-round soil moisture, suggests moisture availability to be a criticalfactor influencing the current lowland distributions of montane taxa.For example, the distribution of the majority of extant species ofHedyosmum in cool, moist habitat of mid-high elevation Andean forestsis expected to be a consequence of constraints induced by poor stemwood hydraulic capacity and susceptibility to drought cavitation(Antonelli and Sanmartín, 2011).Where it occurs in hot and highly sea-sonal lowland regions it is invariably limited tomoistmicrosites, such aspermanently saturated peats (this study),waterlogged sands, and ripar-ian habitats (Todzia, 1988). Thus, our findings are consistent with theidea that the distribution of montane forests and their taxa are highlysensitive to changes in plant water-energy balance (Bruijnzeel andVeneklas, 1998).

Because both ambient temperature and available moisture interactto define plant water-energy balance, downhill distributional shifts ofplant species apparent from the pollen record may not be adequatelyexplained by considering reductions in past temperature in isolation(Bush et al, 2004). Increases in availablemoisture that limit the durationand severity of drought conditions during key seasonal and plant devel-opmental phases may contribute to lower elevation limits of montanetaxa. Downhill niche tracking of a suitable water-energy balance eitheras a result of regional increases in precipitation — as suggested forglacial western Amazonia (Cheng et al., 2013) — or moist micrositeconditionsmight be challenging to distinguish from a pure temperatureresponse (Bush et al., 2004).

We used the Gentry transect data to further test the idea that mois-ture availability can significantly influence the lower-elevation limits ofmontane taxa. Using the same cross-validation procedure described inSection 2.4, we examined model prediction errors of data from 12 Gen-try transect sites located in the Chocó, a lowland region distributed fromsouthern Panama to northern Ecuador (Fig. 1). The region was separat-ed from the Amazon basin due to the uplift of the Eastern Cordillera ofthe Andes Mountains (Behling et al., 1998; Hoorn et al., 2010). Annualprecipitation exceeds 4000 mm/yr and typically reaches N7500 mmnear the foothills of the Pacific slopes of the Andes. Even during the dri-est months droughts persisting longer than one week are extremelyrare (West, 1957). The mean annual temperature is about 26–27 °C,with a minimal 1 °C difference between the coldest and the warmestmonths (Snow, 1976). If increased moisture availability influences theability of montane plant taxa to persist at lower elevational limits,then we might expect these to have lower distributions on rainy west-ern Pacific slopes relative to Amazonian slopes. Indeed, expected eleva-tions based on abundance-weighted community averages of familyelevation optima were significantly higher than expected for 10 out of12 Chocó sites (p b 0.05 as the critical value). Differences between actualand expected elevations ranged from−217m to+695mwith a mean(±sd) of 445 m (±234 sd). The higher expected elevations for thePacific slope lowland sites are the result of the presence of severalplant taxa that are mostly restricted to higher montane forests on theAmazonian side. Using presence/absence data, only four of 12 plotshad significantly higher expected elevations. Differences between ex-pected and actual elevations ranged from −384 m to +525 m with amean (±sd) of 325 m (±268 sd). While we cannot eliminate the

Fig. 3. Elevational distributions of 40 indicator genera on peat andmineral substrates. Box plot hinges extend only to the first and third quartiles and themedian is indicated by a darkenedband. Within respective substrate types taxa are ordered by their mean elevation. The dashed vertical line at 658 m marks the mean of average elevations of all genera.

143J.E. Householder et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 423 (2015) 138–148

possibility that other ecological and historical factors are driving thepatterns observed for the Chocó, this preliminary analysis corroboratesprevious botanical observation of the tendency formontane elements toexhibit lower elevational limits in this pluvial region and for moistureavailability to influence the lower-elevation limits of plant taxa(Gentry, 1982; Hooghiemstra et al., 2012; Van der Hammen andHooghiemstra, 2000). Comparing results of pluvial Chocó sites topeatland sites, we see that while both tend to demonstrate higher ex-pected elevations using family abundance data, proportionally fewerChocó sites show significantly higher (p b 0.05) elevations using pres-ence/absence data. Thus, peatlands are distinguished by their high di-versity of montane taxa that are distributed among a greater numberof families, at least on the spatial scale of small plots considered here.

These results further highlight the important contribution of montaneelements to this lowland habitat type.

4.1.2. Edaphic associationsWhile temperature is largely expected to determine the physiologi-

cal limits of the upper elevational ranges of plant species, their lowerdistributional limits are expected to be determined primarily by compe-tition rather than by excessively high temperature (Coomes and Allen,2007; Grubb, 1971, 1977; Prior and Bowman, 2014). Peatlands in theAmazonian lowlands are structurally and edaphically characterized byshort-statured canopies, decreased productivity, extremely acidic soils,saturated substrates, limited fertility, phenolic soil toxicity, and reduceddecomposition and nitrogen mineralization (Householder et al., 2012;

144 J.E. Householder et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 423 (2015) 138–148

Lähteenoja et al., 2009; Page et al., 1999; Runge, 1983; Waughman,1980). Interestingly, these hydrologically induced edaphic conditionsare quite similar to the climate-driven edaphic conditions characteristicofmontane sites (Bruijnzeel and Veneklas, 1998; Edwards, 1982; Grubb,1977; Tanner et al., 1998; Unger et al, 2010). Confronting potentiallysimilar limitations to growth and survival along increasingly stressfulranges of elevation and hydrological gradients, it is not unreasonableto expect convergent effects on species traits (Chapin et al., 1993;Reich et al., 1997). Montane elements in modern lowlands are likely tobe better competitors in edaphically extreme lowland habitats than ina tall, productive, and diverse terra firme forest matrix. By altering re-source bioavailability, predation and parasitism in such ways thatfavor montane species, edaphic transitions on heterogeneous lowlandlandscapes can shift competitive dynamics at range boundaries and po-tentially significantly decrease the lower limits of montane distributions(Briers, 2003; Gaston, 2003; Holt and Barfield, 2009; Jaeger, 1971;Parmesan, 2006).

With this inmind, to the extent that montane elements invaded pastlowland environments, their migrations may have been largely accom-modated by marginal, highly flooded, stressful, and patchily distributedhabitat types. Indeed, by expanding the geographic window it is clearthat many important montane indicator taxa (e.g., Alnus, Betula,Ericaceae andWeinmannia) are better known as bog and fenwetland in-dicators in their native temperate zones (USDANRCS., 2013). As the eco-logical distributions of montane taxa become increasingly constrainedto specialized edaphic conditions towards the limits of their lower eleva-tion ranges, immigration and extinction processes are likely to be morevariable. Variation in species propagule dispersal, metapopulation dy-namics, and the spatial configuration of appropriate habitats in pastlandscapesmay be important in explaining the extent of past downslopeplant migration (Leibold et al., 2004).

4.1.3. Pollen proxies on heterogeneous Amazonian landscapesEstimates of glacial phase neotropical temperature depression using

fossil pollen proxies vary greatly, ranging from ~5 °C to 9 °C relative tomodern (Ballantyne et al., 2005; Bush et al., 2004; Farrera et al., 1999).While suchwide variability poses problems for elucidating amore accu-rate vision of past climate, we suspect that it is consistentwith high neo-tropical habitat heterogeneity. For example, we have argued that inmodern lowland landscapes habitat tracking of a montane niche canlead to local vegetation communities on particular hydro-edaphicconditions to possess a higher abundance and diversity of montane ele-ments relative to surrounding forest types. Using an adiabatic lapse rateof 6.5 °C/km, the modern flora occupying peat substrates possesses abiogeographic signature resembling forests that are, on average, 2.1–2.3 °C cooler based on the mean differences between observed and ex-pected elevations of presence–absence and abundance models respec-tively. While clearly not reflecting modern paleoecological methods —pollen sums reflect both local vegetation as well as that derived fromsurrounding uplands— it does show a strong effect of habitat heteroge-neity on the biogeographic composition of local plant communities.However, because fossilization can leave a better record of certain

Appendix 1

Alphabetical list of Gentry transect data used to asses family-level biogwere obtained from botanical voucher specimens accessed through Tropicostory sites located within the Chocó biogeographic province (italicized) are

Site Latitude Longitude Country Elevation (m)

Achupall 3°27′S 78°22′W Ecuador 2090Allpahua 3°50′S 73°25′W Peru 130Altodemi 10°55′N 73°50′W Colombia 1180Altosapa 7°10′N 75°54′W Colombia 2660Antado 7°15′N 75°55′W Colombia 1560

habitat types than others, a cool bias in lowland paleotemperature esti-mates may be more prevalent than appreciated to date. Assuming amontane bias in similar neotropical environments in the past, our dataestablish the possibility— at least under depositional conditions charac-terized by the accumulation of peat — that pollen assemblages inAmazonian lowlands can potentially appear to be more “montane”,and thus cooler, independent of plant response to temperature. In thissense our interpretations are consistent with other findings that havesuggested habitat-related errors in estimating past temperatures fromplant macrofossils. For example, in lowland Ecuadorean Amazon,Burnham et al. (2001) demonstrated that trait analysis of woody plantsgrowing around lakes and rivers leads to an underestimation of meanannual temperatures by 2.5 °C–5 °C due to the high proportion oftoothed leaves in these flooded habitats.

The ecological processes that sort montane elements on heteroge-neous lowland landscapes may point towards a potential cool bias ofpollen proxies in reconstructing past temperature in the Neotropics. In-deed, reviews of pollen-based proxies of glacial tropical South Americatend to show greater minimum (cooler) temperature depressions rela-tive to both non-pollen proxies (e.g., Stute et al, 1995) and other tropicalregions (Ballantyne et al., 2005; Farrera et al., 1999).While explanationsof these patternsmay be revealed by a refined understanding of how cli-mate works, efforts to improve our understanding of how tropical eco-systems and their biota uniquely evolve and interact along localgradients may be informative as well.

Acknowledgments

The manuscript was developed and written with funding provided tothe corresponding author by the Conselho Nacional de DesenvolvimentoCientífico e Tecnológico (CNPq—process number 141727/2011-0). Wethank the Max Planck Institute for Chemistry and Maria TeresaFernandez Piedade for providing a productive, hospitable work environ-ment. Field research was made possible by funding from the DiscoveryFund of Fort Worth, Texas, the Gordon and Betty Moore Foundation(Grant no. 484), the U.S. National Science Foundation (grant no.0717453), and the Programa de Ciencia y Tecnologia—FINCYT (co-fi-nanced by the Banco Internacional de Desarollo, BID) grant numberPIBAP-2007-005. We thank Sy Sohmer, Cleve Lancaster, Pat Harrison,Will McClatchey, the board of directors, administration, development,and general staff of the Botanical Research Institute of Texas (BRIT) forproviding important institutional support and infrastructure over theyears. We are grateful to Jason Wells, Amanda Neill, Keri Barfield, andRenan Valega for logistical support, assistance, and discussion duringvarious phases of this research. We also thank Fernando Cornejo, PiherMaceda, Javier Huinga, Angel Balarezo, and Fausto Espinoza for assistingin different phases of field research. Finally, we appreciate various insti-tutions for ongoing research, collection, and export permits during thecourse of this project, including the former Instituto Nacional deRecursos Naturales (INRENA), the Dirección General de Flora y FaunaSilvestre (DGFFS), the Ministerio de Agricultura (MINAG), and theMinisterio del Ambiente (MINAM).

eographic patterns along the elevational gradient. Habitat descriptions. Boldfaced sites were used in the indicator analysis (Section 2.5). Inven-appended at the end.

Habitat Description

Tepui-like bromeliad sward with scattered small treesTropical moist forest (Iquitos–Nauta road)Tropical premontane wet forest (Magdalena)Montane oak forestCloud forest

Appendix 1 (continued)

Site Latitude Longitude Country Elevation (m) Habitat Description

Araracua 0°25′S 72°20′W Colombia 200 Mature forest over yellow clayArcating 0°25′S 72°19′W Colombia 250 High campina forest on white sand soil near air field.Cabezade 10°20′S 75°18′W Peru 320 Sandy river terracesCampano 11°08′N 74°01′W Colombia 1685 Tropical lower montane wet forestCandamo 13°30′S 69°50′W Peru 800 Ridge top forest with cloud forest aspectCarpanta 4°35′N 73°40′W Colombia 2900 High Andean cloud forestCeroneb1 0°50′N 66°11′W Venezuela 140 Mature forest on sandy “ultisol”Ceroneb2 0°50′N 66°11′W Venezuela 140 Mature forest on sandy “ultisol”Cerroayp 4°35′S 79°32′W Peru 2750 Montane cloud forestCerroesp 10°28′N 72°50′W Colombia 2560 Tropical montane wet forestChirinos 5°25′S 78°53′W Peru 1780 Remnant patch of Podocarpus forestChorrobl 6°10′S 78°45′W Peru 2430 Cloud forestCochacas 11°52′S 71°22′W Peru 380 Mature forest on alluvial soilColorado 9°58′N 75°10′W Colombia 240 Disturbed moist forestColosoi 9°30′N 75°22′W Colombia 300 Forest on limestone.Constanc 4°15′S 72°45′W Peru 160 Primary forest on white clay soil with thick root mat. Not inundated.Cuangos 3°29′S 78°14′W Ecuador 1500 Cloud forestCueras 6°40′N 76°30′W Colombia 1670 Cloud forestCueva 6°40′N 76°30′W Colombia 350 Mature forest on rich soil on rocksCutervo 6°10′S 78°40′W Peru 2200 Montane cloud forestCuyas 4°32′S 79°44′W Peru 2410 Disturbed cloud forestCuzcoama 12°35′S 69°09′W Peru 200 Mature forest on alluvial soilDureno 0°05′N 75°45′W Ecuador 300 Tropical moist forest (Napo)Elcorazo 0°36′N 77°42′W Ecuador 3200 Disturbed montane cloud forest on volcanic soilElpargo 6°30′S 79°31′W Peru 3000 High Andean forestEncanto 14°38′S 60°42′W Bolivia 240 Subtropical moist forestFarall 3°30′N 76°35′W Colombia 1930 Tropical lower montane moist forestFincam 2°16′N 76°12′W Colombia 2280 Mature montane forestFincaz 3°32′N 76°35′W Colombia 1960 Tropical lower montane moist forestHachimal 3°38′N 76°33′W Colombia 1860 Tropical lower montane moist forestHuamani 0°40′S 77°40′W Ecuador 1150 Tropical premontane moist forestHumboldt 8°45′S 75°02′W Peru 270 Tropical moist forestIncahuar 15°55′S 67°35′W Bolivia 1560 Steep cloud forest with dry seasonIndiana 3°30′S 73°02′W Peru 130 Well-drained forest on good soil.Jatunsac 1°04′S 77°36′W Ecuador 450 Non-alluvial, red clay soilJenarohe 4°55′S 73°45′W Peru 115 Upland forest on well-drained soilKennedy 11°05′N 74°01′W Colombia 2600 Tropical montane wet forestLagenoa 11°05′S 75°25′W Peru 1140 Mature forest remnantLaplanad 1°08′N 77°58′W Colombia 1750 Premontane wet forestLaraya 8°20′N 74°55′W Colombia 20 Tropical moist forestMadidi 13°35′S 68°46′W Bolivia 280 Subtropical wet forestMadidiri 13°35′S 68°40′W Bolivia 370 Ridge topManaure 10°22′N 73°08′W Colombia 540 Streamside moist forest remnantMaquipuc 0°07′N 78°37′W Ecuador 1630 Mature cloud forestMariquit 5°15′N 74°50′W Colombia 560 Disturbed wet forest remnantMiazi 4°18′S 78°40′W Ecuador 850 Floodplain forest along Rio NangaritzaMishnfl 3°56′S 73°27′W Peru 130 Non-inundated lowland forestMishws 3°59′S 73°27′W Peru 150 White sand forestMurri 6°35′N 76°50′W Colombia 980 Mature cloud forestNangarit 4°18′S 78°40′W Ecuador 930 ForestPasochoa 0°28′S 78°25′W Ecuador 3020 Mature montane forestRioheath 12°50′S 68°50′W Peru 200 Forest near edge of savannahRiomanso 7°30′N 76°05′W Colombia 200 Tropical wet forestRionegro 9°50′S 65°40′W Bolivia 100 Mature forest on sandy clayRiotavar 13°21′S 69°40′W Peru 400 Moist lowland forest on ridgeSabanaru 10°30′N 72°55′W Colombia 2940 Forest just below paramoSacram 16°18′S 67°48′W Bolivia 2450 Dense ridge top cloud forestShiringa 10°20′S 75°10′W Peru 300 Wet lowland forest on claySietecue 4°35′N 73°40′W Colombia 2350 Andean cloud forestSucusari 3°15′S 72°55′W Peru 140 Mature terra firme forestTahuampa 3°57′S 73°31′W Peru 130 Seasonally inundated tahuampa forestTamblat2 12°50′S 69°17′W Peru 250 Tropical premontane wet forest (swamp trail)Tambo 12°50′S 69°17′W Peru 200 Tropical premontane wet forest (Rio Tambopata at mouth of Río D'Orbigny)Tamboall 12°50′S 69°17′W Peru 260 Alluvial forestTambupl 12°49′S 69°43′W Peru 280 Terra FirmeTarapoto 6°40′S 76°20′W Peru 500 Dry forested slopes overlooking Río HuallagaUchire 10°10′N 65°32′W Venezuela 150 Tropical lower montane wet forestVencer 5°45′S 77°40′W Peru 1850 Wet lower montane forestYanam1 3°39′S 72°57′W Peru 105 Mature non-inundated forest on lateriteYanam2 3°28′S 72°50′W Peru 130 Mature forest on yellowish lateritic soilYanamtah 3°28′S 72°50′W Peru 120 Seasonally inundated tahuampa forestAmotape 4°09′S 80°37′W Peru 700 ChocóAnchicay 3°45′N 76°50′W Colombia 300 ChocóBilsa 0°37′N 79°51′W Ecuador 280 ChocóCalima 3°56′N 77°08′W Colombia 50 Chocó

(continued on next page)

145J.E. Householder et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 423 (2015) 138–148

Appendix 1 (continued)

Site Latitude Longitude Country Elevation (m) Habitat Description

Centinel 0°37′S 79°18′W Ecuador 650 ChocóEsmerald 0°54′N 79°37′W Ecuador 150 ChocóJauneche 1°16′S 79°42′W Ecuador 70 ChocóPerromue 1°36′S 80°42′W Ecuador 480 ChocóRiopal1 0°36′S 79°22′W Ecuador 200 ChocóRiopal2 0°35′S 79°22′W Ecuador 200 ChocóSansebas 1°36′S 80°42′W Ecuador 550 ChocóTutunend 5°46′N 76°35′W Colombia 80 Chocó

Family Elevational optima (abundance weighted) Elevational optima (presence/absence) Total number of stems

Anacardiaceae 593.3854 652.6316 855Annonaceae 441.0999 647.193 333Apocynaceae 493.0688 642.8182 4Aquifoliaceae 2355.0417 1853.2609 181Araceae 1595.0441 1347.9268 2Araliaceae 2038.2402 1365.0962 79Arecaceae 789.9211 876.6406 763Asteraceae 2410.2626 1598.2222 1Begoniaceae 1952 2032.5 5Bignoniaceae 398.3834 646.6379 487Boraginaceae 695.7576 798.5106 3Burseraceae 291.1667 345.4545 34Calophyllaceae 359.7126 494.5652 40Campanulaceae 1610 1618.3333 2Celastraceae 634.698 760.6731 4Chloranthaceae 2491.7787 2170.4762 1087Chrysobalanaceae 295.0432 460 1Clusiaceae 1434.398 1158.1579 97Combretaceae 297.0323 348.5294 7Connaraceae 229.5041 255.9375 2Cyatheaceae 1710.7642 1280.8889 63Dichapetalaceae 727.5397 595.5769 1Dilleniaceae 214.1739 268.4848 8Ebenaceae 268.2 320.3333 3Elaeocarpaceae 890.0362 772.0833 50Erythroxylaceae 440.3409 382.9167 281Euphorbiaceae 874.8231 817.7344 64Fabaceae 379.2353 705.1587 116Hypericaceae 1483.1818 1201.25 4Lamiaceae 1195.2027 880.1852 2Lauraceae 1375.1708 1078.913 104Lecythidaceae 394.5533 576.8421 10Linaceae 468.4146 423.9286 1Loganiaceae 274.6711 248.5714 2Lythraceae 240 240 7Magnoliaceae 1632.5 1454 11Malpighiaceae 511.0177 483.2292 1Malvaceae 542.9225 621.3158 383Marantaceae 258.4 292 2Marcgraviaceae 1129.4505 1071.4516 14Melastomataceae 1875.4962 1148.1343 540Meliaceae 798.6979 876.0833 5Menispermaceae 342.2519 452.1429 2Monimiaceae 1080.4032 901.75 1Moraceae 611.4446 784.0476 417Myristicaceae 507.1563 592.1569 26Myrtaceae 1132.7434 911.3492 151Nyctaginaceae 560.4492 487.7778 9Ochnaceae 430.8333 386.6667 6Onagraceae 2628.4211 2606.6667 14Passifloraceae 1180 750 1Phyllanthaceae 1504.2733 1199.3243 5Piperaceae 1661.148 1186.0377 30Polygalaceae 1607.2727 2271.1111 4Polygonaceae 565.5191 1286.7308 64Primulaceae 2041.0519 717.3171 97Rosaceae 2322.2581 360 3Rubiaceae 1413.2514 1906.6667 231Rutaceae 555.1961 970.4795 1

Appendix 2

Alphabetical list of families present in peat communities, their calculated elevational optima, and their total number of stems.

146 J.E. Householder et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 423 (2015) 138–148

Appendix 1 (continued)

Family Elevational optima (abundance weighted) Elevational optima (presence/absence) Total number of stems

Salicaceae 637.0153 1326.1429 15Sapindaceae 669.0143 2200 48Sapotaceae 395.5526 757.9839 4Siparunaceae 713.794 326.1364 23Smilacaceae 829.2857 848.9189 8Solanaceae 2162.6165 1180 31Strelitziaceae 100 318.75 3Thymelaeaceae 1243.3333 2005.5556 1Trigoniaceae 248.5714 1227.1429 1Urticaceae 721.6878 271.9231 4Violaceae 358.3953 470 1Vochysiaceae 726.7901 1155 8

Appendix 2

147J.E. Householder et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 423 (2015) 138–148

References

Ackerly, D., 2003. Community assembly, niche conservatism, and adaptive evolution inchanging environments. Int. J. Plant Sci. 164, 164–184.

Ackerly, D., 2004. Adaptation, niche conservatism, and convergence: comparative studiesof leaf evolution in the California Chaparral. Am. Nat. 163, 654–671.

Antonelli, A., Sanmartín, I., 2011. Mass extinction, gradual cooling, or rapid radiation?Reconstructing the spatiotemporal evolution of the ancient angiosperm genusHedyosmum (Chloranthaceae) using empirical and simulated approaches. Syst.Biol. 60, 596–615.

Ballantyne, A., Lavine, M., Crowley, T., Liu, J., Baker, P., 2005. Meta-analysis of tropical sur-face temperatures during the Last Glacial Maximum. Geophys. Res. Lett. 32, L05712.

Behling, H., Hooghiemstra, H., Negret, A., 1998. Holocene history of the Chocó rain forestfrom Laguna Piusbi, Southern Pacific lowlands of Colombia. Quat. Res. 50, 300–308.

Berrio, J., Hooghiemstra, H., Behling, H., Botero, P., Van der Borg, K., 2002. Late-Quaternarysavanna history of the Colombian Llanos Orientales from Lagunas Chenevo andMozambique: a transect synthesis. The Holocene 12, 35–48.

Briers, R., 2003. Range limits and parasite prevalence in a freshwater snail. Proc. R. Soc.Lond. B 270, 178–180.

Bruijnzeel, L., Veneklas, J., 1998. Climatic conditions and tropical montane forest produc-tivity: the fog has not yet lifted. Ecology 79, 3–9.

Burnham, R.J., Pitman, N.C.A., Johnson, K.R.,Wilf, P., 2001. Habitat-related error in estimat-ing temperatures from leaf margins in a humid tropical forest. American Journal ofBotany 88, 1096–1102.

Bush, M., Colinvaux, P., Wiemann, M., Piperno, D., Liu, K., 1990. Late Pleistocene temper-ature depression and vegetation change in Ecuadorian Amazonia. Quat. Res. 345,330–345.

Bush, M., Silman, M., Urego, D., 2004. 48,000 years of climate and forest change in a bio-diversity hot spot. Science 303, 827–829.

Campbell, K., Frailey, C., Romero-Pittman, L., 2006. The pan-Amazonian Ucayali pene-plain: late Neogene sedimentation in Amazonia and the birth of the modern Amazonriver system. Palaeogeogr. Palaeoclimatol. Palaeoecol. 239, 166–219.

Chapin, F., Autumn, K., Pugnaire, F., 1993. Evolution of suites of traits in response to envi-ronmental stress. Am. Nat. 142, S78–S92.

Cheng, H., Sinha, A., Cruz, F., Wang, X., Edwards, R.L., d'Horta, F., Ribas, C., Vuile, M., Stott,L., Auler, A., 2013. Climate change patterns in Amazonia and biodiversity. Nat.Commun. 4, 1411.

Colinvaux, P., de Oliveira, P., Moreno, J., Miller, M., Bush, M., 1996. A long pollen recordfrom lowland Amazonia: forest cooling in glacial time. Science 274, 85–88.

Colinvaux, P., De Oliveira, P.E., Bush, M.B., 2000. Amazonian and neotropical plant com-munities on glacial time-scales: the failure of the aridity and refuge hypotheses.Quat. Sci. Rev. 19, 141–169.

Coomes, D., Allen, R., 2007. Effects of size, competition and altitude on tree growth. J. Ecol.95, 1084–1097.

Dufrene, M., Legendre, P., 1997. Species assemblages and indicator species: the need for aflexible asymmetrical approach. Ecol. Monogr. 67, 345–366.

Edwards, P., 1982. Studies of mineral cycling in a montane rain forest in New Guinea:rates of cycling in throughfall and litterfall. J. Ecol. 70, 807–827.

Enquist, B., Niklas, K., 2001. Invariant scaling relations across tree-dominated communi-ties. Nature 410, 655–660.

Enquist, B., Haskell, J., Tiffney, B., 2002. General patterns of taxonomic and biomasspartitioning in extant and fossil plant communities. Nature 419, 610–613.

Farrera, I., Harrison, S., Prentice, I., Ramstein, G., Guiot, J., Bartlein, P., Bonnefille, R., Bush,M., Cramer, W., von Grafenstein, U., Holmgren, K., Hooghiemstra, H., Hope, G., Jolly,D., Lauritzen, S., Ono, Y., Pinot, S., Stute, M., Yu, G., 1999. Tropical climates at theLast Glacial Maximum: a new synthesis of terrestrial palaeoclimate data. I. Vegeta-tion, lake-levels and geochemistry. Clim. Dyn. 15, 823–856.

Foster, R., Parker, T., Gentry, A., Emmons, L., Chiccon, A., Schulenberg, T., Rodriguez, L.,Lamas, G., Ortega, H., Icochea, J., Wust, W., Romo, M., Alban Castillo, J., Phillips, O.,Reynel, C., Kratter, A., Donahue, P., Barkley, L., 1994. The Tambopata–Candamo–RioHeath Region of Southeastern Peru: A Biological Assessment. University of ChicagoPress, Chicago, IL (184 pp.).

Gaston, K., 2003. The Structure and Dynamics of Geographic Ranges. Oxford UniversityPress, Oxford, UK.

Gentry, A., 1982. Neotropical floristic diversity: phytogeographical connections betweenCentral and South America, Pleistocene climaticfluctuations, or an accident of Andeanorogeny. Ann. Mo. Bot. Gard. 69, 557–593.

Goulding, M., Smith, N., 2007. Palmeras: Centinelas para la conservación de la Amazonía.Asociación para la Conservación de la Cuenca Amazónica (ACCA), Lima, Peru(356 pp.).

Groot, M., Bogotá, R., Lourens, L., Hooghiemstra, H., Vriend, M., Berrio, J., Tuenter, E., Van,J., der Plicht, B., Van, Geel, Ziegler, M., Weber, S., Betancourt, A., Contreras, L., Gavíria,S., Giraldo, C., González, N., Jansen, J., Konert, M., Ortega, D., Rangel, O., Sarmiento, G.,Vandenberghe, J., Van, T., der Hammen, M., der Linden, Van, Westerhoff, W., 2011.Ultra-high resolution pollen record from the northern Andes reveals rapid shifts inmontane climates within the last two glacial cycles. Clim. Past 7, 299–316.

Grubb, P., 1971. Interpretation of the ‘Massenerhebung’ effect on tropical mountains.Nature 229, 44–45.

Grubb, P., 1977. Control of forest growth and distribution onwet tropical mountains: withspecial reference to mineral nutrition. Annu. Rev. Ecol. Syst. 8, 83–107.

Harrison, S., Grace, J., 2007. Biogeographic affinity helps explain productivity–richnessrelationships at regional and local scales. Am. Nat. 170, S5–S15.

Holt, R., Barfield, M., 2009. Trophic interactions and range limits: the diverse roles ofpredation. Proc. R. Soc. B 276, 1435–1442.

Hooghiemstra, H., Berrio, J., Groot, M., Bogotá, R., Moscol-Olivera, M., González-Carranza,Z., 2012. The dynamic history of the upper forest line ecotone in the northern Andes.In: Randall, R.W. (Ed.), Ecotones Between Forest and Grassland. SpringerScience + Business Media, New York, Heidelberg, Dordrecht, London, pp. 229–246http://dx.doi.org/10.1007/978-1-4614-3797-0_10 (978-1-4614-3796-3 (printedbook), ISBN 978-1-4614-3797-0 (eBook), DOI 10.1007/978-1-4614-3797-0).

Hoorn, C., Wesselingh, P., ter Steege, H., Bermudez, M., Mora, A., Sevink, J., Sanmartín, I.,Sanchez-Meseguer, A., Anderson, C., Figueiredo, J., Jaramillo, C., Riff, D., Negri, F.,Hooghiemstra, H., Lundberg, J., Stadler, T., Särkinen, T., Antonelli, A., 2010. Amazoniathrough time: Andean uplift, climate change, landscape evolution, and biodiversity.Science 330, 927–931.

Householder, J., Janovec, J., Tobler, M., Lähteenoja, O., Page, S., 2012. Peatlands of theMadre de Dios River of Peru: distribution, geomorphology and habitat diversity.Wetlands 32, 359–368.

Jaeger, R., 1971. Competitive exclusion as a factor influencing the distributions of twospecies of terrestrial salamanders. Ecology 52, 632–637.

Janovec, J.P., Householder, J.E., Tobler, M., et al., 2013. Evaluación de los Actuales Impactosy Amenazas Inminentes en Aguajales y Cochas de Madre De Dios, Perú. WorldWildlife Fund (WWF), Lima, Peru (244 páginas, 180 color image and map figures,http://peru.panda.org/?217330/minamywwfpresentanimportanteestudiosobrelasituaciondeloshumedalesdemadrededios).

Juggins, S., 2012. rioja: analysis of Quaternary science data, R package version 0.7-3.http://cran.r-project.org/package=rioja.

Körner, C., 2007. The use of ‘altitude’ in ecological research. Trends Ecol. Evol. 22, 569–574.Kubitzki, K., 1989. The ecogeographical differentiation of Amazonian inundation forests.

Plant Syst. Evol. 162, 285–304.Lähteenoja, O., Ruokolainen, K., Schulman, L., Alvarez, J., 2009. Amazonian floodplains

harbour minerotrophic and ombrotrophic peatlands. Catena 79, 140–145.Leibold,M., Holyoak,M.,Mouquet, N., Amarasekare, P., Chase, J., Hoopes, M., Holt, R., Shurin,

J., Law, R., Tilman, D., Loreau, M., Gonzalez, A., 2004. The metacommunity concept: aframework for multi-scale community ecology. Ecol. Lett. 7, 601–613.

Liu, K., Colinvaux, P., 1985. Forest changes in the Amazon basin during the last glacialmaximum. Nature 318, 556–557.

Marchant, R., Almeida, L., Behling, H., Berrio, J., Bush, M., Cleef, A., Duivenvoorden, J.,Kappelle, M., de Oliveira, P., Texeira, A., de Oliveira, Filho, Lozano-García, S.,Hooghiemstra, H., Ledru, M., Ludlow-Wiechers, B., Markgraf, V., Mancini, V., Paez,M., Prieto, A., Rangel, O., Salgado-Labouriau, M., 2002. Distribution and ecology of par-ent taxa of pollen lodged within the Latin American Pollen Database. Rev. Palaeobot.Palynol. 121, 1–75.

Marchant, R., Harrison, S., Hooghiemstra, H., Markgraf, V., van Boxel, J., Ager, T., Almeida,L., Anderson, R., Baied, C., Behling, H., Berrio, J., Burbridge, R., Björk, S., Byrne, R., Bush,

148 J.E. Householder et al. / Palaeogeography, Palaeoclimatology, Palaeoecology 423 (2015) 138–148

M., Cleef, A., Duivenvoorden, J., Flenley, J., de Oliveira, P., van Geel, B., Graf, K., Gosling,W., Harbele, S., van der Hammen, T., Hansen, B., Horn, S., Islebe, G., Kuhry, P., Ledru,M., Mayle, F., Leyden, B., Lozano-García, S., Melief, A., Moreno, P., Moar, N., Prieto,A., van Reenen, G., Salgado-Labouriau, M., Schäbitz, F., Schreve-Brinkman, E., Wille,M., 2009. Pollen-based biome reconstructions for Latin America at 0, 6000 and18000 radiocarbon years. Clim. Past 5, 369–461.

Oliveira-Filho, A., Ratter, J., 1995. A study of the origin of central Brazilian forests by theanalysis of plant species distribution patterns. Edinb. J. Bot. 52, 141–194.

Page, S., Rieley, J., Shotyk, O., Weiss, D., 1999. Interdependence of peat and vegetation in atropical peat swamp. Philos. Trans. R. Soc. Lond. 354, 1885–1897.

Parmesan, C., 2006. Ecological and evolutionary responses to recent climate change.Annu. Rev. Ecol. Syst. 37, 637–669.

Phillips, O., Miller, J., 2002. Global Patterns of Plant Diversity: Alwyn H. Gentry's ForestTransect Data. Missouri Botanical Garden Press.

Pillon, Y., Munzinger, J., Amir, H., Lebrun, M., 2010. Ultramafic soils and species sorting inthe flora of New Caledonia. J. Ecol. 98, 1108–1116.

Prance, G., 1979. Notes on the vegetation of Amazonia III. The terminology of Amazonianforest types subject to inundation. Brittonia 31, 36–38.

Prior, L., Bowman, D., 2014. Across a macro-ecological gradient forest competition isstrongest at the productive sites. Front. Plant Sci. 5, 260.

Punyasena, S., 2008. Estimating neotropical palaeotemperature and palaeoprecipitationusing plant family climatic optima. Palaeogeogr. Palaeoclimatol. Palaeoecol. 265,226–237.

Punyasena, S., Eshel, G., McElwain, J., 2008. The influence of climate on the spatial pattern-ing of neotropical plant families. J. Biogeogr. 35, 117–130.

Punyasena, S., Dalling, J., Jaramillo, C., Turner, B., 2011. Comment on “The response ofvegetation on the Andean flank in western Amazonia to Pleistocene climate change”.Science 333, 1825.

Pyke, C., Condit, R., Aguilar, S., Lau, S., 2001. Floristic composition across a climatic gradi-ent in a neotropical lowland forest. J. Veg. Sci. 12, 553–566.

R Core Team, 2013. R: A Language and Environment for Statistical Computing. R Founda-tion for Statistical Computing, Vienna, Austria (URL http://www.R-project.org/).

Reich, P., Walters, M., Ellsworth, D., 1997. From tropics to tundra: global convergence inplant functioning. Science 94, 13730–13734.

Roberts, D., 2010. labdsv: ordination and multivariate analysis for ecology. R packageversion 1.4-1. http://CRAN.R-project.org/package=labdsv.

Roucoux, K., Lawson, I., Jones, T., Baker, T., Honorio Coronado, E., Gosling, W., Lähteenoja,O., 2013. Vegetation development in an Amazonian peatland. Palaeogeogr.Palaeoclimatol. Palaeoecol. 374, 242–255.

Runge, M., 1983. Physiology and ecology of nitrogen nutrition. In: Lange, O.L., Nobel, P.S.,Osmond, C.B., Ziegler, H. (Eds.), Physiological Plant Ecology III: Responses to theChemical and Biological Environment. Springer Verlag, Berlin, Germany p. 163–200.

Slivertown, J., Dodd, M., Gowing, D., 2001. Phylogeny and the niche structure of meadowplant communities. J. Ecol. 89, 428–435.

Snow, J., 1976. Northern South America; Columbia; Pacific coast section. In: Schwerdtfeger,W. (Ed.), Climates of Central and South America. World Survey of Climatology vol. 12.Elsevier, Amsterdam, pp. 370–376.

Steyermark, J., 1979. Flora of the Guayana Highland: endemicity of the generic flora of thesummits of the Venezuela Tepuis. Taxon 28, 45–54.

Stute, M., Forster, M., Frischkorn, H., Serejo, A., Clark, J., Schlosser, P., Broecker, W., Bonani,G., 1995. Cooling of tropical Brazil (5 degrees C) during the Last Glacial Maximum.Science 269, 379–383.

Tanner, E., Vitousek, P., Cuevas, E., 1998. Experimental investigation of nutrient limitationof forest growth on wet tropical mountains. Ecology 79, 10–22.

ter Braak, C., van Dam, H., 1989. Inferring pH from diatoms: a comparison of old and newcalibration methods. Hydrobiologia 178, 209–223.

Todzia, C., 1988. Flora Neotropica Vol 48, Chloranthaceae: Hedysosmum. New YorkBotanical Gardens, New York (138 pp.).

Unger, M., Leuschner, C., Homeier, J., 2010. Variability of indices of macronutrient avail-ability in soils at different spatial scales along an elevation transect in tropical moistforests (NE Ecuador). Plant Soil 336, 443–458.

USDA NRCS, 2013. The PLANTS Database (http://plants.usda.gov, 27 November 2013).National Plant Data Team, Greensboro, NC, USA, p. 27401-4901.

Van der Hammen, T., Hooghiemstra, H., 2000. Neogene and Quaternary history of vegeta-tion, climate and plant diversity in Amazonia. Quat. Sci. Rev. 19, 725–742.

van der Voet, H., 1994. Comparing the predictive accuracy of models using a simplerandomization test. Chemom. Intell. Lab. Syst. 25, 313–323.

Waughman, G., 1980. Chemical aspects of the ecology of some south German peatlands.J. Ecol. 68, 1025–1046.

Weng, C., Bush, M., Silman, M., 2004. An analysis of modern pollen rain on an elevationgradient in southern Peru. J. Trop. Ecol. 20, 113–124.

West, R., 1957. The Pacific lowlands of Colombia. Louisiana State University Press.Wiens, J., 2011. The niche, biogeography and species interactions. Philos. Trans. R. Soc.

Lond. 366, 2236–2250.Wiens, J., Donoghue, M., 2004. Historical biogeography, ecology and species richness.

Trends Ecol. Evol. 19, 639–644.Wiens, J., Graham, C., 2005. Niche conservatism: integrating evolution, ecology, and con-

servation biology. Annu. Rev. Ecol. Evol. Syst. 36, 519–539.Wiens, J., Ackerly, D., Allen, A., Anacker, B., Buckley, L., Cornell, H., Damshen, E., Davies, T.J.,

Grytnes, J.A., Harrison, S., Hawkins, B., Holt, R., McCain, C., Stephens, P., 2010. Nicheconservatism as an emerging principle in ecology and conservation biology. Ecol.Lett. 13, 1310–1324.

Wittmann, F., Householder, E., Piedade, M., de Assis, R., Schöngart, J., Parolin, P., Junk, W.,2013. Habitat specificity, endemism and the neotropical distribution of Amazonianwhite-water floodplain trees. Ecography 36, 690–707.