DOI: 10.1126/science.1090795, 827 (2004);303 Science et al.Mark B. Bush

48,000 Years of Climate and Forest Change in a Biodiversity Hot Spot

This copy is for your personal, non-commercial use only.

clicking here.colleagues, clients, or customers by , you can order high-quality copies for yourIf you wish to distribute this article to others

  here.following the guidelines

can be obtained byPermission to republish or repurpose articles or portions of articles

  ): March 28, 2014 www.sciencemag.org (this information is current as of

The following resources related to this article are available online at

http://www.sciencemag.org/content/303/5659/827.full.htmlversion of this article at:

including high-resolution figures, can be found in the onlineUpdated information and services,

http://www.sciencemag.org/content/suppl/2004/02/05/303.5659.827.DC1.html can be found at: Supporting Online Material

http://www.sciencemag.org/content/303/5659/827.full.html#ref-list-1, 4 of which can be accessed free:cites 16 articlesThis article

72 article(s) on the ISI Web of Sciencecited by This article has been

http://www.sciencemag.org/content/303/5659/827.full.html#related-urls13 articles hosted by HighWire Press; see:cited by This article has been

http://www.sciencemag.org/cgi/collection/atmosAtmospheric Science

subject collections:This article appears in the following

registered trademark of AAAS. is aScience2004 by the American Association for the Advancement of Science; all rights reserved. The title

CopyrightAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by theScience

on

Mar

ch 2

8, 2

014

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Mar

ch 2

8, 2

014

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Mar

ch 2

8, 2

014

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Mar

ch 2

8, 2

014

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

48,000 Years of Climate andForest Change in a Biodiversity

Hot SpotMark B. Bush,1*† Miles R. Silman,2*† Dunia H. Urrego1

A continuous 48,000-year-long paleoecological record from Neotropical lowermontane forest reveals a consistent forest presence and an ice-age cooling of�5° to 9°C. After 30,000 years of compositional stability, a steady turnover ofspecies marks the 8000-year-long transition from ice-age to Holocene condi-tions. Although the changes were directional, the rates of community changewere no different during this transitional period than in the preceding 30,000-year period of community stability. The warming rate of about 1°C per mil-lennium during the Pleistocene-Holocene transition was an order of magnitudeless than the projected changes for the 21st century.

The lower montane cloud forests (at eleva-tions of about 1300 to 2000 m) of theeastern Andes are a biodiversity hot spotand among the most threatened habitats onthe planet (1). Steep topography and highproportions of endemic species with narrowaltitudinal distributions make Andean sys-tems especially sensitive to past andpresent climate change. Habitat lossthrough deforestation and an anticipatedwarming of 1° to 4°C in the present century(2) raise concerns for the long-term surviv-al of these systems. Already, the effects ofclimate change on tropical montane sys-tems are evident in the contraction of icecaps (3) and the upslope migration of somebird species (4). Amazonian and high An-dean ecosystems maintained high diversityeven after strong Pleistocene climatic os-cillations (5–7), which might be taken toindicate that they would be resistant tofuture climate change. Here, we show thatalthough the Pleistocene-Holocene warm-ing (about 6°C) was greater than that antic-ipated within the next century, the rate ofclimate change was about an order of mag-nitude less, suggesting that these systemsface unprecedented community change.Our data, from an uninterrupted paleoeco-logical record spanning the past 48,000years from the lower montane forests of theAndes, show that this biodiversity hot spothas a history of profound but not rapidtemperature change.

Lago Consuelo (13°57�S, 68°59�W, el-evation of 1360 m; Fig. 1), Peru, is a

�10-ha lake in lower montane cloud forestthat lies in a small steep-sided catchmentunderlain by coarse-grained white quartzite(8). Parallel sediment cores that are 8.8 min length were collected from the center ofthe lake (water depth of 10.5 m) (9). Pollenanalyses (10) were conducted on the sedi-ments and a chronology was establishedbased on 10 accelerator mass spectrometry14C dates (table S1).

The Lago Consuelo pollen diagram re-veals markedly different Pleistocene and Ho-locene forest types (Fig. 2). Pollen of mon-tane elements that currently do not occuraround the lake, such as Alnus, Bocconia,Hedyosmum, and Vallea are unique to thePleistocene. Similarly, lowland elements,such as Acalypha, Alchornea, Celtis, Trema,and tree taxa of the Moraceae/Urticaceae at-tain their maximum representation in the Ho-locene pollen record. The transition fromPleistocene to Holocene is gradual with aconcurrent decline of montane taxa and anincrease in the abundance of lowland pollentaxa. The first signal of this transition, at ca.21,000 calibrated years before present (cal yrB.P.), is the rise of Moraceae/Urticaceae pol-len representation above its range for theprevious 10,000 years. The first montane tax-

on lost from the pollen record is Bocconia atca. 19,000 cal yr B.P. (Fig. 2). By 17,500 calyr B.P., a clear steady warming is evident inthe pollen record, a trend that continues un-abated until 10,000 cal yr B.P.

Fossil pollen data were analyzed using de-trended correspondence analysis (11) and pur-pose-designed transfer functions (12, 13). In-ferred paleotemperature data suggest two peri-ods of relatively stable temperatures with agradual change between states (Fig. 3). Becausethe pollen record suggests the continuous pres-ence of mesic forest and correspondingly moistair throughout the record, the modern empiri-cally derived lapse rate of 0.0055°C m�1 isapplied to translate elevation difference into achange in temperature (14). Because tempera-ture and moisture do not have independent in-fluences on elevational distributions, we cannotexclude the possibility that some of the Pleis-tocene “cooling” and temperature variabilitymay reflect increased precipitation or fog input.However, if the apparent cooling is exaggeratedby wet conditions, no “warming” marks theperiod at ca. 33,000 cal yr B.P., which was thedriest time of the late Pleistocene. At this time,Lake Titicaca was about 130 m below its mod-ern level (15) and northern Amazonian lakeswere also registering lowstands (7), but at LagoConsuelo, no substantial deviation from sur-rounding Pleistocene conditions is evident inthe transfer function data. We infer that temper-ature exerted a stronger influence than precipi-tation on the Pleistocene forests around LagoConsuelo. We conclude that the late Pleistocenewas at least 5°C, and possibly as much as 9°C,cooler than the present conditions.

The Pleistocene forests around Lago Con-suelo are most similar to modern communitiesat elevations of about 2500 m (Fig. 3) andexhibit a narrow scatter of high probabilitysolutions around that tendency. At ca. 17,500cal yr B.P., the transition to Holocene climatesis clearly evident as the “paleoelevation” de-clines steadily until reaching near modern con-ditions around 10,000 cal yr B.P. During theHolocene, the taxa in forest communitiesaround Lago Consuelo contained species withboth high- and low-elevational ranges. This

1Florida Institute of Technology, 150 West Univer-sity Boulevard, Melbourne, FL 32901–6975, USA.2Department of Biology, Wake Forest University,Box 7325, Reynolda Station, Winston Salem, NC27109–7325, USA.

*These authors contributed equally to this work.†To whom correspondence should be addressed. E-mail: mbush@fit.edu (M.B.B.); silmanmr@wfu.org(M.R.S.)

Sajama Ice core

20°S

0 Km200200 400

LagoConsuelo

10°S

70oW80°W

Titicacapollen site

SouthAmerica

Fig. 1. Location of Lago Consuelorelative to Lake Titicaca and theSajama ice cap.

R E P O R T S

www.sciencemag.org SCIENCE VOL 303 6 FEBRUARY 2004 827

composition results in a broader range of pos-sible temperature solutions compared withthose of the Pleistocene. Holocene elevationestimates are scattered around a mean of1300 m, but the mode is between 700 and1000 m. This apparent discrepancy with LagoConsuelo’s elevation of 1360 m reflects theactual distribution of genera on the Andeanflank and the use of presence-absence data in alowland system. Whereas the high-elevationslopes are rich in palynologically distinct taxawith well-defined elevational ranges, the lowerslopes have few such taxa (fig. S2). Indeed, ofthe genera found in the pollen record, only one,the palm Dictyocaryum, has its peak probabilityof distribution at 1400 m. Given this paucity ofmid-elevation markers, lowland taxa with peakdistributions at 1000 m or less dominate themid-Holocene climatic signal.

The lack of empirical data from this region

(a problem shared by most Andean analyses)does not allow precipitation and temperature tobe estimated separately in the transfer function.Although the transfer function appears to beprimarily driven by temperature, some influ-ence of precipitation cannot be discounted.However, the nearly constant sedimentationrate and absence of abrupt changes in sedimenttype suggest that Lago Consuelo never driedout and that there were no major drought epi-sodes. Although seasonality and temperaturechanged, wet Andean cloud forest was a con-tinuous feature of this landscape, only retreatingupslope during the driest interval of the Holo-cene. During this event, ca. 9000 to 5000 cal yrB.P., the inferred lifting of the cloud basecaused a local replacement of cloud forest taxawith those of wet lowland forest.

Ordination of the fossil pollen data fromLago Consuelo reveals Pleistocene samples that

were clustered together and a separate tightclustering of Holocene samples (fig. S3A), sug-gesting relatively stable communities in thePleistocene and the Holocene, respectively. Thetransitional communities connecting these flo-ras form a nearly linear connection. Plotting thecompositional dissimilarity (16) between pastcommunities and the modern flora reinforcesthe concept of two distinct and stable foresttypes of the Pleistocene and Holocene, as wellas the gradual change between them (fig. S3B).Another important trend is that the amplitude ofcommunity change taking place within thePleistocene is similar to that taking place withinthe Holocene. Indeed, the amount of changebetween samples during the thermally “stable”periods is similar to the change that took placeduring the transitional period (fig. S3, A to C).The only difference between transitional andstable forest types is whether the changes aredirectional. These observations are more con-sistent with hypotheses of stabilizing forcesinfluencing communities (17) than hypothesesof neutral drift (18). During stable periods,shifts in community composition take placecontinually and may be either nondirectional orcorrespond to the vegetation closely trackingshort-term climatic oscillations, whereas intransitional stages most variation results fromclimatically driven vegetation succession.

An important observation from thesedata is that even during the times of great-est community change in response to thePleistocene-Holocene transition, the rate ofchange in forest composition was no great-er than that of the average ongoing changesdue to either community drift or tracking ofsmall-scale climatic fluctuations. Thus, atany given time, an observer monitoringforest structure would find directionalchanges ecologically indistinguishablefrom the more stochastic changes of stableperiods. Unlike temperate systems in whichlong-distance migrations are required fortaxa to remain in equilibrium with climate(19), species migration in Andean systemsprobably did not require long-distance dis-persal events or any special mechanismbeyond normal seed dispersal around a par-ent. Because of the slopes in Andean sys-tems, species can remain in equilibriumwith temperature shifts of 1° to 2°C withmigrations of �1 km. Indeed, we hypothe-size that the ease with which species canmigrate and avoid extinction in response toclimate change may be a major factor pro-moting diversity in these systems.

The slow change in community compo-sition at Lago Consuelo is similar in char-acter to the pollen record of Lake Titicaca,one of the most detailed reconstructions ofthe Pleistocene/Holocene boundary fromthe Andes (20). Titicaca, which lies abovethe tree line and immediately upslope ofLago Consuelo, manifested a similar grad-

Fig. 2. Percentage pol-len diagram of select-ed taxa from LagoConsuelo, Peru. Sumsbased on 300 terres-trial pollen grains.Taxa are ordered ac-cording to their appar-ent sensitivity towarming. Montane el-ements are listed tothe left and lowlandelements to the right.Total taxa obtainedfrom the site include181 terrestrial pollentypes (92% of whichwere identified), 4aquatic pollen types,and 18 terrestrial fernand moss spore types.Aquatic taxa andspores are excludedfrom the pollen sum.

Fig. 3. Most likely equiv-alent modern elevationand inferred paleotem-perature for pollen spec-tra from Lago Consuelo,Peru. Presence-absencedata from 26 taxa withwell-characterized mod-ern distributions. Colorintensity (blue low to or-ange high) reflects proba-bility that a sample con-taining a given set of taxacame from that elevation.

R E P O R T S

6 FEBRUARY 2004 VOL 303 SCIENCE www.sciencemag.org828

ual warming between 21,000 and 11,000cal yr B.P. Our data are also consistent withother Andean sites where pollen recordsshow abrupt transitions between forest andpuna vegetation (5, 6, 21).

The transition at a given site betweenthe tree line and puna takes place within aclimate space of �1°C (22). Thus, a nota-ble change in biome type can be recorded inpollen records without the full Pleistocene-Holocene temperature shift coinciding withthe obvious transition between biomes. Ourdata are consistent with the isotopic recordfrom the only regional ice core that spansthe past 20,000 years (14). The stable ox-ygen isotopic record of the Pleistocene-Holocene boundary from Sajama, Bolivia,exhibits a steady increase in �18O(ice) ratiosinterrupted by one substantial abrupt oscil-lation in the terminal Pleistocene. Thoughinitially aligned to the Younger Dryasevent (3), this oscillation has subsequentlybeen suggested to reflect increased precip-itation associated with the formation ofPaleolake Coipasa rather than tempera-ture change (15). Thus, no evidence-exists of abrupt temperature change in theSajama record.

In the Holocene, the record at Lago Con-suelo is consistent with those of the highAndes (23), coastal Peru (24), and the Boliv-ian rainforest-savanna ecotone (25) and isconsistent with a strong mid-Holocene dryevent. Paleotemperature estimates based onspecies distributions and lapse rate from ca.8200 and 4000 cal yr B.P. average 1° to 3°Chigher than the modern temperature with rel-atively little statistical error (Fig. 3). Howev-er, the apparent temperature increases, man-ifested by little change in the lowland taxaand lack of montane taxa in the pollen record,could also reflect moisture changes. If thiswas a time of reduced Andean precipitation,the absence of cloud forest species could beexplained by drier conditions, causing thecloud base to lift, rather than or in addition toa simple warming.

These data demonstrate the importanceof the buffering force of cloud on theseecosystems. During the strongest dryingevent of the Pleistocene, aquatic pollentaxa show that the lake level at Lago Con-suelo responded somewhat to the drought,but the cloud forest showed no desiccationeffect at an elevation of 1360 m (Fig. 2).However, in the mid-Holocene, when thecloud base had lifted to its modern position,a more modest drying event caused thedecline of cloud forest elements around thelake. Clearly, although temperature canforce elements to migrate vertically,drought will have its most immediateeffect on the lower limit of the cloud for-est by eliminating moisture intercep-tion from clouds, which can provide up to

37% of cloud forest moisture budgets (26).The Lago Consuelo record establishes

several key points. First, cloud forest, al-though changing between stable communi-ty states, existed at or close to this elevationfor the past 48,000 years. Second, thetemperature change from Pleistocene toHolocene began ca. 19,000 cal yr B.P. andprogressed steadily, with no sudden accel-erations. Third, rates of forest communitychange during the most substantial climatechange of the past were no greater thanduring the relative stability of the pre-Industrial Holocene.

Abrupt movement of the Andean tree lineat the Pleistocene-Holocene boundary and thepresence of many endemic species may pro-vide a misleading impression of communityresilience to climate change. Our data suggestthat climate changes in these systems havebeen gradual, perhaps averaging less than1°C per millennium even during the height ofthe Pleistocene-Holocene transition. Conser-vatively, if temperatures change only 1°C thiscentury (the minimum International Panel onClimate Change estimate), that rate would bean order of magnitude higher than the fastestrate observed in this record; the projectedclimate change in the next 100 years will befundamentally different from any in the last50,000 years. Given the relatively short geo-graphic distances between elevations and theconcomitantly short migration distances re-quired to move among them, Andean plantswith broad elevational distributions should beable to remain in equilibrium with climate.For taxa with narrow elevation ranges, how-ever, the predicted rate of climate changemay move them completely outside of theirclimatic niche space within only one or twoplant generations. Coupled with habitat de-struction preventing colonization from adja-cent metacommunities, Andean plant com-munities may experience greatly increasedextinction rates.

References and Notes1. N. Myers, R. A. Mittermeier, C. G. Mittermeier, G. A. B.

da Fonseca, J. Kent, Nature 403, 853 (2000).2. International Panel on Climate Change, IPCC Summa-

ry for Policymakers. Climate Change 2001: Impacts,Adaptation and Vulnerability. A Report of WorkingGroup II. (Cambridge Univ. Press, Cambridge, 2001).

3. L. G. Thompson et al., Science 298, 589 (2002).4. J. A. Pounds, M. P. L. Fogden, J. H. Campbell, Nature398, 611 (1999).

5. H. Hooghiemstra, A. Cleef, in Biodiversity and Con-servation of Neotropical Montane Forests, E. Forero,Ed. ( The New York Botanical Garden, New York,1995), pp. 35–49.

6. P. A. Colinvaux, M. B. Bush, M. Steinitz-Kannan, M. C.Miller, Quat. Res. 48, 69 (1997).

7. M. B. Bush, M. C. Miller, P. E. De Oliveira, P. A.Colinvaux, J. Paleolimnol. 27, 341 (2002).

8. The persistence of the cloud base forming at about1350 m was evident in the transition from organic-poor soils to those with a duff layer above thiselevation. Rocotal, Kosnipata Valley, Peru, is the clos-est weather station at approximately the same ex-posure and elevation on the east Andean slope at

2010 m and 13°06�47��S, 71°34�14��W. The 2-yearprecipitation average is 5452 mm.

9. P. A. Colinvaux, P. E. De Oliveira, J. E. Moreno, AmazonPollen Manual and Atlas (Harwood Academic Press,New York, 1999).

10. K. Faegri, J. Iversen, Textbook of Pollen Analysis(Wiley, Chichester, ed. 4, 1989).

11. DECORANA (A FORTRAN program for detrended cor-respondence analysis and reciprocal averaging), Ecol-ogy and Systematics; M. O. Hill, Cornell University.

12. Modern pollen studies indicate that within Neotro-pical forested settings, pollen representation far fromthe parent tree is statistically insignificant and thatdistributions of taxa used in these transfer functionsare reflected faithfully in pollen occurrence (27, 28).

13. Searches were conducted using TROPICOS (www.mobot.org/tropicos/spectest7.xml). Nonparametricempirical probability distributions were fit for 26pollen taxa in S-PLUS 6.1 (Insightful Corp., 2002) andare presented in the supporting online material text.We use modern distributional data from herbariumcollections grouped at the same level as observedpollen taxa to estimate the probability of a collectionof that taxon coming from a given altitude. Taxawere included in the elevation model for a samplewhen they exceeded 10% or their maximum repre-sentation. Cut-offs of 5 and 15% did not affect theconclusions for the model (29). We first estimatedthe probability Pr (taxon A � elevation). Bayes’ theo-rem was then used with a uniform prior to estimatethe most likely elevation from which a given set oftaxa originates, Pr (elevation � taxa A, B, C). Thismethod gives a probability distribution of elevationfor a group of observed pollen taxa that can also beused to infer a “paleoelevation” for fossil pollenspectra. Taxa used were: Alchornea, Alnus, Alsophylla,Begonia, Bocconia, Celtis, Clethra, Cyathea, Cystop-teris, Dictyocaryum, Gaiadendron, Hedyosmum, Hy-eronima, Ilex, Juglans, Myrica, Myrsine, Piper, Podo-carpus, Polylepis, Symplocos, Trema, Vallea, Wein-mannia, Wettinia, and Zanthoxylum.

14. Measurement of the adiabatic lapse rate of localmoist air is based on an elevational transect ofdataloggers in the nearby forests of the KosnipataValley, Peru, that measured temperature and humid-ity over an 8-month period.

15. P. A. Baker et al., Nature 409, 698 (2001).16. The compositional dissimilarity assumes that the

maximum wet season (December-January-February)insolation equates with periods of maximum precip-itation.

17. J. S. Clark, J. S. McLachlan, Nature 423, 635 (2003).18. S. P. Hubbell, The Unified Neutral Theory of Biodiver-

sity and Biogeography (Princeton Univ. Press, Prince-ton, NJ, 2001).

19. J. S. Clark, M. Lewis, L. Horvath, Am. Nat. 157, 537(2001).

20. G. M. Paduano, M. B. Bush, P. A. Baker, S. L. Fritz, G. O.Seltzer, Palaeogeogr. Palaeoclimatol. Palaeoecol.194, 259 (2003).

21. B. C. S. Hansen, D. T. Rodbell, Quat. Res. 44, 216(1995).

22. R. A. J. Grabandt, thesis, University of Amsterdam(1985).

23. P. A. Baker et al., Science 291, 640 (2001).24. D. H. Sandweiss, K. A. Maasch, D. G. Anderson, Sci-

ence 283, 499 (1999).25. F. E. Mayle, R. Burbridge, T. J. Killeen, Science 290,

2291 (2000).26. J. Cavelier, M. Jaramillo, D. Solis, D. deLeon, J. Hydrol.193, 83 (1997).

27. M. B. Bush, Quat. Res. 54, 132 (2000).28. C. Weng et al., J. Trop. Ecol. 20, 1 (2004).29. B. W. Silverman, Density Estimation for Statistics and

Data Analysis (Chapman and Hall, London, 1986).30. Supported by grants from NSF (DEB-0237573,

-0237684, and -9732951).

Supporting Online Materialwww.sciencemag.org/cgi/content/full/303/5659/827/DC1SOM TextFigs. S1 to S3Table S1

25 August 2003; accepted 7 January 2004

R E P O R T S

www.sciencemag.org SCIENCE VOL 303 6 FEBRUARY 2004 829