Responses of Amazonian ecosystems to climatic and atmospheric carbon dioxide changes since ...

16 Pages·2004·0.28 MB·English

Publishedonline11February2004
Responses of Amazonian ecosystems to climatic
and atmospheric carbon dioxide changes since the
last glacial maximum
Francis E. Mayle1*, David J. Beerling2, William D. Gosling1,3 and Mark B. Bush3
1Department of Geography, University of Leicester, Leicester LE1 7RH, UK
2Department of Animal and Plant Sciences, University of Sheffield, Sheffield S10 2TN, UK ([email protected])
3Department of Biological Sciences, Florida Institute of Technology, 150 West University Blvd, Melbourne, FL 32901, USA
([email protected])
Theaimsofthispaperaretoreviewpreviouslypublishedpalaeovegetationandindependentpalaeoclimatic
datasets together with new results we present from dynamic vegetation model simulations and modern
pollen rain studies to: (i) determine the responses of Amazonian ecosystems to changes in temperature,
precipitation and atmospheric CO concentrations that occurred since the last glacial maximum (LGM),
2
ca. 21000years ago; and (ii) use this long-term perspective to predict the likely vegetation responses to
futureclimatechange.AmazoniaremainedpredominantlyforestedattheLGM,althoughthecombination
ofreducedtemperatures,precipitationandatmosphericCO concentrationsresultedinforestsstructurally
2
andfloristicallyquitedifferentfromthoseoftoday.Cold-adaptedAndeantaxamixedwithrainforesttaxa
in central areas, while dry forest species and lianas probably became important in the more seasonal
southern Amazon forestsand savannahsexpandedat forest–savannahecotones. Net primaryproductivity
(NPP)andcanopydensityweresignificantlylowerthantoday.EvergreenrainforestdistributionandNPP
increased during the glacial(cid:1)Holocene transition owing to ameliorating climatic and CO conditions.
2
However,reducedprecipitationintheEarly–Mid-Holocene(ca.8000–3600yearsago)causedwidespread,
frequentfiresinseasonalsouthern Amazonia,causingincreased abundanceofdrought-tolerant dryforest
taxa and savannahs in ecotonal areas. Rainforests expanded once more in the Late Holocene owing to
increasedprecipitationcausedbygreateraustralsummerinsolation,althoughsomeofthisforestexpansion
(e.g. in parts of the Bolivian Beni) is clearly caused by palaeo Indian landscape modification. The plant
communitiesthatexistedduringtheEarly–Mid-Holocenemayprovideinsightsintothekindsofvegetation
responseexpectedfromsimilarincreasesintemperatureandariditypredictedforthetwenty-firstcentury.
We infer that ecotonal areas near the margins of the Amazon Basin are liable to be most sensitive to
future environmental change and should therefore be targeted with conservation strategies that allow
‘natural’ species movements and plant community re-assortments to occur.
Keywords:Amazontropicalforest; palaeoclimate; carbondioxide; lastglacialmaximum; Holocene
1. INTRODUCTION tolerances,differentspecieswouldbepredictedtorespond
individualistically to climate change. Consequently, not
The Amazon Basin is predicted to experience an increase
only would one expect biome shifts (e.g. replacement of
in temperatures by ca. 3°C coupled with a reduction in
rainforest by seasonally dry forest or savannah), but also
precipitation by ca. 20% over the twenty-first century
significantreassortmentofspecieswithin plantcommuni-
(Houghton et al. 2001). These climatic changes would
ties in response to such changes.
reduce plant water availability and thereby increase
Insights into the nature of such anticipated vegetation
drought stress for many Amazonian species. Climatically
responses can beobtained from an understanding of how
sensitiveecotonalareasarelikelytobemostvulnerableto
Amazonian ecosystems have responded to environmental
increased drought and warmth; i.e. the rainforest–
changes in the past. The aim of this paper is therefore to
savannah and rainforest–dry forest ecotones at the north-
determine the responses of Amazonian ecosystems to the
ern and southern limits of the Basin, and the rainforest–
significant changes in temperature, precipitation and
Andean cloud forest ecotone at the western limit of the
atmospheric CO concentrations that occurred since the
Basin. Given that different species have different climatic 2
LGM, ca. 21000 calendar years before present (cal yr
BP), and synthesize this information to facilitate predic-
tions about likely vegetation responses to future climate
*Author for correspondence ([email protected]). Address from July
change. Towards this aim, we (i) review previously pub-
2004:InstituteofGeography,SchoolofGeosciences,TheUniversityof
Edinburgh,DrummondStreet,EdinburghEH89XP,UK. lished palaeovegetation and pollen rain datasets, inde-
pendent multi-proxy palaeoclimatic data and dynamic
One contribution of 17 to a Theme Issue ‘Tropical forests and global
atmosphericchange’. vegetation modelsimulations; and (ii) presentnew pollen
Phil.Trans.R.Soc.Lond.B(2004)359,499–514 499 2004TheRoyalSociety
DOI10.1098/rstb.2003.1434
500 F. E. Mayle and others Amazonian vegetation changes since the LGM
raindatafromtheBolivianAmazonandvegetationsimul- Figure1. (a) Map showing the location of sites discussed in
ations for the Amazon Basin. the text. The shaded area shows the current distribution of
humid evergreen broad-leaf forest (rainforest). The hatched
area shows the Andes mountains. Lowland unshaded areas
2. AMAZONIA AT THE LAST GLACIAL MAXIMUM represent seasonally dry forests and savannahs. (b) Schematic
broad-scale summary trends of climatic change and/or
(a) Climate and carbon dioxide concentrations
vegetation response for each site since 21000 cal yr BP.
Elucidating species’ responses to Amazonian climate at
N.B. These profiles are intended as simplified cartoons to
theLGMhaslongbeenproblematical,inlargepartowing
summarize the information discussed in the text and should
to thefactthat,in theabsenceofotherkindsofevidence, not be viewed as absolute quantitative depictions. Key for
palaeoclimate and palaeovegetation reconstructions have sites: HU, Huascaran ice core (Thompson et al. 1995;
generally both been obtained from the same proxy data, 9°07(cid:3)S, 77°37(cid:3)W); JU, Lake Junin (Seltzer et al. 2000;
namelyfossilpollenrecords.Asidefromtheobviousprob- 11°S, 76°W); TI, Lake Titicaca (Baker et al. 2001a; 17°S,
lem of circular reasoning, this approach is also flawed 69°W); SA, Sajama ice core (Thompson et al. 1998;
because it rests on the assumption that climate was the 18°06(cid:3)S, 68°53(cid:3)W); SI, Siberia (Mourguiart & Ledru 2003;
17°50(cid:3)00(cid:4)S, 64°43(cid:3)08(cid:4)W); PI, Laguna El Pinal (Behling &
only important environmental variable responsible for
Hooghiemstra 1999; 4°08(cid:3)N, 70°23(cid:3)W); LL, Laguna Loma
controllingspeciesdistributions.Itisnowwellestablished
Linda (Behling & Hooghiemstra 2000; 3°18(cid:3)N, 73°23(cid:3)W);
from ecophysiological experiments (Polley et al. 1993; MX, Maxus 4 (Weng et al. 2002; 0°27(cid:3)S, 76°37(cid:3)W); PA:
Cowling & Sage 1998), dynamic vegetation models Lake Pata (Colinvaux et al. 1996; Bush et al. 2002; 0°16(cid:3)N,
(Cowling et al. 2001) and stable carbon isotope records 66°41(cid:3)W); CA, Lago Calado (Behling et al. 2001; 3°16(cid:3)S,
(Street-Perrott et al. 1997) that the carbon-depleted 60°35(cid:3)W); VE, Porto Velho/Humaita (De Freitas et al.
atmosphereoftheLGM(ca.180–200p.p.m.atmospheric 2001; 8°43(cid:3)S, 63°58(cid:3)W to 7°38(cid:3)S, 63°04(cid:3)W); BV, Laguna
CO versus pre-industrial values of 280p.p.m.; Monnin Bella Vista (Mayle et al. 2000; Burbridge et al. 2004;
2
et al. 2001) significantly reduced the water-use efficiency 13°37(cid:3)S, 61°33(cid:3)W); CH, Laguna Chaplin (Mayle et al.
(waterlostperunitofcarbonuptake)ofC plants(i.e.all 2000; Burbridge et al. 2004; 14°28(cid:3)S, 61°04(cid:3)W); SC, Serra
3 dos Carajas (Absy et al. 1991; Sifeddine et al. 2001; 6°35(cid:3)S,
woodyplants).StudiesbyHuangetal.(2001)suggestthat
49°30(cid:3)W); FA1, Amazon Fan (Haberle & Maslin 1999;
thiswasespeciallyimportantinthehighlyseasonalforest–
5°12.7(cid:3)N, 47°1.8(cid:3)W); FA2, Amazon Fan (Maslin & Burns
savannah ecotones at the northern and southern limits of
2000; 5°45(cid:3)N, 49°06(cid:3)W); GA, Laguna La Gaiba
the Basin, where species already under climate-induced (17°47(cid:3)00(cid:4)S, 57°43(cid:3)00(cid:4)W); MA, Laguna Mandiore´
water stress were most vulnerable to additional water- (18°05(cid:3)31(cid:4)S, 57°33(cid:3)46(cid:4)W); SO, Laguna Soco´rros
stress brought about by low CO2 concentrations. Clearly, (16°08(cid:3)30(cid:4)S, 63°07(cid:3)00(cid:4)W); RP, Rio Piray (Servant et al.
Amazonian species, especially those in the distinctly sea- 1981; 17°55(cid:3)S, 63°15(cid:3)W); OS, Ostra (Andrus et al. 2002;
sonalsouthernhalfoftheBasin,wouldhavebeenaffected, 8°55(cid:3)S); SS, Siches (Andrus et al. 2002; 4°40(cid:3)S).
not only by different climatic conditions, but also low
atmospheric CO conditions.
2
Fortunately, an increasing body of evidence from inde- palaeolimnological data from a cluster of lakes (Lakes
pendent climate proxies has enabled more reliable recon- Pata, Verde and Dragao) on an inselberg in northwest
structions of Amazonian LGM climate to be made. Amazonia(0°16(cid:3)N,66°41(cid:3)W; figure1) toinferthatlake
Analysisofnoblegasconcentrationsinfossilgroundwater levels were relatively high at the LGM (although below
in eastern Brazil (Stute et al. 1995) and tropical SST present-day levels) and that humid conditions existed in
reconstructions from Barbados corals (Guilderson et al. northwest Amazonia at this time. Corroboratingevidence
1994) show that the Amazon Basin at the LGM was ca. for a humid climate is supported by complete absence of
5°Ccoolerthantoday.Thisestimateissupportedby(cid:2)18O charcoalinthesedimentrecordfromthesesites(M.Bush,
analyses of Andean ice-core records at Huascaran, Peru unpublished data), indicating that the climate was too
(9°07(cid:3)S, 77°37(cid:3)W; Thompson et al. 1995) and Sajama, humidtofavourfrequentforestfires.Furtherevidencefor
Bolivia (18°06(cid:3)S, 68°53(cid:3)W; Thompson et al. 1998; fig- wet climatic conditions comes from palaeolimnological
ure 1). data from Lake Titicaca (16–17.5°S, 68.5–70°W; Baker
Precipitation values and patterns at the LGM are less et al. 2001a), located on the Peruvian–Bolivian Altiplano
well understood, for several reasons. First, in contrast to just beyond the southwest margin of the Amazon water-
the temperature and CO variables, precipitation is likely shed. The water-level reconstruction from this site shows
2
to have been highly spatially variable across the Amazon that precipitation on the Altiplano was even higher than
Basin, as it is today. Thus, large-scale extrapolations today, and that the lake overflowed southwards to form
beyond the immediate site of investigation may well be the huge palaeolake ‘Tauca’ on the central Altiplano
highlyerroneous.Second,inadditiontodifferencesinthe (Baker et al. 2001b). Hastenrath & Kutzbach (1985) and
magnitude of precipitation, the direction of precipitation Blodgett et al. (1997) simulated the hydrological con-
changeovertimemayalsohavedifferedacrossAmazonia, ditions on the Altiplano by using different models,
given that the orbital forcing of insolation between the incorporating surface energy and water budgets, which
Northern and Southern Hemispheres of the Basin are in showed that precipitation must have been 20–75% above
anti-phase(Berger &Loutre1991)and therelativedomi- modern levels to produce Pleistocene lake Tauca,
nance of different climate drivers is likely to have varied assuminga5°Ctemperaturedrop.Additionalcorroborating
in different parts of this vast area. Notwithstanding these data for a wet Altiplano come from geochemical and ice-
problems, significant progress has been made in recent accumulation-rate data from the Sajama Mountain ice
years in reconstructing the precipitation regime across core (Thompson et al. 1998), south of Lake Titicaca in
Amazonia at the LGM. Bush et al. (2002) used theBolivianAndes, a sitethat alsoreceivesthebulk ofits
Phil.Trans.R.Soc.Lond.B(2004)
Amazonian vegetation changes since the LGM F. E. Mayle and others 501
(cid:12)(cid:1)(cid:13) (cid:14)(cid:5)(cid:3)(cid:4) (cid:6)(cid:2)(cid:3)(cid:4) (cid:6)(cid:5)(cid:3)(cid:4) (cid:1)(cid:2)(cid:3)(cid:4) (cid:1)(cid:5)(cid:3)(cid:4) (cid:2)(cid:2)(cid:3)(cid:4) (cid:2)(cid:5)(cid:3)(cid:4) (cid:11)(cid:2)(cid:3)(cid:4)
(cid:9)(cid:5)(cid:3)(cid:10) (cid:9)(cid:5)(cid:3)(cid:10)
(cid:8)(cid:2)(cid:21)
(cid:2)(cid:3)(cid:10) (cid:8)(cid:2)(cid:9) (cid:2)(cid:3)(cid:10)
(cid:6)(cid:7)
(cid:5)(cid:5)
(cid:5)(cid:3) (cid:3)(cid:4) (cid:6)(cid:2) (cid:5)(cid:3)
(cid:10)(cid:2)
(cid:1)(cid:1)(cid:1)(cid:1)
(cid:2)(cid:3)(cid:8) (cid:2)(cid:3)(cid:8)
(cid:1)(cid:10)
(cid:11)(cid:12)
(cid:13)(cid:1) (cid:14)(cid:14)(cid:15)(cid:15)
(cid:9)(cid:5)(cid:3)(cid:8)
(cid:9)(cid:5)(cid:3)(cid:8)
(cid:16)(cid:16)(cid:15)(cid:15)
(cid:18)(cid:11)
(cid:10)(cid:14)
(cid:9)(cid:2)(cid:3)(cid:8)
(cid:9)(cid:2)(cid:3)(cid:8) (cid:17)(cid:17)(cid:7)(cid:7) (cid:1)(cid:13)
(cid:20)(cid:2)
(cid:1)(cid:1)(cid:2)(cid:2) (cid:1)(cid:1)(cid:7)(cid:7) (cid:19)(cid:6) (cid:3)(cid:2)
(cid:5) 41 (cid:1)(cid:5)(cid:5)
(cid:7)(cid:5)(cid:3)(cid:8)
(cid:7)(cid:5)(cid:3)(cid:8)
(cid:14)(cid:5)(cid:3)(cid:4) (cid:6)(cid:2)(cid:3)(cid:4) (cid:6)(cid:5)(cid:3)(cid:4) (cid:1)(cid:2)(cid:3)(cid:4) (cid:1)(cid:5)(cid:3)(cid:4) (cid:2)(cid:2)(cid:3)(cid:4) (cid:2)(cid:5)(cid:3)(cid:4) (cid:11)(cid:2)(cid:3)(cid:4)
(cid:12)(cid:2)(cid:13) (cid:16) (cid:15)(cid:16) (cid:31)(cid:26) (cid:8)(cid:27) (cid:8)(cid:26)
(cid:5) (cid:5) (cid:5) (cid:5) (cid:5)
(cid:7) (cid:7) (cid:7) (cid:7) (cid:7)
(cid:11) (cid:11) (cid:11) (cid:11) (cid:11)
(cid:1) (cid:1) (cid:1) (cid:1) (cid:1)
(cid:14) (cid:14) (cid:14) (cid:14) (cid:14)
(cid:9)(cid:5) (cid:9)(cid:5) (cid:9)(cid:5) (cid:9)(cid:5) (cid:9)(cid:5)
(cid:9)(cid:7) (cid:9)(cid:7) (cid:9)(cid:7) (cid:9)(cid:7) (cid:9)(cid:7)
(cid:9)(cid:11) (cid:9)(cid:11) (cid:9)(cid:11) (cid:9)(cid:11) (cid:9)(cid:11)
(cid:9)(cid:1) (cid:9)(cid:1) (cid:9)(cid:1) (cid:9)(cid:1) (cid:9)(cid:1)
(cid:9)(cid:14) (cid:9)(cid:14) (cid:9)(cid:14) (cid:9)(cid:14)
(cid:7)(cid:5) (cid:7)(cid:5) (cid:7)(cid:5) (cid:7)(cid:5)
""#(cid:30)(cid:26) (cid:28)(cid:29) (cid:30)(cid:27) (cid:17)(cid:27) (cid:24)(cid:25)
(cid:5) (cid:5) (cid:5) (cid:5) (cid:5)
(cid:7) (cid:7) (cid:7) (cid:7) (cid:7)
(cid:11) (cid:11) (cid:11) (cid:11) (cid:11)
(cid:1) (cid:1) (cid:1) (cid:1) (cid:1)
(cid:14) (cid:14) (cid:14) (cid:14) (cid:14)
(cid:9)(cid:5) (cid:9)(cid:5) (cid:9)(cid:5)
(cid:9)(cid:7) (cid:9)(cid:7) (cid:9)(cid:7)
(cid:9)(cid:11) (cid:9)(cid:11) (cid:9)(cid:11)
(cid:9)(cid:1) (cid:9)(cid:1) (cid:9)(cid:1)
(cid:9)(cid:14) (cid:9)(cid:14) (cid:9)(cid:14)
(cid:7)(cid:5) (cid:7)(cid:5)
$(cid:24)#(cid:17) (cid:8)(cid:17) !(cid:27)(cid:9) !(cid:27)(cid:7) %&’((cid:23)(cid:21)
(cid:5) (cid:5) (cid:5) (cid:5)
(cid:7) (cid:7) (cid:7) (cid:7) (cid:23)(cid:20))(cid:20)**(cid:20)(cid:18)
(cid:11) (cid:11) (cid:11) (cid:11)
(cid:1) (cid:1) (cid:1) (cid:1) &+(*,)(-((cid:21)(cid:20)(cid:21)(cid:19)&*
(cid:14) (cid:14) (cid:14) (cid:14)
.’/
(cid:9)(cid:5) (cid:9)(cid:5) (cid:9)(cid:5) (cid:9)(cid:5)
(cid:9)(cid:7) (cid:9)(cid:7) (cid:9)(cid:7) (cid:9)(cid:7) 0(cid:20)’1
(cid:9)(cid:11) (cid:9)(cid:11) (cid:9)(cid:11) (cid:9)(cid:11)
(cid:9)(cid:1) (cid:9)(cid:1) (cid:9)(cid:1) 2&3.
(cid:9)(cid:14) (cid:9)(cid:14) (cid:18)(cid:19)(cid:20)(cid:21)(cid:22)(cid:23) (cid:9)(cid:14)
(cid:7)(cid:5) (cid:7)(cid:5) (cid:7)(cid:5) 0((cid:21)
Figure1. (Caption opposite.)
Phil.Trans.R.Soc.Lond.B(2004)
502 F. E. Mayle and others Amazonian vegetation changes since the LGM
precipitation today from the Amazon Basin. It should be SSTsintheequatorialAtlanticwouldbeexpectedtohave
notedthough,thatalthoughtheseice-coredatashowthat reducedoceanicmoisturetransfertotheBasin.Thiscom-
precipitation was continuously high between 25000 and binationoffactorsmayexplaintherelativelyhighlakelev-
16000 cal yr BP,comparedwith the subsequentlate gla- elsinnorthwestAmazonia(i.e.LakePata)comparedwith
cial and Holocene, maximum precipitation occurred low levels and/or lake desiccation in southwest (e.g.
immediately before (25000–22000 cal yr BP) and after Laguna Chaplin and Laguna Bella Vista) and southeast
(18000–17000 cal yr BP) the LGM. (e.g. Carajas) Amazonia. The difference in lake-level rec-
However, sedimentary hiatuses occurred between ords between lowland Amazonian Bolivia and the Alti-
2200014CyrBPand1300014CyrBP(15600calyrBP; plano is reproduced in sensitivity tests with an
Stuiver et al. 1998) at two lakes on the Serra dos Carajas atmosphericGCM(Genesis2.01;Hostetler&Mix1999)
(6°35(cid:3)S, 49°30(cid:3)W, Para State, east Amazonia) and usingforaminiferal-based SST reconstructions (Mix et al.
between38000 14CyrBP and 1100014C yrBP(13000 1999), which simulate a stronger Walker circulation (an
calyrBP)atLagunaBellaVista(13°37(cid:3)S,61°33(cid:3)W,nor- equatorial zonal pattern of ascending air over Indonesia,
theastBolivia,southwestAmazonia).Theselakes,located South America and Africa, and descending air over the
very close to the current rainforest limit (figure 1), there- eastern Pacific, Atlantic and western Indian oceans
fore dried up, indicating that precipitation at the LGM (Hostetler & Mix 1999)). It is likely that westerlies from
musthavebeenbelowpresent-daylevels(Absyetal.1991; the eastern Pacific or strengthened ‘surazos’ from Pata-
Mayle et al. 2000; Sifeddine et al. 2001). Supporting evi- gonia during the austral winter were the dominant mech-
dence comes from Laguna Chaplin (14°28(cid:3)S, 61°04(cid:3)W; anisms for moisture transport to the Altiplano at the
Mayleet al.2000; Burbridge etal.2004), located 100km LGM, rather than easterlies from the Amazon Basin.
south of Laguna Bella Vista. The Isoetes record from this
site shows that water levels were lower than today at the (b) Vegetation responses inferred from palaeodata
LGM, but precipitation cannot have been drastically (i) Forest versus savannah
lower because the sediment stratigraphy and age–depth Accumulating palaeoecological evidence indicates that
plot (Burbridge et al. 2004) show that the lake, which is mostoftheAmazonBasinremainedforestedattheLGM
currently only 2m deep in the dry season, has not dried (figure 1), contrary to the rainforest refugia hypothesis of
up for any significant period (i.e. several millennia) over Haffer (1969). Colinvaux et al. (1996) and Bush et al.
the past 50000years. (2002) provide pollen data to show that the Lake Pata
This LGM scenario of the southern Amazon lowlands catchment in northwest Amazonia was forested at this
beingcooleranddrierthantoday,andtheAltiplanobeing time,and that,furthermore, ithasbeen continuously for-
cooler and wetter than today might, at first glance, seem ested over the past 170000years. Soil stable carbon iso-
difficulttoexplain,giventhattheAltiplanotodayreceives tope analyses by De Freitas et al. (2001) show that
the bulk of its precipitation in the austral summer from savannah islands within southern Amazonian rainforests
the lowland Amazon Basin to the northeast. However, (between Porto Velho, Rondonia State and Humaita,
analysis oflocal meteorological data, incombination with Amazonas State, Brazil, 8°43(cid:3)S, 63°58(cid:3)W–7°38(cid:3)S,
a GCM, has led Garreaud et al. (2003) to infer that the 63°04(cid:3)W) were no more extensive at the LGM than
critical factor determining precipitation variability on the today.
Altiplanois notmoisture changein the sourcearea perse, However, in contrast to these sites in northwest and
butinsteadtheintensityanddirectionofupper-levellarge- south Amazonia, records from sites closer to the margins
scalezonalmoisturetransport(i.e.easterlywindsenabling of the Basin do reveal changes in forest–savannah distri-
moisture transport from Amazonia versus dry westerly butionbetweentheLGMandtoday.The50000yearpol-
winds from the Pacific). Orbital forcing by the 19000– len record from Laguna Chaplin (Mayle et al. 2000;
22000year precession cycle of the Milankovitch Astro- Burbridge et al. 2004) shows that Amazonian rainforest
nomic Theory would be predicted to have maximized communities at the LGM were located at least 30 km
austral summer insolation at 10–15°S at the LGM, ther- north of their current southern limit in eastern Bolivia
eby increasing deep convection over central South Amer- (figure1),andthatthisecotonalareawasthendominated
ica, which would in turn maximize seasonal southerly by open savannahs, with rainforest and/or dry forest spe-
migration of the ITCZ and intensify the Bolivian High (a cies probably restricted to riverine gallery forests. Pollen
prominent upper-level anticyclonic circulation centred at evidence from Laguna El Pinal (4°08(cid:3)N, 70°23(cid:3)W), in
roughly 15°S, 65°W; Lenters & Cook 1997). These the Colombian savannahs of the Llanos Orientales at the
phenomena would not only increase precipitation in the oppositeendoftheBasin(Behling&Hooghiemstra1999;
southern half of Amazonia (which lies in the Southern figure1),shows thatthissite,whichiscurrentlybordered
Hemisphere), but also increase the intensity of moisture by localized gallery forests, was then surrounded by vir-
transport to the Altiplano by strengthening the easterly tually treeless savannah. Mourguiart & Ledru (2003)
winds. However, the expanded Antarctic ice sheet and present pollen data from a site, near the village of Siberia
Patagonian glaciers at the LGM would be expected to (17°50(cid:3)0(cid:4)S, 64°43(cid:3)08(cid:4)W) at the upper cloud forest eco-
havesteepenedthelatitudinaltemperaturegradientacross tone in the Bolivian Andes, which shows an open grass-
South America, despite the 5°C cooling in the Amazon dominated landscape at the LGM, interpreted by the
Basin, thereby counteracting the effects of precessional authors as a vegetation response to increased aridity.
orbitalforcingandreducingwetseasonconvectiveactivity However, such an interpretation is at odds with the
over the southern Amazon Basin, thus producing greater palaeoprecipitationrecordfromtheAltiplano(Bakeretal.
precipitation in the northern half of the Basin compared 2001a,b), and a far more likely explanation, argued by
with the south, relative to today. Furthermore, reduced Baker et al. (2004), is that the grass-dominated LGM
Phil.Trans.R.Soc.Lond.B(2004)
Amazonian vegetation changes since the LGM F. E. Mayle and others 503
pollenassemblagesreflectexpansionofhighAndeanpuna 320tons Cha(cid:1)1 (ha, hectare, =104m2) for tropical rain-
vegetation and lowering of the tree-line, primarily in forest versus 260 tons Cha(cid:1)1 for seasonally dry forest
response to climatic cooling. (Adams&Faure1998).Consequently,significantchanges
Although it would be unwarranted to extrapolate these intheirrelativedistributionsandbiomassovertimewould
isolatedpollenrecordstotheentireAmazonBasin,pollen beexpectedtohaveresultedincorrespondingshiftsinthe
datafromAmazonFansediments(5°12.7(cid:3)N,47°1.8(cid:3)W) magnitudeoftheAmazoncarbonstore,whichin turnare
canbe considered a more reliableindicatorof Basin-wide likelyto havehadimplications forthe globalcarboncycle
changes in vegetation, because these pollen assemblages (Mayle & Beerling, 2004). Second, determining their
have been deposited from the entire Amazon river catch- respective biogeographic distributions at the LGM pro-
ment. Pollen spectra from these Amazon Fan cores show videsinsights intounderstanding theimportanceofcyclic
nosignificant changes in therelative proportions offorest Quaternary climatic changes in explaining their present
versus savannah pollen taxa between the LGM and the floristic composition, biodiversity and structure. Previous
Holocene(Haberle1997;Hoorn1997;Haberle&Maslin debate has dwelt on the relative distribution of rainforest
1999). These findings corroborate the isolated terrestrial versus savannah at the LGM (rainforest refugia hypoth-
pollenrecords,showingthatalthoughthereisevidencefor esis; Haffer 1969), and yet the possibility of dry forest
more widespread savannahs at the northern and southern expansion within Amazonia at this time has largely been
Amazonian margins relative to today, most of the Basin ignored,aspointedoutbyPenningtonetal.(2000),which
remained forested at the LGM. is surprising given that both dry forests and cerrado sav-
annahs grow under similar climatic conditions (generally
(ii) Seasonally dry forests versus rainforests 700–1800mmyr(cid:1)1; Furley 1999).
Evidence from a variety of sources indicates that these
environmental changes during the LGM had significant (c) Semi-deciduous dry forest versus evergreen
impacts upon the floristic composition and structure of rainforest modern pollen rain data
theforestcommunities.Thefossilpollendata(albeitfrom Characterizing current rainforest and dry forest com-
very few sites) indicate that the species composition of munities by their surface pollen spectra is crucial to
theseglacial-ageforestshasnomodernanalogue.Thecli- determining whether or not these different vegetation
matecoolingcausedtaxathatarecurrentlylargelyrestric- types can be reliably distinguished by their pollen signa-
ted to the Andes (e.g. Podocarpus and Alnus) to spread tures, and hence whether or not their Amazonian distri-
throughouttheAmazonBasinandformmixedcommuni- butions at the LGM can be elucidated by fossil pollen
ties with lowland Amazonian forest taxa (Colinvaux et analysis. Although the available pollen data show that
al. 2000). LGMplantcommunitieswerewithoutmodernanalogues,
Notwithstanding these Andean elements, it has gener- owing to their unique mixture of Andean and lowland
allybeenassumed(Colinvauxetal.1996)thatthesecom- tropical forest taxa, modern pollen rain studies should at
munities were dominated by evergreen rainforest taxa. least give the potential to determine whether or not such
However,recentbiogeographicalresearchraisesthepossi- communities were dominated by evergreen species or
bility that much of the Amazon Basin was instead occu- instead deciduous/semi-deciduous species.
piedbyseasonallydrytropicalforests(i.e.semi-deciduous Severalmodernpollenrainstudieshavebeenpublished
and/or deciduous forests) during the last glacial period. fromevergreenrainforeststhroughouttheNeotropics,e.g.
Seasonallydryforests todayexistas isolated,disjunctdis- Central America (Bush 2000), the tropical Andes (Weng
tributions,widelyseparatedfromeachother;forexample, etal.2004),Amazonia(Bush1991;Bushetal.2001)and
the Caatinga forests of eastern Brazil, the Chiquitano dry theAtlanticcoastofBrazil(Behlingetal.1997).However,
forestofeasternBoliviaandtheAndeanpiedmontdryfor- the onlydata from semi-deciduous dryforests come from
ests of southern Bolivia/northwest Argentina (Dinerstein Panama (Bush 1991). Here, we present the first modern
et al. 1995). Prado &Gibbs (1993) and Pennington et al. pollen rain data from the Chiquitano semi-deciduous dry
(2000, 2004) found that these widely disjunct dry forest forestsoflowlandeasternBolivia,togetherwithadditional
regions share many widespread species and inferred that rainforest pollen rain data from the Bolivian Amazon
they were therefore ancient refugia of a formerly much (figure 2). The samples have been collected from both
more extensive, contiguous distribution during the last artificial pollen traps (Gosling et al. 2003) and sediment–
glacialperiod,whichtheytermedthe‘PleistoceneDryForest water interface lake samples. It is immediately apparent
Arc’.Pennington etal. (2000) argued thatfull glacialpol- that pollen percentages of the predominantly rainforest
len spectra from Amazonia, previously interpreted as a family,Moraceae,aresignificantlyhigherinevergreenfor-
rainforest signal (albeit with Andean elements), could est(40–60%)thaninsemi-deciduous/deciduousforest(7–
equally well be interpreted as a predominantly dry forest 19%). Although such high Moraceae pollen percentages
signalowing to thefact thatmostdryforest familiesarea are a consistent, reliable signature of the seasonal humid
subsetofrainforestfamiliesandbecausemostpollentypes evergreenforests of northeast Bolivia (Mayle et al. 2000),
cannot be identified to species level, thus frustrating comparisonwithotherpollenraindatashowsfarlesscon-
attemptstodistinguishtheseecosystems(atleastinterms sistency in abundance of this taxon in rainforests from
of diagnostic species). other parts of the Neotropics (e.g. 10–50%, Bush et al.
Determiningthechanges inspatial extentofdryforests 2001).Clearly,abundanceofthisfamily,byitself,cannot
versus rainforests in response to changing climate and serve as a reliable ecosystem indicator when considering
atmosphericCO concentrationssincethelastglacialper- an area as large as the Amazon Basin. Abundance of
2
iod is important for several reasons. First, these two Poaceae pollen is also unreliable as a means of dis-
ecosystems have different carbon storage values; i.e. ca. tinguishing between these ecosystems in the fossil record,
Phil.Trans.R.Soc.Lond.B(2004)
504 F. E. Mayle and others Amazonian vegetation changes since the LGM
(a) (b) (c) these species cannot be distinguished palynologically, we
Laguna La Gaiba are confident that most pollen grains in these samples
belongtoA.colubrina,becausethisisbyfarthemorecom-
Laguna Mandioré monofthetwospeciesinneotropicaldryforests.Prado&
Gibbs(1993)foundthatA.colubrinawaseitherdominant
Laguna Socórros or frequent in all areas of South American seasonally dry
tropicalforests,withtheexceptionoftheCaribbeancoasts
Acuario vegetation plot ofColombiaandVenezuela.Theyusedthecontemporary
distributionofthistaxonasakeybasisfortheirargument
Laguna Chaplin thatthesepatchesaredisjunctdistributionsthatmusthave
beenformerlyconnected.Thistaxonreaches19%inarti-
Laguna Bella Vista
ficial traps and 2–10% in lake surface sediments of the
Chiquitano Dry Forest (figure 2). It is completely absent
Los Fierros vegetation plot
from any neotropical rainforest modern pollen rain spec-
0 20 40 0 20 40 0 20 tra, with the exception of Laguna Bella Vista and Laguna
terrestrial pollen sum (%) Chaplin, which are situated only a few kilometres north
of the dry forest ecotone, and are therefore likely to con-
Figure2. Modern pollen rain percentages of (a) Moraceae– tain low-level occurrences of a few dry forest taxa, which
Urticaceae, (b) Poaceae and (c) Anadenanthera pollen from
no doubt account for the negligible percentages (1% and
semi-deciduous dry forest and humid evergreen forest sites
2%, respectively) of Anadenanthera pollen in the surface
in lowland Bolivia. The Acuario (15°14(cid:3)58(cid:4)S, 61°14(cid:3)42(cid:4)W)
spectra of these two sites. Given that most seasonally dry
and Los Fierros (14°34(cid:3)50(cid:4)S, 66°49(cid:3)48(cid:4)W) sites are
20m×500m permanent vegetation study plots. Pollen data forestsaresubjecttofrequentfires,whereasrainforestsare
not, a further distinguishing character between these two
from artificial pollen traps from these two plots (Gosling et
al. 2003) constitute mean values of counts from 10 evenly types of forest is the presence of charcoal within all these
spaced traps from each plot sampled over three consecutive dryforestsurfacespectra,butabsenceofcharcoalinrain-
years. Pollen samples from the five lakes were collected from forest spectra. It should be noted, however, that certain
the uppermost centimetre of sediment at the sediment–water kindsofneotropicaldryforestecosystem,suchasthexeric
interface. See Mayle et al. (2000) and Burbridge et al. caatingas of eastern Brazil, do not experience frequent
(2004) for more detailed pollen data of the surface samples fires, borne out by their sizeable populations of cacti,
from Laguna Chaplin and Laguna Bella Vista.
which are poorly adapted to fire.
The LGM pollen spectra of the Lake Pata record con-
ascogentlyarguedbyBush(2002a).Thisislargelyowing tain no Anadenanthera pollen and no charcoal, suggesting
to theunknowncontribution ofshore-lineaquaticgrasses eitherthat semi-deciduous–deciduous forests were absent
tothepollenassemblage(terrestrialandaquaticgrasstaxa fromthevicinity,orthatifsuchforesttypeswerepresent,
cannot be distinguished palynologically). Although low theycomprisedcommunitiesoftaxathatwerequitediffer-
Poaceae pollen percentages (i.e. less than 5–10%) are ent floristically and/or structurally from those of today.
indeed generally characteristic of most rainforest surface
pollenspectra,ourartificialpollentrapsfromtheChiquit-
anosemi-deciduousdryforests(figure2)yieldequallylow (d) Vegetation simulations
Poaceae percentages. Interestingly, the surface pollen (i) Application of the University of Sheffield Dynamic
samples from the semi-deciduous dry forest sites Laguna Global Vegetation Model
LaGaiba,LagunaMandiore´ andLagunaSoco´rros(figure Thepoorspatialresolutionofthefossilpollendatasev-
2)allcontaininexcessof30%grasspollen,althoughsuch erelyhamperseffortstomapdistributionalchangesindif-
high values may at least partly be attributed to localized ferentecosystemsacrossthevastAmazonBasinovertime.
patchesof open cerradosavannahon neighbouringridges Dynamic vegetation models can therefore serve as useful
(L. Mandiore´) or nearby patches of savannah marsh toolsfor making process-based predictions about the nat-
and/or small local clearances (L. Soco´rros and L. La ureofvegetationresponsestopastatmosphericCO levels
2
Gaiba). The large surface areas of these lakes (e.g. L. and climates across Amazonia. Here, we use data
Mandiore´, 18km×8km) would suggest that the relative extracted from global equilibrium simulations using the
contribution from shore-line aquatic grasses to the grass SDGVM (Woodward et al. 1995; Beerling 1999; Beer-
pollen sumwas minimal. Similarly, highgrass pollen per- ling & Woodward 2001). The SDGVM calculates veg-
centageshavebeenrecordedfromopensavannahs(Ledru etationpropertiesundersteady-stateconditionsofclimate
2002;W.D.Gosling,unpublisheddata),furtherunderlin- and CO , and represents the physiological processes of
2
ing the poor diagnostic value of Poaceae pollen abun- plant nutrient uptake, C and C photosynthesis,
3 4
dance. respiration and stomatal control of canopy transpiration.
Incontrasttotheotherpollentaxainthesesurfacepol- Above-ground primary productivity is fully coupled to a
len spectra, Anadenanthera pollen emerges as a key indi- below-groundmodelofsoilcarbonandnitrogendynamics
cator taxon of semi-deciduous–deciduous tropical forest. so that plant litter (leaves and surface roots) is subsumed
Although Anadenanthera colubrina is restricted to season- throughdecomposition–nutrientcycling(Woodwardetal.
ally dry forests (Prado & Gibbs 1993), the other species 1998; Beerling & Woodward 2001). The model predicts
in this mimosoid genus, Anadenanthera peregrina, is also the distribution of different plant functional types on the
common in the southern cerrado savannahs of Sao Paulo basis of annual net primary production and biomass,
state, Brazil (J. A. Ratter, unpublished data). Although competition for light and other resources, as well as
Phil.Trans.R.Soc.Lond.B(2004)
Amazonian vegetation changes since the LGM F. E. Mayle and others 505
(a) (d)
(b) (e)
(c) (f)
t C ha–1 yr–1
BAR c3 c4 ebl enl dbl dnl 0 1.5 3.0 4.5 6.0 7.5 9.0 10.5 12.0
Figure3. SDGVM model simulations forced with the UGAMP GCM. See §2d(i). Key: c3, C grasses; c4, C grasses; ebl,
3 4
evergreen broad-leaf forest; enl, evergreen needle-leaf forest; dbl, deciduous broad-leaf forest; dnl, deciduous needle-leaf forest.
Amazonian vegetation distribution: (a) Pre-Industrial; (b) Mid-Holocene; (c) LGM; NPP of Amazonian vegetation: (d) Pre-
Industrial; (e) Mid-Holocene; (f) LGM. (tCha(cid:1)1yr(cid:1)1, tonnes of carbon per hectare per year, where 1 hectare is 104m2.)
probability of disturbance, and succession after disturb- discussed in the context of Amazon carbon cycling else-
ance (Cramer et al. 2001). where (Mayle & Beerling 2004).
In this analysis, we focused on three time intervals for TheUGAMPGCMclimateresponsetotheLGMtem-
comparison: the Pre-Industrial, the Mid-Holocene and perature(5°Cbelow present) and atmosphericCO con-
2
theLGM.ThePre-Industrialsimulationsofpotentialveg- ditions (180p.p.m.) is a basin-wide MAP of 1660mm,
etation in the Amazon Basin were made by forcing the which is 20% (338mm) below present. The GCM com-
SDGVM with the historical land-surface climatologies putesmeanmonthlyprecipitationvalues(i.e.precipitation
produced by New et al. (1999, 2000) and averaging the seasonality as well as MAP), which are used by the
resultsfortheperiodAD1901–1910.Thetwopalaeoveg- SDGVM.
etation simulations were forced by using the UGAMP
GCM-derived climate (Hall et al.1996a,b; Hall &Valdes (ii) Changes in biome distribution
1997),whichhasaspatialresolutionof2.8°latitude×2.8° The first key finding from these model simulations is
longitude.The UGAMPGCM is basedon theEuropean that the Amazon Basin remained predominantly forested
Centre for Medium-Range Weather Forecasting model attheLGM(figure3c),inagreementwiththepollendata
(ECMWF) and uses SSTs prescribed from palaeodata. discussed in §2b(i). The second significant result is that
Model bias was minimized by calculatinglocally differing evergreenbroad-leafforestsaresimulatedtohavecovered
monthly climate anomalies, i.e. GCM control (present- the northern half of Amazonia, while deciduous broad-
day)runminusGCMpalaeorun,andimposingtheseonto leaf forests covered the southern half. These simulations
a modern underlying climatology (cf. Beerling 1999). suggestthatthereducedprecipitationinthesouthernhalf
Atmospheric CO concentrations for the Pre-Industrial, of Amazonia was sufficient to result in competitive
2
Mid-Holocene and LGM were 300, 280 and replacement of evergreen species by the drought-adapted
180p.p.m., respectively. semi-deciduousanddeciduousspeciesfoundinseasonally
Figure3showsthepredicteddistributionofplantfunc- dry tropical forests. They also support the hypothesis of
tional types and vegetation productivity in the Amazon Pennington et al. (2000), based on their phytogeographic
Basin for the (a) present day (Pre-Industrial), (b) Mid- studies,thatmuchoftheBasinwascoveredbyseasonally
Holocene (6000 cal yr BP) and (c) LGM (21000 cal yr dry forest rather than rainforest.
BP). The Pre-Industrial simulation is for the potential However, it should be noted that, despite a 20%
vegetation, and does not account for anthropogenic land- decrease in precipitation in northwest Amazonia, our
cover change (e.g. crops, urban areas, forest clearance). model shows that precipitation in this area (which
Changes in land surface carbon storage of the different includes Lake Pata) remained high enough to maintain
ecosystems in Amazonia for these time periods are evergreen broadleaf forest at the LGM, which supports
Phil.Trans.R.Soc.Lond.B(2004)
506 F. E. Mayle and others Amazonian vegetation changes since the LGM
the inferences by Colinvaux et al. (1996, 2000) and our- stressduetoloweredprecipitation.Reducedprecipitation
selves,andcontradictsthesuggestionbyPenningtonetal. (e.g.20%)wouldnodoubthave beensufficienttotip the
(2000)thatitwasregionallysurroundedbyseasonallydry ecologicalbalance infavourofsavannahs attherainforest
forest.Furthermore,ourLGMmodelsimulationdoesnot ecotonesatthenorthernandsouthernAmazonBasinmar-
supportthe‘PleistoceneDryForestArc’hypothesis. This gins, where current rainfall approaches the precipitation
hypothesis predicts that the Chiquitano Dry Forest of thresholdofca.1500–1800mmyr(cid:1)1forevergreenrainfor-
easternBoliviaisaremnantofthecentralcoreofthispos- est. However, the cooler temperatures would have served
tulated expanded Pleistocene distribution. Instead, the to reduce evapotranspiration and thereby prevent more
model simulates C savannah grasses for eastern Bolivia, widespread replacement of forest by savannah in more
4
and shows thatdry forest taxa were located northof their central parts of the Basin.
present-day disjunct distributions and were restricted to Our modelsimulationsofNPP corroboratethe simula-
the southern half of Amazonia, which is currently occu- tions of Cowling et al. (2001) of reduced LAI. Figure 3f
piedbyrainforest.Thisvegetationsimulationiscorrobor- shows that the cool, low CO environment of the LGM
2
atedby the pollen recordfrom Laguna Chaplin.This site severely restricted vegetation carbon uptake and lowered
islocatedatthesouthernmostlimitofAmazonrainforest, NPPcomparedwiththeHolocene (figure3d,e). Irrespec-
only30kmnorthoftheChiquitanoDryForest(figure1), tiveofthecompositionorstructure offorestcommunities
and was predominantlysurrounded by opensavannahs at that occupied the Amazon Basin at this time, it is clear
theLGM,withdeciduous–semi-deciduousforesttaxaper- thatbothmodellingstudiesandpalaeodatarefutethe‘gla-
haps restricted to isolated edaphically suitable areas (e.g. cial refugia hypothesis’, advocated by Haffer (1969),
riverinegalleryforests).Anadenantheradidnotarriveuntil Prance (1982), Whitmore & Prance (1987) and Haffer &
the Early Holocene (Burbridge et al. 2004). These fossil Prance (2001), which states that most of the Amazon
pollen data (albeit from only one site), together with the BasinwascoveredbysavannahattheLGM.Furthermore,
model simulation, suggest that rather than being relict, these studies suggest that, although seasonally dry forests
refugial populations, the current distributions of dry for- appear to have been more extensive within Amazonia at
ests are instead only Holocene in age, having arisen from the LGM than today, their overall spatial coverage
populations spreading southwards from Amazonia since throughout tropical South America does not appear to
the LGM. have been significantly greater at the LGM than today
(beforehumanclearance),but simplyhadadifferentspa-
(iii) Changes to forest structure and composition tial distribution in response to the differing climatic and
Vegetation modellingby Cowling et al.(2001) suggests CO conditions at that time.
2
that changes in vegetation structure could have been at It is important to stress that although much of Ama-
leastasimportantaschangesinvegetationbiomesorplant zoniawasprobablycoveredbytropicalevergreenforestat
functionaltype.Todistinguishbetweenspecies’responses the LGM, the structure, productivity and composition of
to lowered temperature, atmospheric CO concentrations theforestsdifferedsubstantiallyfromthoseofcontempor-
2
and precipitation, Cowling et al. (2001) simulated veg- ary Amazonian rainforests. Insights into the nature of
etation responses at the LGM to differing climate and these ice-age forests may be gained by considering the
atmospheric CO conditions. They undertook multivari- structure and composition of the moist evergreen
2
ablesensitivityexperimentstoshowthatbothconservative rainforests of southwest Amazonia (southern Peru and
(2°C)aswellasextreme(6°C)LGMcoolingconditions, northern Bolivia) near the ecotone with savannahs and
associated with both conservative and extreme values of semi-deciduousdryforests.Thelatter,incontrasttoother
decreased precipitation (20% versus 60% reduction) and forests in more central parts of the Amazon, are domi-
CO (220p.p.m. versus 180p.p.m.) resulted in signifi- natedto a large extent by evergreen lianas (Killeen 1998)
2
cantlyhigherpercentagesofforestcover thanthosesimu- and/or bamboo, which have a greater ability to withstand
lated under present-day temperature conditions. Cowling waterstressthanotherAmazonianrainforesttaxainthese
et al. (2001) attributed this response to improved carbon highly seasonal climates. It is therefore conceivable that
and water balance owing to reduced evapotranspiration theseforestsconstitutetheclosestmodernanaloguetothe
and lowered rates of photorespiratory carbon loss, both evergreen forests that existed in Amazonia at the LGM.
of which would have favoured dominance of forests over Furthermore, the low canopy density of these forests, in
savannahs. These authors further showed that atmos- comparison with those growing in wetter, more central
pheric CO concentrations would have been more parts of Amazonia, is consistent with the low NPP values
2
important than precipitation levels in controlling canopy simulatedinfigure3fandthelowLAIsimulatedbyCow-
density (i.e. vegetation structure). For a decrease in 20% lingetal.(2001).Suchforestswouldnothavebeenappar-
precipitation alone, they simulated an 11% decrease in ent from the fossil pollen records owing to the low pollen
basin-average LAI (an index of canopy density), whereas productivityofbamboo(Bush2002a),thefactthatdiffer-
lowered atmospheric CO concentrations alone (set at entgrasstaxacannot bedistinguishedby theirpollen and
2
LGM values of 200p.p.m.) were simulated to cause a because liana taxa are also generally indistinguishable
34% reduction in LAI. palynologically from evergreen tree taxa.
From these sensitivity simulationswe can infer thatthe
changes in vegetation structure (i.e. reduced canopy (iv) Vegetation–precipitation feedbacks
density) and forest type (i.e. replacement of evergreen by DeterminingtheresponsesofAmazonianecosystemsto
deciduous–semi-deciduous species) were predominantly climate forcing is further complicated by the fact that the
responsestocarbonlimitationandwaterstressduetolow- climate (especially precipitation) is itself determined, to a
ered atmospheric CO concentrations rather than water large degree, by the vegetation. For example, ca. 50% of
2
Phil.Trans.R.Soc.Lond.B(2004)
Amazonian vegetation changes since the LGM F. E. Mayle and others 507
Amazon precipitation is recycled through rainforest can- 3. AMAZONIA DURING THE LATE-GLACIAL
opy transpiration (Shukla & Mintz 1982), thus reducing AND HOLOCENE
the relative importance of moisture transport from the
Atlantic Ocean on total precipitation. Nepstad et al. (a) Climate and carbon dioxide concentrations
(1994) have shown that this precipitation recycling pro- The ice-core record from DomeConcordia (Dome C),
cess is in large part causedby the ability of tropical forest Antarctica (Monnin et al. 2001) shows that atmospheric
treestoaccessdeepsoilmoisturewithrootsinsomecases CO concentrations rose by 76p.p.m. across the last gla-
2
asdeepas18m.Fieldstudieshaveshownthatthisability cial–Holocene transition, from a mean of 189p.p.m.v.
to draw upon large stores of water allows such forests to between 18100 and 17000 cal yr BP to a mean of
continue transpiration and maintain their foliage during 265p.p.m.v. between 11100 and 10500 cal yr BP, and
extended dry periods. Kleidon & Lorenz (2001) included that this rise took place in four incremental steps. How-
rooting depth as an additional parameterin climate mod- ever, the pattern of CO changes across the Younger
2
elsandtherebydeterminedtheimpactofdeep-rootedveg- Dryas chronozone (GS-1, 12900–11500 cal yr BP;
etation upon climate simulations for Amazonia at the Bjo¨rck et al. 1998) determined from the stomatal charac-
LGM. They found that access to large soil-water stores tersoffossilleaves remainscontroversial (McElwainetal.
by deep-rooted vegetation caused greater evapotranspir- 2002). The Antarctic Taylor Dome ice core (Indermuhle
ation in the dry season, which in turn produced a 4°C etal.1999)showsthatCO concentrationsrosegradually
2
surface cooling (owing to latent heat of evaporation) and from 260p.p.m.v. to 285p.p.m.v. through the Holocene
an increase in precipitation of up to 2mmday(cid:1)1. These until the Industrial Period.
results show that despite an anticipated reduction in pre- Multi-proxydata fromseveralsites(figure1)showthat
cipitation coming from a cooler tropical Atlantic Ocean, the temperature and precipitation changes in Amazonia
precipitationrecyclingbyvegetationwouldhavebeensuf- through the last glacial–Holocene transition were com-
ficient to maintain humid conditions, and consequently plex. Oxygen isotope, ice-accumulation and aerosol con-
evergreen forest, throughout most of the Amazon Basin. centrations from Sajama Mountain (Thompson et al.
Althoughtheseauthorsneglectedtoincludetheadditional 1998), together with geochemical, diatom, sedimentolog-
water stress expected from lowered LGM CO concen- ical and modelling studies from Lake Titicaca (Baker et
2
trations, which could have tipped the ecological balance al.2001a;Crossetal.2001),providehigh-resolutionesti-
fromevergreentosemi-deciduousdryforest,itisnonethe- mates of temperature and precipitation changes on the
less clear that the vegetation itself would have counter- Bolivian Altiplano, above the southwest rim of the Ama-
acted to a significant extent any climate aridity inferred zonBasin.Theserecordsshowthat,aftertheLGM,cold,
from independent climate data from outside the Amazon wetconditionspersisteduntil15500calyrBP,whentem-
Basin, e.g. equatorial Atlantic SSTs. peraturesandaridityrapidlyincreaseduntilca.14300cal
Thesestudieshighlight thefact thatthe seasonaldistri- yr BP. However, new sedimentological data from Lake
bution of precipitation (i.e. the length and severity of the Titicaca and Lake Junin, Peruvian Andes (ca. 11°S,
dry season) is the critical control upon vegetation, rather 76°W; Seltzer et al. 2002) show that the onset of this
thanmeanannualprecipitation.Forexample,thedecidu- deglacial warming trend may have begun as early as
ous monsoon forests of India receive up to 5000mm 22000–19500 cal yr BP, coincident with the generally
MAP,whereastheevergreenequatorialforestsofpeninsu- accepted date of the global LGM (i.e. 21000 cal yr BP)
lar Malaysia receive only 2000mm MAP (O. Phillips, inferredfromthemaximumextentoftheNorthAmerican
unpublished data). Sternberg (2001) developed a simple Laurentide ice sheet. This episode of increasing tempera-
dynamicmodelsimulatingthedynamicfeedbackbetween tures and aridity correlates with lower lake levels in the
dry-season precipitation and forest area. His simulations PeruvianAndes(Seltzeretal.2000)andcentralAmazonia
showed that a minimum of 100mm dry-season precipi- (Bush et al. 2002) and rising equatorial Atlantic SSTs
tation (total precipitation over the driest three months) (Guildersonetal.1994),indicativeofsimilarclimaticcon-
was required to maintain tropical evergreen forest, ditionsthroughouttheAmazonBasin.Subsequently,tem-
although only 50mm imported dry-season precipitation peraturesdecreasedandprecipitationincreasedoncemore
was required, given the ability of such forests to recycle on the Altiplano, culminating in a cold, wet climate
50% of the rainfall. between13000and11500calyrBP,correlativewiththe
In light of these studies, it is therefore important to NorthAtlanticYoungerDryasstadial(GS-1;Bjo¨rcketal.
point out that the palaeoprecipitation records gleaned 1998). However, this cold interval in lowland Amazonia
fromlake-levelreconstructionsprovideinformationabout appears to have been accompanied by the most arid con-
past changes in MAP but unfortunately cannot provide ditionsovertheentireperiodofstudy(i.e.thepast21000
information about the crucial parameter of dry-season years). At Lake Pata, northwest Amazonia, lowest water
precipitation.Furthermore,althoughourSDGVMsimul- levelssincethe LGM occurredat 12000 cal yrBP(Bush
ations are based on precipitation seasonality, they do not etal.2002),coincidentwithapeakindunegrowthnearby
incorporate any vegetation-climate feedbacks. Conse- (00°23(cid:3)S, 64°33(cid:3)W; Filho et al. 2002) and a major
quently, our simulations may underestimate the actual reductioninAmazonriverdischarge(60%belowpresent;
spatial cover and productivity of the rainforests that Maslin&Burns2000).Furthermore,thelake-levelrecord
existed at the LGM. Clearly, understanding and quan- from Lake Junin provides evidence for a peak in aridity
tifyingthesefeedbacksremainsanimportantgoalforveg- during the Younger Dryas chronozone in the central
etation and climate modellers, although progress is being Peruvian Andes above the western rim of the Amazon
made (e.g. the TRIFFID terrestrial carbon cycle model Basin (Seltzer et al. 2000). These regional comparisons
(Cox et al. 2000; Cox 2001; Cowling et al. 2004)). suggestthat,incontrasttotheprecedinginterstadialwarm
Phil.Trans.R.Soc.Lond.B(2004)
508 F. E. Mayle and others Amazonian vegetation changes since the LGM
periodandthepresent-dayclimate,moisture-bearingeast- and Chaplin (Burbridge et al. 2004) all show higher
erliesfromtheAmazonBasinwereconsiderablyweakened precipitation in the Late Holocene than Early–Mid-
duringtheYoungerDryas, withprecipitationinsteadper- Holocene. This trend of increasing precipitation can be
haps predominantly originating from strengthened west- attributed to progressively greater seasonal southerly pen-
erlies and/or anti-cyclones from the Pacific, as simulated etration of the ITCZ, owing to increasing summer inso-
by Hostetler & Mix (1999). lation at 10°S, which can in turn be attributed to the
Proxy-temperature data from all these sites show that precessioncycleoforbitalMilankovitchforcing(Berger&
the termination of the Younger Dryas event was marked Loutre 1991). The increased southerly migration of the
by a very rapid, large increase in temperatures at 11500 ITCZwouldbeexpectedtohavenotonlyincreasedMAP,
cal yr BP to modern values, marking the onset of the but,cruciallywithrespecttothevegetation,shortenedthe
present Holocene interglacial period. In contrast to the duration of the dry season.
relatively stable Holocene climate of mid-high latitudes,
the South American tropics have experienced major fluc- (b) Vegetation responses inferred from palaeodata
tuationsin precipitation throughtheHolocene. Through- Over the last glacial–Holocene transition (ca. 21000–
out southern Amazonia and the Bolivian Altiplano 10000 calyrBP),forestsexpanded insouthernand east-
precipitationintheEarly–Mid-Holocene wassignificantly ernAmazonia.Cloudforestshadbecomefullyestablished
below that of the preceding late-glacial period and sub- on the eastern flanks of the Bolivian Andes between
sequentLateHolocene.Forexample,maximumaridityat 12500 and 10000 14CyrBP (14900 and 11500 cal yr
LakeTiticacaoverthepast25000yearsoccurredbetween BP; Mourguiart & Ledru 2003), most probably owing to
ca. 8500 and 3600 cal yr BP (Cross et al. 2000; Baker et increasingtemperatures(Bakeretal.2004)andCO con-
2
al. 2001a), when precipitation is modelled to have been centrations, coincident with forest expansion in Ron-
40% below modern values (Cross et al. 2001) and lake donia–Amazonas states (based on stable carbon isotope
levels were 90m below present (D’Agostino et al. 2002). data; De Freitas et al. 2001) and Para state (Absy et al.
Furthermore, Wolfe et al. (2001) inferred from oxygen 1991). Populations of Podocarpus increased throughout
isotope data from other Bolivian lakes that summer rela- Amazonia through the late glacial, suggesting that the
tivehumidityvaluesontheAltiplanowere20%lowerthan combination of increasing CO levels and precipitation,
2
present 7500–6000 cal yr BP. There is widespread char- together with sufficiently cool temperatures, had reduced
coal evidence to indicate that this reduction in precipi- water stress sufficiently to allow expansion of this pre-
tationcaused anincrease infires in the BolivianAmazon, dominantly Andeangenus throughoutthe entireBasin. It
i.e. the eastern flank of the Andes (Mourguiart & Ledru should be noted, though, that there is considerable tem-
2003), the vicinity of Santa Cruz de la Sierra just east of poral variation in the occurrence of this Podocarpus peak
the Andes (ca. 17°55(cid:3)S, 63°15(cid:3)W; Servant et al. 1981), across the Amazon Basin (Mayle et al. 2000; Behling
and the northeast Bolivian lowlands (Burbridge et al. 2001; Ledru et al. 2001; Sifeddine et al. 2003; Burbridge
2004), as well as further afield in Para State (eastern et al.2004), mostprobably reflectingthe substantialtem-
Brazilian Amazonia; Soubies 1980; Turcq et al. 1998; poral and/or spatial complexity of precipitation and tem-
Sifeddine et al. 2001). Although this reduction in rainfall perature patterns in Amazonia over this interval. The
appears not to have been great enough to cause extensive temporal resolution of most of these pollen studies is
fires inthe wetter, central and western parts ofthe Basin, insufficient to draw precisecorrelations withthe high-fre-
pollen data do suggest more frequent low-water stands quencyclimaticoscillationsthatcharacterizedthisperiod.
nearManaus(3°16(cid:3)S,60°35(cid:3)W;Behlingetal.2001)and Humid evergreen rainforest remained significantly less
increased drought frequency in lowland Amazonian extensivethantodayinNKMNP(Mayleetal.2000;Bur-
Ecuador (0°27(cid:3)S, 76°37(cid:3)W; Weng et al. 2002). Oxygen bridgeetal.2004),indicatingthatdry-seasonprecipitation
isotope records from the Peruvian Andes (Thompson et was still below present-day levels at this time, at least in
al. 1995) and Peruvian coastal waters (8°55(cid:3)S, 4°40(cid:3)S; the Bolivian Amazon.
Andrusetal.2002)suggestthatthisMid-Holoceneperiod The Early–Mid-Holocene episode of aridity had a sig-
of aridity coincided with a Holocene thermal maximum, nificant impact on plant communities throughout much
centred around 6000–6500 cal yr BP, with mean annual ofAmazonia.Stablecarbonisotopedatafromsoilorganic
SSTs of coastal Peru ca. 3–4°C higher than today. How- matter show expansion of savannah islands at the border
ever, the lack of any apparent temperature change in the of Amazonas and Rondonia states, Brazil (Pessenda et al.
Sajama oxygen isotope profile (Thompson et al. 1998) 1998; De Freitas et al. 2001). The increased fire fre-
calls into question the regional extent of such a warming quencies between 7000 and 3000 14C yr BP (7800 and
and whether or not it affected the Amazon lowlands. In 3200calyrBP)insouthernAmazoniawouldbeexpected
any event, any warming in lowland Amazonia would be to have caused significant structural and compositional
expectedtohavebeenoffsettosomeextentbythecooling changes to the forests, especially increases in drought-
effect of evapotranspiration from the rainforest. tolerant lianas and semi-deciduous tree species. For
There is evidence from all of these sites that precipi- example the peaks in the dry forest taxa A. colubrina,
tation increased in the latter half of the Holocene, Astroniumurundueva and Astronium fraxinifolium,together
especially after ca. 3000 cal yr BP.For example, Amazon with the savannah indicator Curatella americana and
riverdischarge(Maslin&Burns2000)andwaterlevelsat Poaceae pollen, in combination with high charcoal con-
Lake Junin (Seltzer et al. 2000), Lake Pata (Bush et al. centrations,fromLagunaChaplinandLagunaBellaVista
2002) and Lake Titicaca (Cross et al. 2000, 2001), (Burbridge et al. 2004) show that the ecotonal area of
togetherwithmarkeddecreasesincharcoalconcentrations NKMNP in southwest Amazonia, which is currently
at Lake Carajas (Turcq et al. 1998) and lakes Bella Vista dominatedbyevergreenrainforest,wasinsteadcoveredby
Phil.Trans.R.Soc.Lond.B(2004)