The roles of climate and human land-use in the late Holocene rainforest crisis of Central Africa

Contents lists available atScienceDirect

Earth

and

Planetary

Science

Letters

www.elsevier.com/locate/epsl

The

roles

of

climate

and

human

land-use

in

the

late

Holocene

rainforest

crisis

of

Central

Africa

Germain Bayon

a,b,

∗

,

Enno Schefuß

c

,

Lydie Dupont

c

,

Alberto

V. Borges

d

,

Bernard Dennielou

a

,

Thibault Lambert

e

,

Gesine Mollenhauer

f

,

Laurence Monin

b

,

Emmanuel Ponzevera

a

,

Charlotte Skonieczny

a,g

,

Luc André

b

a_{IFREMER,MarineGeosciencesUnit,F-29280Plouzané,France}

b_{RoyalMuseumforCentralAfrica,DepartmentofEarthSciences,B-3080Tervuren,Belgium} c_{MARUM–CenterforMarineEnvironmentalSciences,BremenUniversity,D-28359Bremen,Germany} d_{ChemicalOceanographyUnit,UniversitédeLiège,B-4000,Belgium}

e_{InstituteofEarthSurfaceDynamics,UniversityofLausanne,CH-1015,Switzerland} f_{AlfredWegenerInstituteforPolarandMarineResearch,D-27570Bremerhaven,Germany}

g_{LaboratoireGeosciencesParis-Sud,UMRCNRS8148,UniversitédeParis-Sud,UniversitéParis-Saclay,F-91405Orsay,France}

a

r

t

i

c

l

e

i

n

f

o

a

b

s

t

r

a

c

t

Articlehistory:

Received15June2018

Receivedinrevisedform12October2018 Accepted14October2018 Availableonlinexxxx Editor: D.Vance Keywords: neodymiumisotopes suspendedparticulates CongoBasin ENSO humanland-use deforestation

There is increasing evidence that abrupt vegetationshifts and large-scale erosive phases occurredin Central Africa during the third millennium beforepresent. Debate exists as to whetherthese events were caused by climatechange and/or intensifying human activitiesrelated to the Bantu expansion. In this study, we report ona multi-proxyinvestigation of asediment core (KZR-23) recoveredfrom the Congo submarine canyon. Our aim was to reconstruct climate, erosion and vegetation patterns in the Congo Basinfor the last10,000 yrs, with aparticular emphasison the late Holocene period. Samplesofmodernriverinesuspendedparticulateswerealsoanalyzed tocharacterizesedimentsource geochemical signatures from across the Congo watershed. We find that a sudden increase of bulk sediment aluminium-to-potassium (Al/K) ratios and initial radiocarbon ages of bulk organic matter occurredafter2,200 yrsago,coincidentwithapollen-inferredvegetationchangesuggestingforestretreat and developmentofsavannas.Althoughhydrogenisotopecompositionsofplantwaxes(δDwax)donot revealasubstantialhydroclimateshiftduringthisperiod,neodymiumisotopesandrareearthelements indetritalfractionsindicateprovenancechangesforthesedimentexportedfromtheCongoBasinatthat time,hencesuggestingareorganizationofspatialrainfallpatternsacrossCentralAfricaduringthisevent. Taken together,these findingsprovideevidence for changinglandscapes inCentralAfrica from about 2,200 yrsago,associatedwithsynchronouseventsofvegetationchangesandenhancederosionof pre-aged and highlyweatheredsoils.Theseeventscoincidedremarkably wellwiththe arrivalofIron Age communitiesintotherainforest,asinferredfromcomparisontoregionalarchaeologicalsyntheses.While thehumanimpactontheenvironmentremainsdifficulttoquantifyatthescaleofthevastCongoBasin, wetentativelyproposethatstrengtheningofElNiño-SouthernOscillation(ENSO)variabilityatthattime playedakeyroleintriggeringtheobservedenvironmentalchanges,andpossiblyactedasadriverfor theeastwardmigrationofBantu-speakingpeoplesacrossCentralAfrica.

©2018ElsevierB.V.Allrightsreserved.

1. Introduction

About5,000yrsago,earlyfarmersoriginatingfromWestAfrica startedtomigratesouthward,leadingwithinafewthousandyears to widespread diffusion of Bantu languages and cultural innova-tions across sub-Saharan Africa (Diamond and Bellwood, 2003;

*

Correspondingauthorat:IFREMER,MarineGeosciencesUnit,F-29280Plouzané, France.

E-mailaddress:gbayon@ifremer.fr(G. Bayon).

Phillipson, 2005). The Bantu migrations occurredata time when landscapeswereadjustingtoadrierclimate(Bostoenetal.,2015). Atca.4,000and2,500yrbeforepresent(BP),forestcoverdeclined significantly attheperipheryandcoreofthecentralAfrican rain-forest, respectively(Vincenset al.,1998; Maley,2002; Ngomanda etal.,2009; Bostoenetal.,2015; Maleyetal., 2018),probably fa-cilitatingthemigrationofBantu-speakingpeoplesthroughopening ofsavannacorridors(Grollemundetal.,2015).Abundantevidence for intense sediment reworking and strong erosion is recorded at thistime in soil sequences from thecentral African rainforest https://doi.org/10.1016/j.epsl.2018.10.016

(Thiéblemontetal.,2013,2014;Maleyetal.,2018).Itisgenerally assumedthat both vegetation shiftsand erosive phaseswere re-latedtoclimatechange(Lézineetal., 2013;Bostoenetal., 2015; Maley et al., 2018). However, debate exists as to whetherthese eventscould havepartly resulted fromthe development of agri-cultureandiron smelting technology amongst theBantu popula-tions (Willis et al., 2004; Bayon etal., 2012; Maley et al., 2012; Neumannetal.,2012).Usingvariousgeochemicalproxies,Bayonet al. (2012) identifiedapulseofintenseweatheringinlateHolocene sediments deposited at the Congo deep-sea fan, which they at-tributedto increasing land-use coincidingwiththe settlement of Iron Ageagriculturalists in Central Africa. More recently,a study conducted at Lake Barombi (Cameroon) also reported new evi-denceforanabruptdeforestationeventinWesternCentralAfrica about2,600yrsago,whichwasinterpreted,inabsenceofany par-ticularhydrologicalchange,astheconsequenceofhumanactivities (Garcinetal.,2018).Werehumansorclimate,oracombinationof bothfactors, responsibleforthe rainforest crisis andintense ero-sionalevents? This question remains open, mostly due to a lack of knowledge regarding the environmental history of the Congo Basinduring the last few thousand years, asexemplified by the recent discovery of a much larger than expected peatland com-plex of Holocene age in its central region (Dargie et al., 2017). RainfallpatternsintropicalAfricaarecomplexandsubjectto var-iousinfluencesassociated withlatitudinalvariations ofthe tropi-cal rainbeltandzonaleast–west gradients drivenby Atlanticand IndianOcean sea-surface temperatures(Singarayer andBurrough,

2015). To date, no coherent framework exists for the interpre-tationof late Holocene palaeoenvironmental records in this vast region.

Todetect pastnaturalclimatevariabilityandpotential anthro-pogenic influence in Central Africa, we analyzed sediment core KZR-23recoveredfromtheactive submarinecanyonoftheCongo River(Fig.1).TheCongosubmarinecanyonincisingintothemouth actsasadirectsource-to-sinkrouteforsedimentsfromtheCongo Riverwatershedto thedeep-sea(Babonneauetal.,2004). Conse-quently, sedimentation rates recorded forcore KZR-23 are much higher(about170cm/kyr)comparedtothoseofothernearby ma-rine sites (about 10 cm/kyr) investigated so far (Schefuß et al.,

2005; Bayonetal., 2012). Ittherefore provides acontinuous and uniquehigh-resolutionrecordofsedimentexportand environmen-tal changes in the Congo Basin for the Holocene period. In this study,we report findings froma combined applicationoftracers forchemical weathering intensity (aluminium-to-potassium), soil erosion(soil-derivedlipids andradiocarbonanalyses ofbulk sedi-mentorganicmatter),sediment provenance(neodymium isotopes andrareearthelements),precipitation (hydrogenisotope compo-sitionofleafwaxes),andvegetationchange(pollen).

2. Materialsandmethods

2.1.TheCongoBasin:climate,geologicalsettingandlandcoverpatterns TheCongoRiver BasinisthelargestAfricanriversystem, cov-eringanareaofabout3.8

·

106_km2_{thatrangesacrosstheequator}

fromabout9◦Nto14◦S.Its sourceisinthesouth-eastern partof the Democratic Republic of Congo; a region drained by the Up-perCongo (orLualaba)andpartlyfed bysome ofthe largelakes oftheEast AfricanRiftValley,such asLake TanganyikaandLake Kivu.The Congo River then flows northward andlater westward towards a large depression referred to as the ‘Cuvette Centrale’, coveredbyequatorialrainforestandfloodplains(Fig.1A).This de-pression accounts for almost 50% of the catchment area of the Congo River, hosts an extended peatland complex (Dargie et al.,

2017) and is a large source of greenhouse gases to the atmo-sphere (Borges et al., 2015). It is characterized by a low mean

Fig. 1. StudyareaandlocationofmarinesedimentcoreKZR-23.(A)Landcoverand vegetationpatternsintheCongoBasin(fromLambertetal.,2016).(B)Mapofthe CongoRiverBasinandmaintributaries,withpositionofthesuspendedriver par-ticulatesinvestigated duringearlier work(Allègreetal., 1996;squares)andthis study (TransCongosamples:circles).Therangeofneodymiumisotopic composi-tions(εNd)forsuspendedparticulatesisdenotedtopright.(C)Rainfallchartsfor

averagemonthlyprecipitations(mm)intheOubangui,KasaiandUpperCongo (Lu-alaba)sub-basinsoftheCongoRiverBasin.RainfalldatafromtheCLIMWATclimatic databasemanagedbytheFoodandAgricultureOrganizationoftheUnitedNations (FAO).

slope (<2 cm/km),low flow velocities(Laraqueetal., 2013), and isjoinedby theOubangui,thelargestright-banktributary, which drainsthenorthernpartoftheCongoBasin.Then,theCongoRiver movessouth-westward andisjoined by theKasai River, the

ma-Fig. 2. SchematicpositionofcoreKZR-23intheCongosubmarinecanyon.Modified fromBabonneauetal. (2004).

jor left-bank tributary, originatingfrom the high-plateaus in the southernpartofthewatershed.

Distinct physiographic regions compose the Congo Basin. The Cuvette Centrale is characterized by an equatorial climate,while the marginal regions of the Congo Basin all display tropical cli-matic conditions with pronounced seasonal rainfall variability (Fig. 1C). The northern partofthe basin (drainedby the Ouban-guiRiver)receivesmostrainfallfromMarchtoOctoberwhenthe rainbelt migratesto its northern position. Incontrast, the south-ernandeasternpartsofthewatershed(drainedby theKasaiand theLualabarivers)mainlyreceiveprecipitationfromSeptemberto Aprilwhentherainbeltmovessouth(Fig.1C).Intermsof geolog-icalsetting,thenorthernandeasternpartsoftheCongoBasinare dominated by Archaean and Proterozoic metamorphic basement rocks, while the southwestern part of the watershed is mostly composedofMesozoicandCenozoicsedimentaryrocks (Bayonet al.,2018).NeogenevolcanicrocksassociatedwiththeEastAfrican Riftsystemarealsoencounteredintheeasternpartofthebasin. 2.2. CoreKZR-23andriversuspendedparticulates

CoreKZR-23 (5◦40S,10◦38E;

∼

2220m waterdepth;

∼

20m long) was recovered during the ZAIROV cruise (2000; R/V L’Ata-lante),fromaterrace lyingat

∼

380mabove thethalweg,i.e.the deepest part of the Congo submarine canyon (Fig. 2). KZR-23 is mostly composedof homogeneous organic-richfine-grained sedi-ment,withminorcontributionsfromdiatomsandradiolaria.Most likely, the hightotal organic carbon content (2.7

±

0.2 wt%) re-sulted in active early diagenetic processes within the sediment, even aftercorerecoveryandsubsequent storage,which probably ledtopoorpreservationoftherareforaminiferaandotherbiogenic carbonatematerialsthat wereinitiallypresent. Asa consequence, itischaracterizedbyverylowbulksedimentCaOcontents(mean 0.5

±

0.2 wt%).Amajorturbiditesequencewasidentifiedatabout 17 mbsf(Babonneauetal.,2004),sothatonlytheupper

∼

16mof thecorewereinvestigatedduringthecourseofthisstudyinorder toavoidanalysisofanydisturbedsedimenthorizons.

AseriesofsuspendedriverparticulatesamplesfromtheCongo Riverwatershedwere alsoanalyzed duringthisstudy.These sam-ples were collected during the TransCongo expeditions (PI: A.V. Borges; Fig. 1B). TransCongo I (December 2013) consisted in a 1700-km transect across the river basin from Kisangani to Kin-shasa,whileTransCongoIII(June2015)wasconductedintheKasai River watershed.A total of 17 suspended particulates were col-lectedusing15-Lpolyethylenereservoirsalongthemaincourseof theCongoandtheKasairivers,andattheendpointoftheirlargest tributaries(Fig. 1B; Table 1). One sample was collected fromthe main stem ofthe Congo River atKisangani (#3; sample number refers to Fig. 1B and Table 1) and is assumed to be

representa-tive of the sediment exported from the upstream Upper Congo Tab

le 1 Geoc hemical com position of ri v er suspende d se diments fr om the Cong o Basin. #R iv e r Ar ea (10 3km 2) Lat ◦(N) Long ◦(E) Bulk Al/K 143 Nd/ 144 Nd cla y 2 SE (10 − 6) εNd La Ce Pr Nd Sm Eu Gd Tb Dy Ho Er Yb Lu (Nd/Yb) N a T ra n sC o n go ri ve r p a rt ic u la te s 1 T shopo 19 0 .56 25. 12 5.7 0 .511968 7 − 12.9 43.82 73.69 10. 14 35.38 6.51 1. 38 5.33 0.80 4.63 0.87 2.49 2.52 0.37 1. 30 2 Lindi 39 0 .57 25. 12 7. 2 0 .511642 11 − 19.3 3 0.78 46.59 5.72 18.67 3.22 0.77 2.66 0.40 2.23 0.43 1. 20 1. 21 0. 17 1. 42 3 Lualaba 1035 0.58 24.76 6.8 0 .512044 10 − 11 .4 36.04 79.33 7. 01 23. 16 4. 14 0.96 3.28 0.51 2.88 0.55 1. 55 1. 64 0.23 1. 30 4 Lomami 116 0.77 24.27 9.6 0 .512034 8 − 11 .6 29.79 48.87 6.02 19.67 3.48 0.79 2.93 0.45 2.62 0.57 1. 71 1. 88 0.27 0.96 5 A ruwimi 122 1. 23 23.62 14. 1 0 .511466 12 − 22.7 26.08 49.90 4.86 16.07 2.79 0.64 2.21 0.33 1. 85 0.37 1. 07 1. 18 0. 17 1. 26 6 Itimbiri 5 7 2 .06 22.70 14.6 0 .511542 5 − 21 .2 59.02 82.65 10.20 32.41 5.23 1. 07 3.89 0.57 3. 15 0.61 1. 74 1. 78 0.25 1. 68 7 M ong a la 53 1. 91 19.82 7. 3 0 .511987 22 − 1 2 .5 – – – – –––– –-–– 8 Lulonda 86 0.68 18.40 11 .6 0.511775 6 − 1 6 .7 – – – – –––– –-–– 9 R uki 179 0.07 18.32 13.8 0 .511847 9 − 1 5 .3 – – – – –––– –-–– 10 Oubangui 640 − 0.49 17 .70 12.8 0 .511744 9 − 17 .3 27 .01 39.72 4.89 16.43 2.89 0.63 2.45 0.38 2.33 0.48 1. 41 1. 54 0.23 0.98 11 Kasai 890 − 3. 18 16.20 11 .2 0.511875 6 − 14.7 37 .46 61 .48 7.47 25.92 4. 17 0.94 3.42 0.43 2.57 0.51 1. 49 1. 54 0.23 1. 54 12 Kw ang o 176 − 3.41 17 .33 – 0.511942 9 − 1 3 .4 – – – – –––– –-–– 13 Kwilu 9 6 − 3.39 17 .41 – 0.511910 12 − 1 4 .0 – – – — –––– –-–– 14 Loang e 4 2 − 4.30 20.03 – 0.512130 15 − 9 .8 – – – – –––– –-–– 15 U pper K asai 245 − 4.36 20.58 – 0.511771 17 − 1 6 .8 – – – – –––– –-–– 16 Sank uru 148 − 4.28 20.44 – 0.511701 1 3 − 1 8 .1 – – – – –––– –-–– 17 Kasai 890 − 3. 17 16.21 – 0 .511822 25 − 1 5 .8 – – – – –––– –-–– Lak e Ki vu se diment − 1. 90 29.00 n.d. 0 .512234 7 − 7. 7 26.40 53.26 5.33 18.64 3.24 0.76 2.68 0.34 1. 79 0.33 0.93 0.93 0. 14 1. 84 a No rm a li ze d to Wo rl d Ri v e r Av e ra g e Silt (WRAS; Ba y on et al., 2015 ).

Fig. 3. AgemodelforcoreKZR-23.(A)CoreKZAI-01Al/Kprofile(Bayonetal.,2012) andagecontrolpointsusedforKZR-23agemodel.(B)CoreKZR-23Al/Kprofile(this study),AMS14_{Cdatesonmixedforaminiferfractions,andagecontrolpointsderivedfromcoreKZAI-01.(C)Agevsdepthrelationshipfor1)Mixedforaminiferafractionsand}

agecontrolpoints(inred),and2)bulkCorgsamples.(D)CalibratedageprofileforAl/KincoreKZR-23andcoreKZAI-01(revised).(Forinterpretationofthecolorsinthe

figure(s),thereaderisreferredtothewebversionofthisarticle.)

(Lualaba)watershed.Aftersettling,suspendedparticleswere trans-ferredintosmallerpolyethylenebottles,priortobeingcentrifuged atthelaboratorywithin50mlcentrifugetubesandovendried.In additiontothesesuspendedparticulatesamples,one surface sed-imentfromLake Kivu(withinthe catchment area ofthe Lualaba River)collectedin2013wasalsoanalyzed.

2.3.Agemodel

Thechronology forcoreKZR-23 relieson two 14C-AMS radio-carbondatesformixedplanktonicforaminiferafractions(150–160 depthand800–820 depth), andtuning ofthe Al/K signal to that ofcoreKZAI-01(Fig.3),alsorecoveredofftheCongoRivermouth (Fig.1B). One additional 14C agewas determinedon echinoderm testfragmentsfromthe800–820depthinterval,whichconfirmed thecorresponding14Cageonmixedforaminifera.These radiocar-bon dates were converted into calendar ages with the Calib 7.1 program(Stuiveretal.,2017),applyingtheMARINE13radiocarbon

age calibration curve (Table S1). Additional constraints on sedi-mentationrates,obtainedfrom14C-AMSagesof21bulksamples, were also used for establishing the age model (Fig. 3). Constant sedimentationrateswere assumedbetweenthe203–961 cmand 961–1481cmdepthintervals,onthebasisthat bothsediment in-tervals exhibited linear ‘depth’ versus ‘bulk 14_C

org’ relationships.

Whilethereisinherentuncertaintyassociatedwiththeagemodel ofKZR-23,duetoits particulardepositionalsettingdominatedby terrigenousinputsandthepoorcarbonatepreservationwithinthe sediment, the timing ofthe high-Al/K episodein the upper part of thecore (which isthe main focus ofthe discussion below) is wellconstrainedbythecombinationoftwo14_{CdatesfromKZR-23}

andKZAI-01(Fig.3B).Notealsothatouragemodelindicatesthat the last

∼

1300yrsare missing fromthesediment record atsite KZR-23 (Fig. 3D); a featurethat we tentatively attribute toa de-creaseofsedimentdynamicsintheCongosubmarinecanyonover therecentperiod.

2.4. Majorelementinorganicanalyses

A total of 140 bulk sediment samples from core KZR-23 wereanalyzed for majorelementconcentrationsatIfremer,using wavelength-dispersiveX-rayfluorescence(WD-XRF).Theprecision of measured Al/K ratios was assessed from replicate analyses of various KZR-23 samples during the course of5 different analyti-cal sessions(<5%; N

=

15).The geochemicalcomposition ofbulk suspendedsedimentscollectedduringtheTransCongoIexpedition was determined by inductively coupled plasma–atomic emission spectroscopy (ICP–AES) at the Royal Museum for Central Africa, with average accuracies typically below 5%. The results for Al/K ratiosinTransCongosuspendedparticulatesandcoreKZR-23 sam-plesarelistedinTable1andTableS2,respectively.

2.5. RareearthelementsandNdisotopicmeasurements

Allbulksedimentandriversuspendedparticulatesampleswere treatedusingasequentialleachingprocedure toremovebiogenic, Fe–Mnoxyhydroxideandorganiccomponents(Bayonetal.,2002). Clay-sizefractionswereseparatedfromtheresidualdetrital mate-rial using low-speed centrifugation(Bayon et al., 2015). Between 40–80 mg of the obtained clay-size fractions were digested by alkalinefusionforrareearthelements(REE)andNdisotopic mea-surements.Notethat duetosmallsample weight (<15 mg),two suspended particulatesamples collected fromrivers draining the central depression of the Congo Basin (Lulonda and Ruki) were digested using concentrated HF:HCl solution without any initial leachingorgrain-sizeseparation.

Rare earth element concentrations were determined on a ThermoScientific QuadX-Series 2atthe RoyalMuseum for Cen-tralAfrica.TheREEabundancesweredeterminedwithaprecision betterthan5%andnormalized toWorldRiverAverageSilt(WRAS; Bayonet al., 2015). Neodymium was isolated using conventional ion chromatography andisotopic measurements were performed atthePôleSpectrométrieOcéan(Brest),usingaThermoScientific Neptunemulti-collectorICPMS.Ndisotopiccompositionswere de-termined using sample-standard bracketing, by analyzing JNdi-1 standardsolutions every twosamples.Mass-biascorrected values for143_Nd/144_{NdwerenormalizedtoaJNdi-1valueof}143_Nd/144_Nd

=

0.512115(Tanakaetal., 2000).Analysesof aninhouse (SPEX-Nd)standardsolutionduringthecourseofthisstudy(2analytical sessions)gave143Nd/144Ndof0.511709

±

0.000007(2SD,n

=

16), corresponding to an external reproducibility of

∼ ±

0.15

ε

(2 SD). Epsilon Nd values (

ε

Nd) were calculated using 143Nd/144Nd

=

0.512630(Bouvier etal., 2008). Neodymium isotopicdata are re-portedin Table1 (TransCongosuspended particles)andTable S2 (KZR-23samples). Together with resultspublished earlieron an-other series of suspended particulate samples from the Congo River (Allègre et al., 1996), these datasets were used to calcu-late end-member Nd isotopic compositions for the three main tributariesof the Congo River (Lualaba, Oubangui andKasai; see Table S3).

2.6. Calculationofapparentinitialradiocarbonages

The apparent initial ages ofthe terrestrial organic matter ex-ported by the Congo River were calculated using bulk AMS ra-diocarbon data and corresponding depositional ages in KZR-23 inferredfromourmodelage(Table S4). Theseapparentages cor-respondto theageof asedimentary organicfractionat thetime of deposition. An apparent initial age of several thousand years (as will be reported below for core KZR-23) generally indicates that rivers were transportingpre-agedplant organicmatter from soils at the time of sediment deposition (Schefuß et al., 2016;

Winterfeldetal.,2018).Followingtheapproachdescribedinthese latter studies, theapparent initial ages (yr) were calculated after correction ofthe atmosphericradiocarbon contentat thetime of sedimentdeposition(14_C

atm),derivedfromtheIntcal13southern

hemisphereatmosphericdata(TableS4):

14_{C age} apparent

(yr)

= −

8,033

×

ln



1

+ 

14Cinitial

/1000



1

+ 

14Catm

/1000



where



14Cinitialcorrespondstotheinitialradiocarboncontentof

bulkKZR-23samplescalculatedusingthefollowingequation (Stu-iverandPollach,1977):



14Cinitial

(

h) =

1000

×



Fm/e((2015–1950)/8267)

−

1



2.7. n-alkaneisotopeanalysesandBIT-index

Bulk organic carbon contents (%OC) in core KZR-23 were de-termined at KU Leuven using a Thermo Flash HT/EA equipped with a Costechzero-blank analyzer (Table S4). For n-alkane iso-tope analyses, lipids were extracted using a Dionex accelerated solventextractorwith9:1(v/v)dichloromethane(DCM)/methanol (MeOH).Asphalteneswereprecipitatedfromthetotallipidextract and hexane-soluble fractions were subsequently saponified with 6%KOHinMeOHfor2 hat85◦C.Saturatedhydrocarbonand po-lar lipid fractions were obtained from neutral lipids using silica column chromatographyby elution withhexaneand DCM/MeOH (1:1), respectively. Saturated hydrocarbon fractions were subse-quentlyelutedonAgNO3-coatedsilica.Quantificationoflong-chain

n-alkaneswasachievedusinga ThermoFisherScientific Focusgas chromatograph equipped with an Rxi-5 ms 30

×

column (30 m, 0.25 mm,0.25 μm) andaflameionization detector.Identification andquantification ofcompounds was done by comparisonto re-tentiontimesandpeakareasofanexternaln-alkanestandard mix-ture. Maximumquantificationuncertaintyis5%. Then-C29 alkane

was consistently the most abundant plant-wax compound in all samplesandisotopeanalyseswerethereforefocusedonthis com-pound.Hydrogenisotopeanalyses(δD)wereconductedatleastin duplicateonaThermoFisherScientificMAT253isotoperatiomass spectrometer (IRMS)coupledviaaGCIsolinkoperatedat1420◦C toaThermoFisherScientificTraceGCatMARUM,Bremen.

δD

com-positions were measured against calibrated H2 referencegasand

δD

valuesarereportedinpermill(

h

)versus ViennaStandardof Mean Ocean Water (VSMOW; Table S5). Average standard devia-tion forreplicate

δD

measurements is 1

h

for the n-C29 alkane.

Theinternalstandard(squalane)yieldedanaccuracyof

+

2

h

and a precision of 3

h

(n

=

94). An external standard mixture with known

δD

valueswasanalyzed repeatedlyeverysixruns, yielding a long-termmeanstandarddeviationof

<3

h

anda mean devia-tionof

<1

h

fromreferencevalues.

Compound-specificstablecarbonisotopeanalyseswerecarried out at least in duplicate on a ThermoFisher Scientific MAT 252 isotope ratio mass spectrometer coupled via a GC-C combustion interface with a nickel catalyzer operated at 1000◦C to a Ther-moFisherScientificTraceGCatMARUM,Bremen.

δ

13Cvalueswere calibrated against CO2 reference gas of known isotopic

composi-tion and

δ

13_{C valuesarereportedin permill(}

_h

_{) againstVienna}

PeeDeeBelemnite(VPDB;TableS5). Averagestandarddeviationof

δ

13C values for the n-C29 alkane is 0.1

h

. The internal standard

(squalane)yieldedanaccuracyof

+

0.4

h

andaprecisionof0.4

h

(n

=

94). An external standard mixture was analyzed repeatedly every 6runsandyieldedalong-termmeanstandarddeviationof 0.2

h

withameandeviationof0.1

h

fromthereferencevalues.

Branched and isoprenoid glycerol dialkyl glycerol tetraether lipids(GDGTs)wereanalyzed inthepolarfractionsfilteredthrough

4 μm pore size PTFE filters. GDGTs were analyzed by high-performance liquid chromatography (HPLC) mass spectrometry (MS)usingamethodslightlymodifiedfromHopmansetal. (2004). GDGTs analyses were conducted using an Agilent 1200 series HPLCsystemcoupledwithanAgilent6120massselectivedetector (MSD)system.Thebranchedandisoprenoidtetraether(BIT)index wascalculatedaccordingtoHopmansetal. (2004).Corresponding BIT-indexvaluesarereportedinTableS5.

2.8.Pollen

Palynologicalsamplepreparationfollowedstandardprocedures, wherebysamplesof0.3–4.2 gdrysedimentwere decalcifiedwith dilutedHClandsubsequentlytreatedwithconcentratedHFto re-movesilicates.Samples were sievedusingultrasonic sievingover a screenremoving particles smaller than 10–15 μm. Pollen sam-ples were mounted in glycerol and microscopically examined at 400

×

and 1000

×

magnification. Pollen percentages were calcu-latedbasedonthetotalnumberofpollenandspores. Confidence intervals(95%) were calculated after Maher(1972). Pollengrains wereidentified usingseveral publications(seesupplementary in-formation),the AfricanPollen Database,andthe reference collec-tionoftheDepartmentofPalynologyandClimateDynamicsofthe UniversityofGöttingen.Alistof114arborealpollentaxa(Vincens etal., 2007) anda listof 22pollen taxa fromthe pioneer forest (Dalibardetal.,2014) aregiveninTableS6.Thepercentagemeans forselectedpollentaxaandgroupsusedinthisstudyarelistedin TableS7.

3. Resultsandproxyinterpretation

3.1.Pastweatheringanderosionchanges:Al/K,BIT-indexandinitial bulkorganicmatterapparent14_Cages

Bulk sediment aluminium-to-potassium ratios (Al/K) in sed-iments and river suspended matter can provide constraints on chemicalweathering intensity.Potassium is highly mobileduring chemical weathering and typically depleted in soils, while alu-miniumisoneofthemostimmobileelementsandisincorporated intosecondary clay minerals (Govinetal., 2012). Direct compar-isonofAl/K valuesmeasured inriversuspended particulatesand marinesedimentsmaybe complicatedby variouslithologicaland grain-sizeeffects(Bouchezetal.,2011).However,theevolutionof Al/K ratios and other weathering proxies in sedimentary records is generally considered to reflect past fluctuations in alteration processescoupledwithchanginghydroclimateandvegetation pat-ternsoncontinents(e.g.Schneideretal.,1997; Bayonetal.,2012; Ivory et al., 2017). At the Congo margin, the utility of Al/K for reconstructing past chemical weathering has been demonstrated usingHfisotopes,whichrepresentanotherpowerfulindicator for silicate weathering intensity (Bayon et al., 2009, 2012). In core KZR-23,bulksedimentAl/K ratiosprogressivelydecreasebetween

∼

10.0and2.2thousand yearsago (from7.6to 6.4;Fig.4A), sug-gestingdecreasingsoilweatheringinresponsetoageneraldrying trend. The most striking feature of the Al/K profile is the sud-denandpronounced increase between2.2and1.8kyr BP (upto Al/K

∼

8.7),whichindicatesexportofhighly weatheredsediments at that time. Unequivocally, this episode corresponds to the so-calledthirdmillenniumBPweatheringeventidentifiedpreviously in Congo fan sediments (Bayon et al., 2012). The timing of the Al/Kpeakslightlydiffersbetweenthetwosedimentrecords:about 2.3 kyrBPforcoreKZAI-01(Bayonetal.,2012) and1.9kyrBPfor KZR-23.This difference arises fromminor agemodel offsets, but therevisedagemodelreportedhereisbetterconstrainedbyAMS radiocarbondata andhencerepresentsa moreaccurate temporal estimatefortheevent(Fig.2).

SedimentationatsiteKZR-23 isalsocharacterized by substan-tialinputsofterrestrialorganicmatter(OM),asinferredfromthe branched and isoprenoid tetraether (BIT) index. In marine sedi-ment records,the BITindexrepresents a measureof therelative proportionofsoil-derived(BIT

∼

0.8–1)versusmarine(BIT

∼

0–0.2) organic matter (Hopmans et al., 2004). In core KZR-23, the BIT index remains high (0.76

±

0.05; 1 SD) throughout the entire Holocene (Fig. 4B), indicating that accumulation of bulk OM at thissitewas dominatedby terrestrialinputs. In thiscontext, the development of theage oforganicmatter deposited atthe stud-ied site (as inferred from the difference between bulk sedimen-tary OM radiocarbon ages and corresponding depositional ages; see section 2.6), can provide useful information on past erosion processesofpre-aged terrestrialOMintheCongo Basin(Fig.4C). CoreKZR-23sedimentsdisplayapparentinitialbulkOMages rang-ingfrom

∼

1400to300014_{Cyr(TableS4),whichsuggestthatthe}

Congo River has been exporting aged organic matter throughout the Holoceneperiodin accordancewithearlierfindings (Schefuß etal., 2016). At siteKZR-23,the ageof soil-derivedOM isabout 2,900

±

200 14_{C yr (1 SD) between 10.0 and 5.0 kyr BP, then}

progressivelydecreases (downto

∼

160014C yr)during the mid-to-late Holocene period, before increasing suddenly after 2.2kyr BP (up to

∼

2,700 14_{C yr). This trend differs fromthat obtained}

frombulkOMandcompound-specificradiocarbonanalysesinthe nearby coreGeoB6518-1(Schefußet al., 2016). The causeofthis discrepancyisunclearbutlikelyrelatestodifferencesin sedimen-tationratesandassociateddiagenetic processesbetweenthetwo sites.Tosomeextent,theinherentuncertaintyassociatedwithour age modelfor KZR-23 could alsoaccount, at leastpartly, forthe observed discrepancybetween KZR-23and GeoB6518-1 bulk OM radiocarbonrecords.Nevertheless,thesuddenshifttowardsgreater initial OMagesafter 2.2kyrBP,coinciding withevidencefor ex-portofhighlyweatheredsediments(asinferredfromAl/Kratios), strongly suggests that enhanced erosion of pre-aged terrigenous OMoccurredintheCongoBasinatthattime.

3.2. Vegetationpatterns:Pollenand

δ

13_C

wax

InformationonpastvegetationpatternsintheCongoBasincan be inferredfromboth pollenassemblagesandcompound-specific stablecarbonisotoperatiosofplantlipids.The mid-Holocene pe-riodappears todisplayslightlyhigherrelative abundancesoftree pollentaxa,whichmainlyreflectincreasedproportionsofdry for-est woodlands relative tothe rainforest trees (Fig. 4D;Table S7). However, pollen assemblages in KZR-23 do not reveal any clear long-term trend during most of the Holocene, with abundances oftreepollen, rainforest andforest/woodlandtaxa displaying rel-ativelylimiteddowncorevariations (Fig.4D).Similarly,stable car-bonvalues (δ13_C)ofthen-C

29alkanes display nearconstant

val-ues in core KZR-23 (

−

34.1

±

0.3

h

VPDB; Table S5). Although higherresolution profiles forpollenand/orn-alkanes

δ

13C would be required to more accurately reconstruct past vegetation pat-ternsintheCongoBasin,theseresultssuggestthatlimited vegeta-tionchanges occurredinthecentral Africanrainforest duringthe Holocene.However,liketheweatheringanderosionproxydata re-ported above, pollen-based vegetation patterns also demonstrate a significant shift after 2.2 kyr BP, characterized by a reduction of tree pollen (Fig. 4D), an increase of oil palm (from near-zero values to

>1%;

Fig. 4E) and forest pioneer species(from

∼

2 up to 6%;Fig. 4E), followed after2 kyr BP by a pronounced expan-sion of grasses (from

∼

6 up to 16%; Fig. 4D). In tropical Africa, oilpalmandotherpioneerspeciescanusuallyrapidlycolonize ar-easthathavebeensubjecttoforestdegradation(Sowunmi, 1999; Maley,2001).Therefore,theirpresenceincoreKZR-23after2.2kyr BPprovidesevidenceforforestdeclineintheCongo Basinduring thelateHolocene.

Fig. 4. CoreKZR-23proxyrecordsfortheevolutionofpasterosionandvegetationpatternsinCentralAfrica.(A)BulksedimentAl/Kratiosindicatepastchangesinchemical weatheringanddegreeofalterationofsoils.NotethattwodifferentY-axisscalesusedforthelateHolocene(from7.5to9),andtherestofcoreKZR-23.(B)BITindex valuescalculatedfromGDGTsdata,representingtherelativeproportionofterrestrial(BIT∼0.8–1)versusmarine(BIT∼0–0.2)OMinbulksediment.(C)Apparentinitial agesofbulksedimentorganiccarbon.(D)Percentagesofpollentaxafortrees,includingrainforestandwoodlands,normalizedtototalpollencounts.Shadingindicates95% confidenceintervals.PercentagesofpollentaxaforPoaceae(grasses).(E)Percentagesofpollentaxafrompioneerforestandoilpalm(Elaeisguineensis).Theverticalyellow bandcorrespondstotheLateHolocene‘weathering’eventinferredfromAl/Kratios.

3.3. Sedimentprovenanceproxies:Ndisotopesandrareearthelements The suite of river particulate samples collected within the CongowatershedwasusedtoidentifythecharacteristicNdisotope andREE signatures of each major sub-basin(Table 1). Measured Nd isotopic compositions range from

ε

Nd

−

22.7 (Aruwimi; #3)

to

−

9.8 (Loange; #14), in agreement with previously published results forother river suspended particles from the Congo Basin (Allègre et al., 1996) anda recentregional synthesis for Nd iso-topesinCentralAfrica(Thiéblemontetal.,2014).Ourdataindicate thattheOubangui,KasaiandUpperCongo(Lualaba)rivers,which represent the most important tributaries of the Congo in terms of solid particulate discharge (Laraque et al., 2009), are charac-terizedby distinctive Nd isotopic signatures (

ε

Nd

= −

17.6

±

0.4;

−

15.5

±

0.7 and

−

11.4

±

0.2, respectively; see Table S3 for the definition of the end-members). While also representing a large partoftheCongowatershed,theriversdrainingtheCuvette Cen-trale(e.g.Ruki,Lulonda,Mongala)carryverylittletotalsuspended material in comparison (Laraque et al., 2009), and hence con-tributenegligibly to the total solid discharge of the Congo River.

The Oubangui, Kasai and Lualaba river particles also display dif-ferences in WRAS-normalized REE patterns, with Oubangui sus-pendedparticlesbeingcharacterizedbyslightlydepletedmidREE (MREE) normalized-abundances relative to light and heavy REE, andparticulatematter fromtheKasaiandLualaba rivers exhibit-ing morepronounced LREE enrichment relative to the other REE (Fig. 5). ThisresultsinsuspendedparticulatesfromtheOubangui, Kasai andLualababeingcharacterized bydistinctive neodymium-to-ytterbiumratios(takenhereasameasureoftherelative enrich-mentofLREEoverHREE),with(Nd/Yb)N of

∼

0.98,1.54,and1.30,

respectively (Table 1). Overall, these characteristics indicate that bothNdisotopesandREEcanbeusedtoreconstructpastchanges between northern (Oubangui) and southern/eastern (Lualabaand Kasai) sedimentaryinputsto siteKZR-23, andthus toprovide in-formation onthegeographic originofthe sedimentsexportedby theCongoRiver.

In coreKZR-23, Ndisotopes display overall homogeneitywith anaverage

ε

Ndvalueof

−

15.6

±

0.1 (1SD;Fig.6A),whichsuggests

that sediment provenance, and hence the mean latitudinal posi-tion of the rainbeltover the Congo Basin, have remainedmostly

Fig. 5. Rareearthelementconstraintsonsedimentprovenancechanges.(A) Rare earthelementabundancesofclay-sizefractionsfromriversuspendedmaterialin theCongoBasin(squares)andcoreKZR-23sediments(gray lines),normalizedto WorldRiverAverageSilt(Bayonetal.,2015).(B)PlotofεNdversus(Nd/Yb)Nratios

forKZR-23detritalfractions.Thethreemainsourcesofterrigenousmaterialtothe solidloadexportedfromtheCongoRiverBasin(Oubangui,KasaiandLualaba)are plottedforcomparison.Mixinglinesbetweenthethreeend-membersareshown, withsmallgray circlescorrespondingto20%,40%,60%and80%relative contribu-tions.

unchangedoverthelast10thousandyears.Minor

ε

Nd fluctuations

occurdowncore,inparticulararound2kyrBP,whenasubtlebut significant shifttowards more radiogenicsignatures (from

−

15.8 to

−

15.2) indicates a period ofenhanced relative sediment con-tributions from the southern and/or eastern parts of the Congo watershed.This assumptionisfurthersupported by the(Nd/Yb)N

profile in KZR-23, which similarly displays a sudden increase at about2.2kyrBP(Fig.6B),indicativeofenhancedrelativesediment input from the Kasai sub-basin. Quantitative constraints on past provenancechangesincoreKZR-23canbeobtainedusingamixing modelthatcombines

ε

Ndand(Nd/Yb)Nend-membercompositions

fortheOubangui, LualabaandKasairivers (Fig.5B).Theseresults suggestthatsedimentationatsiteKZR-23hasbeendominatedby inputsfromtheKasaiRiverduringtheentireHolocene,withan av-eragecontributionofabout68

±

10%,whileOubanguiandLualaba riversaccountedfortheremaining

∼

23

±

8%and9

±

4%, respec-tively(Fig.7A).RelativesedimentaryinputsfromtheOubanguiand Kasai rivers display little long-term fluctuationsbetween 10and 2.2kyr BP (Fig. 7A), in contrast withsediment discharge associ-atedwiththeLualabaRiver,whichdecreasessignificantlyfromthe mid-Holoceneto2.2 kyrBP(fromabout16%to5%).Assuspected from(Nd/Yb)N ratios,our mixingmodel alsoindicates an abrupt

shifttowardsincreasingsedimentcontributionsfromtheKasai re-gionbetween

∼

2.2to2.0kyrBP,immediatelyfollowedbyasharp

decrease (downto

∼

50%) andenhanced export ofparticles from theLualabaandOubanguicatchmentareas(Fig.5B;Fig.7A). 3.4. Hydroclimate

δD

waxrecord

Informationonpasthydrologicalchangescanbe inferredfrom the application of hydrogen isotope analysis of plant wax de-rived biomarkers (δDwax, in

h

relative to VSMOW). In tropical

areas, hydrogen isotope abundances in plant waxes are mainly controlled by rainfall amounts and the precipitation–evaporation balance, therefore reflecting predominantly continental hydrolog-ical changes (Sachse etal., 2012). As already reported in earlier studies(Schefußetal.,2005),thelong-termprecipitationtrend in-ferredfrom

δD

waxincoreKZR-23,fromabout

−

150

h

intheearly

Holoceneto

−

134

h

inthemorerecentperiod(Fig.6C),indicates progressivelydecreasingrainfallsinceabout6.5kyrBPinresponse to declining northern hemisphere summer insolation. This gen-eraldryingtrendispunctuatedbya

∼

10

h

shifttowardsheavier

δD

waxcompositionsbetween7.5and6.5 kyr BP,whichprobably

in-dicatesslightlyreduced precipitationlevels atthat time. Another shift towards slightly more depleted

δD

wax values occurred

be-tween 2.2 and1.4 kyr BP (i.e. the mostrecent sediment in core KZR-23), but its amplitude is small (less than 5

h

), so that it could alsoreflectisotopiceffectsduetochangesinbasin-internal rainfall patterns.At present, precipitation inthe Kasai watershed is isotopically depleted relative to rainfall in the Oubangui and Lualaba catchments(http://wateriso.utah.edu/waterisotopes/media/ IsoMaps/jpegs/h_Africa/Hma_Africa.jpg). A decrease in rainfall in the lattersub-catchments could thus haveled to a shifttowards relativelydepleted

δD

wax signaturerecordedinKZR-23.

4. Discussion

4.1. EvidenceforLateHoloceneclimatechangeintheCongoBasin Takentogether,NdisotopesandREEdataprovideevidencefor climateinstability in theCongo Basin fromabout2.2kyr BP on-wards,presumablyassociatedwithchangingdischargepatternsin different parts of the watershed. In agreement with

δD

wax, Nd

isotopes and REE suggest that the period after

∼

2.2 kyr BP was associated with decreasing rainfall in the northern and eastern regionsdrainedbytheOubanguiandLualaba,respectively,and rel-atively elevated discharge from the Kasai sub-basin. At the scale ofthevastCongo Basin, thiseventdoesnotappearto havebeen related to any substantial change in total basin-integrated pre-cipitation levels,as inferred fromthe relatively smallshift in

δD

compositions of plant leaf waxes. For this time interval, there is plenty of evidence for contrasting patterns of climate change acrossCentralAfrica.InwesternequatorialAfrica,thisperiodwas accompaniedbyintenselandsurfaceprocessesassociatedwiththe formationofcoarse-graineddeposits,followedbyrapiddeposition of a cover horizon of potential aeolian origin, which were both interpreted as the consequence of abrupt climate deteriorations (Thiéblemont et al., 2013, 2014; Maley et al., 2018). Instead, in eastern tropical Africa,precipitation proxyrecordsgenerally indi-catewetterconditionsduringthisperiod(e.g.Tierneyetal.,2011; Foersteretal.,2012).Inparticular,ourinferredlate Holocene hy-droclimatic shift in the Congo Basin coincides with a period of markedly increasing rainfallat theChew BahirBasin insouthern Ethiopia (Foerster et al., 2012; Fig. 7B); an event that was at-tributed tostrengthening of El Niño-SouthernOscillation(ENSO). The distribution ofprecipitation in tropical EastAfrica is indeed partlydrivenbytropicalIndo-PacificclimateandaffectedbyENSO (Preethietal.,2015).While easternequatorialAfricatends to ex-perience greaterthannormalprecipitationduringan ENSOevent, south-easternAfricanregionsdisplaydrierthannormalconditions

Fig. 6. ProxyrecordsforsedimentprovenanceandhydroclimateinKZR-23duringtheHolocene.(A)Ndisotopes(εNd)infine-graineddetritalfractionsrepresentrelative

sedimentinputsfromtheUbangui(−17.6),Kasai(−15.5),andLualaba(−11.4)sourceareasandprovideinformationonpastshiftsoftherainbeltfromnorthernto south-ern/easternpartsofthewatershed,respectively.Analyticaluncertaintyifshowningray shading.(B)Evolution(Nd/Yb)Nindetritalfractionsasproxyforsedimentprovenance.

(C)Hydrogenstableisotopecompositionofleaf-waxn-C29alkanes(δDwax;inhVSMOW),withgray shadingcorrespondingtoanalyticaluncertainty.Morenegativevalues

indicatewetterconditions.TheverticalyellowbandcorrespondstotheLateHolocene‘weathering’eventinferredfromAl/Kratios.

(Ropelewski and Halpert, 1987; Nicholson and Kim, 1997). One of the presumed mechanisms behind these climate anomalies is thatENSOeventsareassociatedwithaneastwardmigrationofthe Congo Air Boundary (CAB), i.e.the convergence zone that marks theconfluence ofIndian Ocean airwithair massesderived from the Congo Basin (Ropelewski and Halpert, 1987; Tierney et al.,

2011). Todate,the impactofENSO events oncentral African cli-materemainslargely unknown,dueto alackofcontinuous long-term meteorological data for the Congo Basin. However, such a linkhasbeenalreadyevoked (Mbuebueetal.,2016),whichcould possiblyaccount forsubstantial rainfallchanges inCentralAfrica, drawingmoisture away fromthecentralandeastern partsofthe CongobasintowardseasternequatorialAfrica(Tierneyetal.,2011; Jungingeretal., 2014). Whilethemid-Holocene periodwas glob-allycharacterizedby weak ElNiñoactivity,palaeoclimate records fromkey ENSO-sensitiveregions generallyindicatemuchstronger and/ormorefrequent ElNiñoeventsbetween2000and1000yrs ago(Luetal., 2018;see Fig.7C–Eandreferencestherein). Alate HolocenestrengtheningofENSOwouldthusbecoherentwithour finding, based on Nd isotopes and REE, that precipitation levels abruptlydecreasedinboththe LualabaandOubanguiwatersheds from

∼

2200yrBPleadingtoarelativeincreaseofthesignalfrom the Kasai sub-basin. To summarize, we tentatively propose here that increasingEl Niñoactivityduring the lateHolocenepossibly played an importantrole in drivingthe sudden reorganization of rainfallpatternsacrosstheCongo watershedafter

∼

2.2kyrBP,as inferredfromourhydroclimateproxyrecordsincoreKZR-23. 4.2. AclimaticforcingfortheeastwardBantumigrationacrossCentral Africa?

Collectively,ourproxyrecordsforpastvegetation,erosionand weatheringpatternsinCentralAfricaclearly showthat major en-vironmentaldisturbances occurredbetweenabout2.2and1.6kyr BP, hence contemporaneously with the period of presumed

cli-mate instability inferred fromour Nd isotope andREE data. The timing of this environmental crisis is slightly younger, but con-sistent with previous evidence froma numberof terrestrialsites across CentralAfrica, which all point to important forest retreat and erosive surface processes between about2.6 and2.0 kyr BP (Vincensetal.,1998; Maley,2002,2018;Thiéblemontetal.,2013; Bostoen et al., 2015). Importantly, this period also coincides re-markablywell withthefirstmajor immigrationwave ofIronAge communities into the rainforest, as inferred from recent synthe-ses of radiocarbon dates for archaeologicalsites in western Cen-tral Africa (Fig. 7F; Oslisly etal., 2013; Morin-Rivat et al., 2014; Garcin et al., 2018). Evidence for simultaneous occurrence of changesinhydroclimate,vegetationanderosionwithintheCongo Basin during this time interval thus suggests a close connection withtheBantumigrationsacrossthecentralAfricanrainforest.We speculate that changing hydroclimate patternsin the Congo’s in-land basin during the third millennium BP may have played an importantroleindrivingtheBantumigrationacrossCentralAfrica. Theevolutionofsedimentprovenancechangesreconstructedfrom NdisotopesandREEatsiteKZR-23suggeststhatprecipitation lev-elsinthenorthernandeasternregionsoftheCongobasinbecame severely reduced from

∼

2.2 and2.0kyr BP, while comparatively wetter conditionsprevailedintheKasaicatchment area(Fig. 7A). Asproposedearlier(Bostoenetal.,2015),suchaclimatic reorgani-zation, whichwetentativelyattribute toenhancedENSO variabil-ity,couldhaveledtoforestretreatintheseregions, possibly trig-gering large-scale displacement of Bantu-speaking populations at thattimethroughopeningofsavannacorridors.Albeitspeculative, thishypothesiswouldbecoherentwithphylogeneticinvestigations oftheirdispersalroute,whichsuggestthatafterfollowinga south-eastward directionfirst,theBantu farmersthen movedeastwards to reach theeastern fringe ofthe Congo Basinby about2000yr ago(Grollemundetal.,2015).

Fig. 7. ReconstructedpastsedimentprovenancechangesinKZR-23,comparedtootherHolocenerecordsforpalaeoclimateandironagehumansettlementsinCentralAfrica. (A)RelativepercentagecontributionsofsedimentfromtheKasai,OubanguiandLualabacatchmentareascalculatedusingamixingmodelcombiningεNdand(Nd/Yb)N

end-membercompositions.NotethedifferentscalefortheLualaba%sedimentcontribution.Shadingindicatestheuncertaintyonmodeled valueswhentakingintoaccount uncertainties(1SD)ontheNdisotopiccompositionofsedimentend-members.(B)ChewBahirBasinpotassium(K)record(notereversescale)ofpastaridityinsouthern Ethiopia(Foersteret al.,2012).(C–E)Paleo-ENSOsedimentrecordsfromGalapagosLakeElJunco(percentsandfraction;Conroyetal.,2008);LakeTutira,New-Zealand (stormeventspercentury;Gomezetal.,2012);TuggerahLake,Australia(ENSOassociatedchangesinorganiccarbonaccumulationrates;Macreadieetal.,2015).(F)Number ofradiocarbondatesforarchaeologicalsitesinwesternCentralAfrica(Garcinetal.,2018).TheverticalyellowbandcorrespondstotheLateHolocene‘weathering’event inferredfromAl/Kratios.

4.3.ClimateversushumanimpactsonthelateHoloceneenvironments inCentralAfrica

Ourmulti-proxyinvestigation allows usto re-evaluatethe hy-pothesisthathumansmayhavebeeninvolvedinthelateHolocene crisis ofthe central Africanrainforest (Bayonet al., 2012; Garcin etal.,2018).The thirdmillennium BPepisodeofforest clearance

inCentralAfricaisgenerallyattributedtoenhancedaridityand/or more pronouncedclimate seasonality (e.g.Neumann etal., 2012; Bostoenetal.,2015; Maleyetal.,2018).Similarly,increasedstorm activityrelatedtoenhancedrainfallseasonalityhasalsobeen pro-posedasanexplanationfortheintenseerosionaleventsthattook placeduring thelate Holocene, whicharenow preservedasvery coarse-grainedsedimentarydepositsinmanyuppersoilsequences

acrossCentralAfrica(Thiéblemontetal.,2013; Maleyetal.,2018). As suggested above, the late Holoceneenvironmental changes in the Congo Basin, reconstructed from our proxy data, could also be explained by a prolonged phase of climate instability related tostrengtheningofENSO.Toalargeextent,theobservedchanges in coreKZR-23 after 2.2 kyr BP, includingchanges in vegetation patterns,could havebeen driven by an internal basin-wide shift of precipitation. In this context, the erosional material inferred from bulk OM radiocarbon and Al/K data after 2.2 kyr BP could correspond to remobilized pre-aged soilspreviously deposited in the Cuvette Centrale. This important storage reservoir for par-ticulate OM has been forming since the early Holocene and is thoughttobeparticularlysensitivetoprecipitationchangeswithin the watershed (Schefuß et al., 2016; Hemingway et al., 2017; Dargieetal.,2017).IncoreKZR-23,theincreaseinbulkOMinitial

14_{Cagesat2.2kyrBPcoincidedwithmarkedlyreducedsediment}

inputs fromthe Lualaba andthe Oubangui watersheds (Fig. 7A), henceatatimewhendrierconditionsprobablyprevailedin north-ern and eastern Central Africa. This particular climatic context couldhaveresultedinasubstantialdropofthewaterlevelwithin theCuvette, whichwould haveled to remobilizationofpre-aged particulateOMandpossiblyhighlyweatheredsediments.

While our results appear to provide support for the role of climate in the late Holocene rainforest crisis, the magnitude of theobservedenvironmentalsignals,especiallyforAl/Kratios,still remains enigmatic and could point towards human imprint on the rainforest as suggested by Bayon etal. (2012) and more re-centlyby Garcinetal. (2018).ComparedtotheCongoBasin,Lake Barombi has a much smaller catchment area and the develop-ment ofhuman activities in thispartof southwestern Cameroon during the third millennium BP most likely led to an amplified anthropogenicsignal nowrecordedinthe lakesedimentdeposits (Garcin etal., 2018). Incontrast,the vastnessof theCongo Basin andthediversityofvegetationpatternsacrossitswatershed prob-ably acted as a buffer that dampened the anthropogenic signal exported by river sediments and pollen at that time. Neverthe-less, settlements ofIron Agecommunities in the rainforest were accompanied by slash-and-burn practices and iron ore smelting, which locally led to woodland clearance (Oslisly et al., 2013; Lupo et al., 2015). Bayonet al. (2012) showed that the pulse of highly weatheredmaterialexported fromtheCongo Basinduring the late Holocene had been unprecedented for at least the last 40ka,andthiswas takenasevidence forlarge-scalehuman im-pactoncentralAfricanlandscapes.Asproposedearlier(Sowunmi,

1999),theincreasingabundanceofoilpalmpollenrecordedatsite KZR-23after2.2kyrBP(Fig.4E)couldindicatelocalplant cultiva-tion,although its expansion inthe Congo Basincould also relate toclimate deteriorationandassociated forest disturbance(Maley,

2001). WhiletheforestdeclineduringthelateHoloceneprobably openeddispersalroutestowardstheeast(Grollemundetal.,2015), thecombinationofpresumed intensifying land-useand strength-ening of ENSO variability, presumably associated with changing rainfallpatternsintheCongo Basinafter2.2kyrBP,could be re-sponsible, at least locally, for accelerated soil erosion and more intensechemicalweathering.Thisseriesofeventswouldprovidea complementarymechanismpossiblyaccountingforthemagnitude oftheenvironmentaldisturbancesthattookplaceinCentralAfrica during the late Holocene, suggestingthat afteran initialclimatic triggerland-useintensificationalsoplayed aroleintherainforest crisis.

5. Concludingremarks

This studyprovides evidence for a major landscape reorgani-zationinthe CongoBasinfromca.2.2kyrBP onward,associated withforestretreatandexpansionofsavannas.Thiseventoccurred

contemporaneously with a period of climate instability and was characterized by erosion of pre-aged and intensively weathered soils,whichcouldindicateremobilizationofswampdepositsfrom thecentral partofthewatershed,inresponsetodecreasedwater levelsintheCuvetteCentrale.Weproposethatchanging hydrocli-mate patternsin theCongo Basinatthat time were triggered by theonsetofenhancedENSOvariability,possiblyleadingtoopening ofhumandispersalroutestowardstheeastandfavoring the migra-tion ofBantu-speakingpopulations acrossCentralAfrica.While it remains difficult todiscriminate betweenclimateandhuman ac-tivities as the dominant cause of abrupt environmental changes in the vast Congo basin, it isvery likely that thesefactors were closelyinterconnectedduringthelateHolocenerainforestcrisis. Acknowledgements

ThisworkwasfundedviaanIEFMarieCuriefellowshiptoG.B. (Grant No. FP7-PEOPLE-2012-IEF 327778) and the DFG Research Center/Cluster of Excellence “The Ocean in the EarthSystem” at MARUM.WeareverygratefultoS.Bouillonforprovidingbulk or-ganic data and important feedback on thiswork. We also thank J. Etoubleau for major element analyses, the participants of the ZAIROV cruise (PI:Bruno Savoye)for their assistance at sea,and C. Delvauxforproviding sediment fromLakeKivu.TheEditor (D. Vance) andthreeanonymousreviewers arealso greatly acknowl-edged for providing constructive comments on this manuscript. The TransCongo project was funded through a FNRS grant (PDR 14711103)anda European ResearchCouncilStarting Grant (ERC-StG240002AFRIVAL).Alldatareportedinthispaperarelistedin theSupplementaryMaterial.

Appendix A. Supplementarymaterial

Supplementarymaterialrelatedtothisarticlecanbefound on-lineathttps://doi.org/10.1016/j.epsl.2018.10.016.

References

Allègre,C.J.,Dupré,B.,Negrel,P.,Gaillardet,J.,1996.Sr–Nd–Pbisotopesystematics inAmazonandCongoRiversystems:constraintsabouterosionprocesses.Chem. Geol. 131,93–112.

Babonneau,N.,Savoye,B.,Cremer,M.,Bez,M.,2004.Multipleterraceswithinthe deepincisedZaireValley(ZaïAngoProject):aretheyconfinedlevees?Geol.Soc. (Lond.)Spec.Publ. 222,91–114.

Bayon,G.,German,C.R.,Boella,R.M.,Milton,J.A.,Taylor,R.N.,Nesbitt,R.W.,2002. SrandNdisotopeanalysesinpaleoceanography:theseparationofboth detri-talandFe–Mnfractionsfrommarinesedimentsbysequentialleaching.Chem. Geol. 187,179–199.

Bayon,G.,Burton,K.W.,Soulet,G.,Vigier,N.,Dennielou,B.,Etoubleau,J.,Ponzevera, E.,German,C.R.,Nesbitt,R.W.,2009.HfandNdisotopesinmarinesediments: constraintsonglobalsilicateweathering.EarthPlanet.Sci.Lett. 277,318–326. Bayon,G.,Dennielou,B.,Etoubleau,J.,Ponzevera,E.,Toucanne,S.,Bermell,S.,2012.

IntensifyingweatheringandlanduseinIronAgeCentralAfrica.Science 335, 1219–1222.

Bayon,G.,Toucanne,S.,Skonieczny,C.,André,L.,Bermell,S.,Cheron,S.,Dennielou, B., Etoubleau, J., Freslon,N., Gauchery,T., Germain, Y.,Jorry, S.J.,Ménot, G., Monin,L., Ponzevera,E.,Rouget,M.-L.,Tachikawa,K.,Barrat, J.A.,2015.Rare Earthelements and neodymiumisotopes inworldriversediments revisited. Geochim.Cosmochim.Acta 170,17–38.

Bayon,G.,Delvigne,C.,Ponzevera,E.,Borges,A.V.,Darchambeau,F.,DeDeckker,P., Lambert,T.,Monin,L.,Toucanne,S.,André,L.,2018.Thesiliconisotopic compo-sitionoffine-grainedriversedimentsanditsrelationtoclimateandlithology. Geochim.Cosmochim.Acta 229,147–161.

Borges,A.V.,Darchambeau,F.,Teodoru,C.R.,Marwick,T.R.,Tamooh,F.,Geeraert,N., Omengo,F.O.,Guérin,F.,Lambert,T.,Morana,C.,Okuku,E.,Bouillon,S.,2015. GloballysignificantgreenhousegasemissionsfromAfricaninlandwaters.Nat. Geosci. 8,637–642.

Bostoen,etal.,2015.MiddletolateHolocene paleoclimaticchangeandtheearly Bantu expansioninthe rain forestsofWesternCentralAfrica.Curr. Anthro-pol. 56,354–384.

Bouchez,J.,Gaillardet,J.,France-Lanord,C.,Maurice,L.,Dutra-Maia,P.,2011.Grain size control ofriversuspended sediment geochemistry:clues from Amazon Riverdepthprofiles.Geochem.Geophys.Geosyst. 12(3).

Bouvier,A.,Vervoort,J.D.,Patchett,P.J.,2008.TheLu–HfandSm–Ndisotopic compo-sitionofCHUR:constraintsfromunequilibratedchondritesandimplicationsfor thebulkcompositionofterrestrialplanets.EarthPlanet.Sci.Lett. 273,48–57. Conroy,J.L., Overpeck,J.T., Cole, J.E., Shanahan, T.M., Steinitz-Kannan, M., 2008.

HolocenechangesineasterntropicalPacificclimateinferredfromaGalápagos lakesedimentrecord.Quat.Sci.Rev. 27,1166–1180.

Dalibard,M., Popescu,S.M.,Maley,J., Baudin,F.,Melinte-Dobrinescu,M.C.,Pittet, B.,Marsset,T.,Dennielou,B.,Droz,L.,Suc,J.P.,2014.High-resolutionvegetation historyofWestAfricaduringthelast145ka.Geobios 47,183–198.

Dargie,G.C.,Lewis,S.L.,Lawson,I.T.,Mitchard,E.T.A.,Page,S.E.,Bocko,Y.E.,Ifo,S.A., 2017.Age,extentandcarbonstorageofthecentralCongoBasinpeatland com-plex.Nature 542,86–90.

Diamond,J.,Bellwood,P.,2003.Farmersandtheirlanguages:thefirstexpansions. Science 300,597–603.

Foerster,V.,etal.,2012.ClimaticchangerecordedinthesedimentsoftheChew Bahirbasin,southern Ethiopia, duringthe last45,000 years. Quat. Int. 274, 25–37.

Garcin,Y., et al., 2018. Early anthropogenicimpact on WesternCentral African rainforests2,600yago.Proc.Natl.Acad.Sci.USA.https://doi.org/10.1073/pnas. 1715336115.

Gomez,B., Carter,L., Orpin, A.R.,Cobb, K.M.,Page, M.J., Trustrum,N.A., Palmer, A.S., 2012. ENSO/SAM interactions during the middle and late Holocene. Holocene 22,23–30.

Govin,A.,Holzwarth,U.,Heslop,D.,FordKeeling,L.,Zabel,M.,Mulitza,S.,Collins, J.A.,Chiessi,C.M.,2012.DistributionofmajorelementsinAtlanticsurface sed-iments(36N–49S):imprintofterrigenousinputandcontinentalweathering. Geochem.Geophys.Geosyst. 13(1).

Grollemund,R.,Branford,S.,Bostoen,K.,Meade,A.,Venditti,C.,Page,P.,2015.Bantu expansionshowsthathabitataltersthe routeandpaceofhumandispersals. Proc.Natl.Acad.Sci.USA 112,13296–13301.

Hemingway,J.D.,Schefuß,E.,Spencer,R.G.M., Dinga,B.J.,Eglinton, T.I.,McIntyre, C.,Galy,V.V.,2017.Hydrologiccontrolsonseasonalandinter-annual variabil-ityofCongoRiverparticulateorganicmattersourceandreservoirage.Chem. Geol. 466,454–465.

Hopmans, E.C., Weijers, J.W.H., Schefuß, E., Herfort, L., Sinninghe Damsté, J.S., Schouten,S., 2004.Anovelproxyfor terrestrialorganic matterinsediments basedonbranchedandisoprenoidtetraetherlipids.EarthPlanet.Sci.Lett. 224, 107–116.

Ivory,S.J.,McGlue,M.M.,Ellis, G.S.,Boehlke,A.,Lézine, A.M.,Vincens,A.,Cohen, A.S.,2017.EastAfricanweatheringdynamicscontrolledbyvegetation-climate feedbacks.Geology 45,823–826.

Junginger,A.,Roller,S.,Olaka,L.A.,Trauth,M.H.,2014.Theeffectsofsolarirradiation changesonthemigrationoftheCongoAirBoundaryandwaterlevelsof paleo-LakeSuguta,NorthernKenya Rift,duringtheAfricanHumidPeriod(15–5 ka BP).Palaeogeogr.Palaeoclimatol.Palaeoecol. 396,1–16.

Lambert,T.,Bouillon,S.,Darchambeau,F.,Massicotte,P.,Borges,A.V.,2016.Shiftin thechemicalcompositionofdissolvedorganicmatterintheCongoRiver net-work.Biogeosciences 13,5405–5420.

Laraque,A.,Bricquet,J.P.,Pandi,A.,Olivry,J.C.,2009.Areviewofmaterialtransport bytheCongoRiveranditstributaries.Hydrol.Process. 23,3216–3224. Laraque,A.,Castellanos,B.,Steiger,J.,Lòpez,J.L.,Pandi,A.,Rodriguez,M.,Rosales,

J.,Adèle,G., Perez,J., Lagane,C.,2013. Acomparisonofthe suspendedand dissolvedmatterdynamicsoftwolargeinter-tropicalriversdrainingintothe AtlanticOcean:theCongoandtheOrinoco.Hydrol.Process. 27,2153–2170. Lézine,A.M.,Holl,A.F.C.,Lebamba,J.,Vincens,A.,Assi-Khaudjis,C.,Février,L.,Sultan,

E.,2013.TemporalrelationshipbetweenHolocenehumanoccupationand veg-etationchangealongthenorthwesternmarginoftheCentralAfricanrainforest. C.R.Geosci. 345,327–335.

Lu,Z.,Liu,Z.,Zhu,J.,Cobb,K.M.,2018.AreviewofpaleoElNiño-Southern Oscilla-tion.Atmosphere 9,130.

Lupo,K.D.,Schmitt,D.N.,Kiahtipes,C.A.,Ndanga,J.P.,Young,D.C.,Simiti,B.,2015. OnintensivelateHoloceneIronminingandproductionintheNorthernCongo Basinandtheenvironmentalconsequencesassociatedwithmetallurgyin Cen-tralAfrica.PLoSONE.https://doi.org/10.1371/journal.pone.0132632.

Macreadie,P.I.,Rolph,T.C.,Boyd,R.,Schröder-Adams,C.J.,Skilbeck,C.G.,2015.Do ENSOand coastal developmentenhance coastal burialofterrestrialcarbon? PLoSONE 10,0145136.

MaherJr.,L.J., 1972.Nomograms for computing0.95confidencelimits ofpollen data.Rev.Palaeobot.Palynol. 13,85–93.

Maley,J.,2001.Elaeisguineensis Jacq.(oilpalm)fluctuationsincentralAfricaduring thelateHolocene:climateorhumandrivingforcesforthispioneeringspecies? Veg.Hist.Archaeobot. 10,117–120.

Maley,J.,2002.AcatastrophicdestructionofAfricanforestsabout2,500yearsago stillexertsamajorinfluenceon presentvegetationformations. IDSBull. 33, 13–30.

Maley,J.,Giresse, P.,Doumengue,C., Favier,C.,2012. Commenton “Intensifying weatheringandlanduseinIronAgeCentralAfrica”.Science 337,1040.

Maley,J.,Doumenge,C.,Giresse,P.,Mahé,G.,Philippon,N., Hubau,W.,Lokonda, M.O.,Tshibamba,J.M.,Chepstow-Lusty,A.,2018.LateHoloceneforest contrac-tionandfragmentationincentralAfrica.Quat.Res. 89,43–59.

Mbuebue,J.M.T.,Muliwavyo,A.M.,Mwamba,V.L.,Phuati,E.P.,Bantu,A.K.M.,Keto, F.T.,2016.Time-scalecharacteristicsofKasairiverhydrologicalregime variabil-ityfor1940–1999.arXiv:1611.07317.

Morin-Rivat,J.,Fayolle,A.,Gillet,J.F.,Bourland,N.,Gourlet-Fleury,S.,Oslisly,R., Bre-mond,L.,Bentaleb,I.,Beeckman,H.,Doucet,J.L.,2014.Newevidenceofhuman activities duringthe Holoceneinthe lowlandforests ofthe northernCongo basin.Radiocarbon 56,209–220.

Neumann,K.,etal.,2012.Commenton“Intensifyingweatheringandlandusein IronAgeCentralAfrica”.Science 337,1040.

Ngomanda,A.,Neumann,K.,Schweizer,A.,Maley,J.,2009.Seasonalitychangeand thethirdmillenniumBPrainforestcrisisinsouthernCameroon(CentralAfrica). Quat.Res. 71,307–318.

Nicholson,S.E.,Kim,J.,1997.TherelationshipoftheElNiño-SouthernOscillationto Africanrainfall.Int.J.Climatol. 17,117–135.

Oslisly,R.,White,L.,Bentaleb,I.,Favier,C.,Fontugne,M.,Gillet,J.F.,Sebag,D.,2013. Climaticand culturalchangesintheWestCongobasin forestsoverthe past 5000years.Philos.Trans.R.Soc.Lond.B,Biol.Sci. 368,20120304.

Phillipson,D.W.,2005.AfricanArchaeology,3rdedition.CambridgeUniversityPress, Cambridge.

Preethi,B.,Sabin,T.P.,Adedoyin,J.A.,Ashok,K.,2015.ImpactsoftheENSOModoki andothertropicalIndo-Pacificclimate-driversonAfricanRainfall.Sci.Rep. 5.

https://doi.org/10.1038/srep16653.

Ropelewski,C.F.,Halpert, M.S.,1987.Global andregionalscaleprecipitation pat-ternsassociatedwiththeElNiño/SouthernOscillation.Mon.WeatherRev. 115, 1606–1626.

Sachse, D., et al., 2012. Molecular paleohydrology: interpreting the hydrogen-isotopic composition oflipid biomarkers from photosynthesizing organisms. Annu.Rev.EarthPlanet.Sci. 40,221–249.

Schefuß,E.,Schouten,S.,Schneider,R.R.,2005.ClimaticcontrolsoncentralAfrican hydrologyduringthepast20,000years.Nature 437,1003–1006.

Schefuß,E.,Eglinton,T.I.,Spencer-Jones,C.L.,Rullkötter,J.,DePol-Holz,R.,Talbot, H.M.,Grootes,P.M.,Schneider,R.R.,2016.Hydrologiccontrolofcarboncycling andagedcarbondischargeintheCongoRiverbasin.Nat.Geosci. 9,687–690. Schneider,R.R.,Price,B.,Muller,P.,Kroon,D.,Alexander,I.,1997.Monsoon-related

variationsinZaire(Congo)sedimentloadandinfluenceoffluvialsilicatesupply onmarineproductivityintheeastequatorialAtlanticduringthelast200,000 years.Paleoceanography 12,463–481.

Singarayer,J.S.,Burrough,S.L.,2015.InterhemisphericdynamicsoftheAfrican rain-beltduringthelateQuaternary.Quat.Sci.Rev. 124,48–67.

Sowunmi,M.A.,1999.Thesignificanceoftheoilpalm(Elaeisguineensis Jacq.)inthe lateHoloceneenvironmentsofwestandwestcentralAfrica:afurther consider-ation.Veg.Hist.Archaeobot. 8,199–210.

Stuiver, M.,Polach,H.A.,1977. Discussionreportingof14_{Cdata. Radiocarbon 19,} 355–363.

Stuiver,M.,Reimer,P.J.,Reimer,R.W.,2017.CALIB7.1.http://calib.qub.ac.uk/calib/.

Tanaka,T.,etal.,2000.JNdi-1:aneodymiumisotopicreferenceinconsistencywith LaJollaneodymium.Chem.Geol. 168,179–181.

Thiéblemont,D., Flehoc,C.,Ebang-Obiang,M., Rigollet,C., Prian,J.P.,Prognon,F., 2013.Geochronologicalargumentsforacloserelationshipbetweensurficial for-mationprofilesandenvironmentalcrisis(c.3000–2000BP)inGabon(Central Africa).C.R.Geosci. 345,272–283.

Thiéblemont, D., Guerrot, C., Négrel, P., Braucher, R., Bourlès, D.L., Thieblemont, R., 2014. Nd-isotope evidence for the distal provenance of the historical (c. < 3000 BP)lateriticsurfacecoverunderlyingtheEquatorialforestinGabon (WesternAfrica).Aeol.Res. 15,177–192.

Tierney,J.E.,Russell,J.M.,SinningheDamsté,J.S.,Huang,Y.,Verschuren,D.,2011. LateQuaternarybehavioroftheEastAfricanmonsoonandtheimportanceof theCongoAirBoundary.Q.Sci.Res. 30,798–807.

Vincens,A.,etal.,1998.Biomereconstructionfrompollenand plantmacrofossil dataforAfricaandtheArabianpeninsulaat0and6000years.J.Biogeogr. 25, 1007–1027.

Vincens,A.,Lézine,A.M.,Buchet,G.,Lewden,D.,LeThomas,A.,2007.Africanpollen databaseinventoryoftreeandshrubpollentypes.Rev.Palaeobot.Palynol. 145, 135–141.

Willis,K.J., Gillson,L., Brncic,T.M., 2004.How“virgin” isvirginrainforest? Sci-ence 304,402–403.

Winterfeld,M.,Mollenhauer,G.,Dummann,W.,Köhler,P.,Lembke-Jene,L.,Meyer, V.D., Hefter, J., McIntyre, C., Wacker, L., Kokfelt, U., Tiedemann, R., 2018. Deglacialmobilizationofpre-agedterrestrialcarbonfromdegradingpermafrost. Nat.Commun. 9,3666.