Late Oligocene–early Miocene birth of theTaklimakan DesertHongbo Zhenga,b,1, Xiaochun Weic,1, Ryuji Tadad, Peter D. Clifte, Bin Wangf, Fred Jourdang, Ping Wanga,and Mengying Hea

aSchool of Geography Science, Nanjing Normal University, Nanjing 210023, China; bCenter for Excellence in Tibetan Plateau Earth Sciences, ChineseAcademy of Sciences, Beijing, 100101, China; cSchool of Earth Sciences and Engineering, Nanjing University, Nanjing 210023, China; dDepartment of Earthand Planetary Science, University of Tokyo, Tokyo 113-0033, Japan; eDepartment of Geology and Geophysics, Louisiana State University, Baton Rouge, LA70803; fTourism and Environment College, Shaanxi Normal University, Xi’an 710119, China; and gWestern Australian Argon Isotope Facility, Department ofApplied Geology and John de Laeter Centre, Curtin University, Perth, WA 6845, Australia

Edited by Zhonghe Zhou, Chinese Academy of Sciences, Beijing, China, and approved May 11, 2015 (received for review December 22, 2014)

As the world’s second largest sand sea and one of the most im-portant dust sources to the global aerosol system, the formationof the Taklimakan Desert marks a major environmental event incentral Asia during the Cenozoic. Determining when and how thedesert formed holds the key to better understanding the tectonic–climatic linkage in this critical region. However, the age of theTaklimakan remains controversial, with the dominant view beingfrom ∼3.4 Ma to ∼7 Ma based on magnetostratigraphy of sedi-mentary sequences within and along the margins of the desert. Inthis study, we applied radioisotopic methods to precisely date avolcanic tuff preserved in the stratigraphy. We constrained theinitial desertification to be late Oligocene to early Miocene, be-tween ∼26.7 Ma and 22.6 Ma. We suggest that the TaklimakanDesert was formed as a response to a combination of widespreadregional aridification and increased erosion in the surroundingmountain fronts, both of which are closely linked to the tectonicuplift of the Tibetan–Pamir Plateau and Tian Shan, which had re-ached a climatically sensitive threshold at this time.

Taklimakan Desert | late Oligocene | early Miocene | tectonic uplift |desertification

Surrounded by the Tian Shan to the north, the Pamir to thewest, and the West Kunlun to the south, the Taklimakan

Desert is the world’s second largest sand sea (Fig. 1). Located farfrom any major source of moisture, and shadowed by Tibetanand central Asian mountain ranges, the Taklimakan is deprivedof rainfall, with mean annual precipitation not exceeding 50 mm(1). Provenance studies suggest that mineral dust from theTaklimakan Desert contributes substantially to the global aero-sol system, allowing it to play a significant role in modulatingglobal climate on various time scales (2, 3). The formation of theTaklimakan Desert therefore marked a major environmental eventin central Asia during the Cenozoic, with far-reaching impacts.Furthermore, determining when and how the desert formed holdsthe key to better understanding the nature of tectonic–climaticlinkage in this critical region. However, the time at which theTaklimakan Desert came into existence has been strongly debated,with estimates ranging from only a few hundreds of thousands to afew million years ago (1, 4–8).In the context of this study, we define desertification to rep-

resent not only the formation of a significant sand sea but alsothe generation of a dynamic eolian system that supplied volu-minous mineral dust on regional and even global scales. Evi-dence of desertification in the geological past is preserved insedimentary sequences within, and along the margins of, thepresent-day Taklimakan Desert that lies within the Tarim Basin(Fig. 1). Recent geochronological work, mostly based on themagnetostratigraphy of these sedimentary sequences, proposed∼3.4 Ma to 7 Ma for the initiation of the Taklimakan (4–8).Precise dating of these terrestrial rocks has largely been ham-pered by lack of dateable material. In this study, we reporta newly identified volcanic tuff from two sedimentary sections

along the southwestern margin of the Tarim Basin. Radioiso-topic dating of the volcanic minerals provides a robust age forthe sections, and therefore we are able to determine the age ofthe Taklimakan Desert more precisely.

Geological Setting and LithostratigraphyFrom latest Cretaceous to early Paleogene, much of the westernTarim Basin was influenced episodically by a shallow sea, whichwas connected to the Paratethys, an epicontinental marine sea-way covering large parts of Europe and southern central Asia(9). Shallow marine strata of this age are observed extensivelyalong the western and southwestern margins of the Tarim Basin,extending to Hotan to the east and Ulugqat to the north (Fig. 1).Five major marine incursions have been recognized in the broadregion, with the fourth being the final one at Aertashi, where it isrepresented by the Wulagen Formation (Fig. 2 A and B and SIText). Recent biostratigraphic work at Aertashi has suggested thatthe Paratethys Sea finally retreated from this locality at ∼41 Ma (9,10), a process that has been attributed to global eustasy coupledwith tectonism associated with the Indo–Asia collision. Together,these have amplified the aridification of the Asian interior (9, 11).Syntectonic foreland basin deposition followed the marine

regression immediately and led to accumulation of thick conti-nental red beds (Fig. 2 A and B). The Bushiblake Formation ischaracterized by red, fine-gained mudstone and fine sandstone, withcommon evaporative gypsum, typical of low-energy, meanderingriver, lacustrine, and playa deposits (SI Text). The overlyingWuqia Group (SI Text) consists of thickly bedded sandstones with

Significance

The formation of the Taklimakan Desert marked a major geo-logical event in central Asia during the Cenozoic, with far-reaching impacts. Deposition of both eolian sand dunes in thebasin center and the genetically equivalent loessite along thebasin margins provide evidence for the birth of the TaklimakanDesert. This paper resolves a long-standing debate concerningthe age of the Taklimakan Desert, specifically whether it datesto ∼3.4–7 Ma, currently the dominant view. Our result showsthat the desert came into existence during late Oligocene–earlyMiocene, between ∼26.7 Ma and 22.6 Ma, as a result of wide-spread regional aridification and increased erosion in the sur-rounding mountain fronts, both of which are closely linked tothe tectonic uplift of the Tibetan–Pamir Plateau and Tian Shan.

Author contributions: H.Z. designed research; H.Z., X.W., R.T., P.D.C., B.W., P.W., and M.H.performed research; X.W. and F.J. analyzed data; and H.Z., X.W., and F.J. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence may be addressed. Email: zhenghb@njnu.edu.cn or xcwnju@gmail.com.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1424487112/-/DCSupplemental.

7662–7667 | PNAS | June 23, 2015 | vol. 112 | no. 25 www.pnas.org/cgi/doi/10.1073/pnas.1424487112

Dow

nloa

ded

by g

uest

on

June

25,

202

0

minor mudstone. The overall content of sand increases upward, asdoes the grain size.Lithofacies change to distal alluvial deposits, consisting of con-

glomerate, sandstone, and siltstone crossing the boundary from theWuqia Group to the Artux Formation (SI Text). The conglomeratelayers in the Artux Formation are thin-bedded debris flow deposits,containing medium-sized, angular to subrounded pebbles, of whichmore than half are sedimentary rocks. The overlying Xiyu For-mation (SI Text) is up to 3 km thick and consists of massiveboulder to cobble-grade conglomerate with increasing volumesof igneous and high-grade metamorphic clasts. The Xiyu For-mation is typical of proximal diluvial fan deposits derived fromunroofed mountain belts. Red beds passing upward into upward-coarsening conglomerate and debris flow deposits recorded thechange in paleoslope and sediment supply related to uplift of thenorthern margin of the Tibetan Plateau and Pamir.Massive siltstone lenses intercalated in the Artux and Xiyu

Formations at Kekeya and many other localities are particularlynoteworthy (Fig. 2 C and D). Previous sedimentological studies,including facies investigations and grain size (Fig. S1 A and B)and geochemical analyses, suggest that these siltstone lenses areeolian loessite, having been sourced from the desert, deposited,and preserved on an intermittent diluvial fan system (4), a pro-cess resembling that in the Taklimakan Desert today.The modern Taklimakan Desert is surrounded by a series of

giant diluvial fans that link the uplifting mountain chains with theTarim Basin. Sediments shed off the mountain fronts have beeneroded, mostly but not exclusively, by landsliding and glaciation.They are then delivered down the slope to the basin through thefan systems, and sorted by eolian and fluvial processes into siltand sand fractions, which become the constituents of loess anddesert, respectively. These processes of dust and sand productionmust have been in operation since the deposition of the Artuxand Xiyu formations. We therefore conclude that widespreadaccumulation of loessite in the Artux and Xiyu formations alongthe margins of the Taklimakan Desert strongly suggests that the

source area had become fully arid and desertified, and wassupplying dust to the proximal region at that time.Direct evidence of desertification is found in the exposed

sedimentary sequence at the Mazatagh Range, in the centralTaklimakan Desert (Figs. 1 and 2E). The lithostratigraphy at thislocation has been well established and can be generally corre-lated to the basin margin sequences (Fig. 2G) (5, 6), except thatthe lithologies are generally finer and sedimentation rates arelower. Red cross-bedded eolian sandstone (Fig. 2H and Figs. S1C and D and S2) is present in the red bed unit as isolatedsandstone lenses, likely indicating an initial stage of a desertifiedenvironment that was composed of low-energy fluvial, playalakes and associated lunette dunes. The most prominent sedi-mentary unit in the Mazatagh sequence is the well-developed,cross-bedded, yellowish sandstone (Fig. 2F). It is about 200 mthick and stratigraphically continuous and is interpreted as typ-ical desert sand dune deposits.Deposition of both eolian sand dunes in the basin center and

the genetically equivalent loessite along the basin margins pro-vides two lines of evidence to suggest that the Taklimakan Desertcame into existence around that time.We have identified volcanic ash in the Aertashi and Kekeya

sections (Fig. 2C, Fig. S3, and SI Text), where it is intercalated inthe Xiyu Formation, the basal age of which has previously beendetermined by magnetostratigraphy to be of Plio-Pleistocene age(12). In this study, 40Ar/39Ar and U–Pb dating of the volcanic ashhas provided, for the first time, to our knowledge, an anchorpoint to better constrain the strata, and therefore the age ofdesertification. Petrologic investigation showed that the com-position of the ash includes mainly vitroclastic, sanidine, aegir-ine–augite, and biotite, typical of alkaline magmatic rocks (Fig.S3). These volcanic minerals are ideal for radioisotopic dating.In addition, we also carried out field investigations together withpetrologic studies and found that the volcanic activity associatedwith the Tashkorgan alkaline complex (13) is most likely re-sponsible for providing the Xiyu Formation ash.

Yark

and

Rive

r

Kara

kax

Rive

r

Keriy

a R

iver

Yuru

ngka

x Rive

rH

otan

Riv

er

Kirzlesu River

Yecheng

Mazar

Kashgar

Ulugqat

Kalpin

Muji

Mustagh

Tashkorgan

Hotan

Tarim Basin

Tibetan Pleteau

Taklimakan Desert

Karakorum Fault Karakax Fault

Main Pamir Thrust

West Kunlun Shan

Pamir

Tian Shan

Mazatagh

Tibetan Plateau

India

Pamir

Tian Shan

CLPQA

TB

JB

Gobi & DesertLoess

Nomal fault

Thrust fault

Stike-slip fault

Town

Section

0m

3000m1000m

5000m8524m

N

Aertashi

Kekeya

0km 200km100km

76 78 80

36

38

40

BA

74

Fig. 1. Location map. (A) Topographic map showing the western portion of the Taklimakan Desert (Tarim Basin) and the surrounding mountain ranges withmajor active faults. The location of the studied Cenozoic sedimentary sections at Aertashi, Kekeya, and Mazatagh are shown by stars. (B) Map showing thelocation of the Tarim Basin (TB), Junggar Basin (JB), and Chinese Loess Plateau (CLP). Qin’an loess section (QA) is marked by a red star.

Zheng et al. PNAS | June 23, 2015 | vol. 112 | no. 25 | 7663

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

Dow

nloa

ded

by g

uest

on

June

25,

202

0

40Ar/39Ar Dating of Biotite from Volcanic AshWe selected three samples (YC10-19, AT10-34, and AT10-35) fromtwo locations (Kekeya section and Aertashi section) for 40Ar/39Ardating and separated unaltered, 200- to 800-μm-size biotite. These

minerals were separated using a Frantz magnetic separator, andthen carefully handpicked under a binocular microscope. The se-lected biotite crystals of sample AT10-35 were classified accordingto their color, with green, red, and brown populations.

Mudstone

Siltstone

Sandstone

Conglomerate

Marl

Gypsum

Eolian sand

Tuff

Shell

Eolian loess

Dis

tal d

iluvi

al fa

nP

roxi

mal

dilu

vial

fan

Del

ta +

pla

ya la

ke +

fluv

ial

Kas

hi G

.B

ashi

bula

k F.

Wuq

ia G

roup

Atu

x Fo

rmat

ion

Xiy

u Fo

rmat

ion

Mar

ine

km

1

2

3

4

0

Kekeya

Ash ~11 Ma

~26.7 or 22.6 Ma

Conglo-merate

Loess

Conglome-rate

Sandstone

Loess

MudstoneSandstoneGypsum

Mudstones

Sandstone

MarlMudstoneGypsum

40

35

30

25

20

15

10

GTS

2012

Ma

D

Mazatagh

Fluv

ial +

pla

yaW

uqia

Gro

upK

ashi

GA

tux

Form

atio

nM

arin

e D

eser

t +flu

vial

1

2km

0

MudstoneRed eolian sand

Yellowish eolian sand

~26 MaMudstoneSandstoneGypsumMarlMudstoneGypsum

ConglomerateYellowish-grey sand, not well exposed

35

GTS

2012

Ma

30

25

20

15

G

1

2

3

4

0

5

6

7km

D

elta

+ p

laya

lake

+ fl

uvia

lM

arin

enoita

mroFuy i

Xnoita

mroFxut

Apuo r

Gaiq u

W.F

kal ubihsaB

.Gihs a

K B

raid

ed +

Mea

nder

ing

river

Dis

tal a

lluvi

al fa

nP

roxi

mal

allu

vial

fan

Aertashi

MudstoneSandstoneGypsum

Final sea retreat~41 Ma

Conglom-erate

Sandstone

Conglo-merate

Mudstone

Sandstone

Sandstone

Mudstone

Brown sandstone

MarlMudstoneGypsum

40

35

30

25

20

15

10

GT

S201

2 (re

f. 19

)

Ma

A

Ash ~11 Ma

Yellowish eolian sandstone

F

Red eolian sandstone

H

Mazatagh range

Taklimakan Desert

Red eolian sandstone

Yellowish eolian sandstone

EE

Gypsum

Marl + limestone + sandstoneMudstone + sandstone

ContinentalB MarineAertashi Kashi G.

Loess

Volcanic ash

Conglomerate

C Kekeya Xiyu F.

Mazatagh

Mazatagh

Fig. 2. Stratigraphy and magnetostratigraphy of Cenozoic sedimentary sequences within and along the southwestern margin of the Tarim Basin (Takli-makan Desert). The magnetostratigraphy of all sections was correlated to GTS2012 (19). (A) Stratigraphy and magnetostratigraphy of the Aertashi sectionbased on age control from volcanic ash and paleontology (9). (B) Shallow marine deposits and continental red beds at Aertashi. (C) Volcanic ash and loessiteintercalated in Xiyu conglomerate at Kekeya section. (D) Stratigraphy and revised magnetostratigraphy for the Kekeya section based on age control of thevolcanic ash combined with data from ref. 12. The solid line and dotted line indicate two alternative magnetostratigraphic correlations for this section(see SI Text for more information). (E ) Exposed Mazatagh section. (F ) Yellowish desert sandstone at Mazatagh section. (G) Stratigraphy and revisedmagnetostratigraphy based on ref. 6 in the context of regional stratigraphic correlation. (H) Red eolian sandstone within the red beds at Mazatagh. SeeFig. 1 for locations.

7664 | www.pnas.org/cgi/doi/10.1073/pnas.1424487112 Zheng et al.

Dow

nloa

ded

by g

uest

on

June

25,

202

0

The 40Ar/39Ar analyses were performed at the Western Aus-tralian Argon Isotope Facility at Curtin University. Due to thepossibility of crystal xenocrysts in the volcanic tuff layers, eachbiotite crystal was fused in one step using a 110-W Spectron LaserSystem, with a continuous laser rastered over the sample during1 min to ensure complete melting (see SI Text for more detail).Ar isotopic data corrected for blank, mass discrimination andradioactive decay are given in Datasets S1–S3. Individual errorsin Datasets S1–S3 are given at the 1σ level. Since the crystals in asingle tuff can yield vastly different ages and we did not step heatthe samples, we did not apply the standard plateau age criteria.Rather, we looked for age convergence of the youngest pop-ulation with a minimum of four grains yielding indistinguishableage. This procedure is similar to the approach used for individualU–Pb zircon age to calculate a magmatic age. All sources of un-certainties are included in the final age proposed for a given sample.Eighteen crystals were analyzed for sample YC10-19. Fig. S4A

shows a continuum of age from ∼11 Ma to 13 Ma. The sixyoungest ages yielded a concordant age population with aweighted average age of 11.17 ± 0.15 Ma (mean square weighteddeviation (MSWD) = 1.2, possibility (P) = 0.33), interpreted asthe age of the youngest eruption bearing the crystals. Those re-sults also indicate the occurrence of older eruptions, possibly ofup to 13 Ma, whose crystals have been entrained during theyoungest eruption at ∼11.2 Ma; however, since this could be dueto minute excess 40Ar* incorporated in the crystal, we refrainfrom making assumptions about the magmatic activity for thethree samples. For sample AT10-34 (from the upper layer), weanalyzed 14 biotite crystals. This sample includes two distinct agegroups, with the oldest group (n = 7) having ages ranging from120 Ma to 180 Ma and whose significance is beyond the scopeof this paper. The youngest group (n = 7) shows a range of age(Fig. S4B), with the four youngest grains giving a concordant agepopulation with a weighted mean age of 10.94 ± 0.20 Ma(MSWD = 0.03, P = 0.99) (Fig. S4C). This age is interpreted asthe eruption age of the ash layer from which AT10-34 is derived.The last sample, AT10-35, consists of 21 crystals, and 40Ar/39Arresults show that this sample also contains crystals from mucholder events (Fig. S4D). Apparent ages of the older group (allred or brown crystals) range from 114 Ma to 214 Ma. Fivecrystals from this group yield a homogenous age population witha weighted mean age of 189.9 ± 1.4 Ma (MSWD = 1.4, P = 0.24)likely to date a Jurassic magmatic event in the region. Theyoungest population (n = 13) yields a range of ages from 17 Mato 10 Ma, suggesting episodic activity during this period. The 10youngest grains belong mostly to the green group and yield aconcordant age population with a weighted mean age of 11.49 ±0.34 Ma (MSWD = 0.70, P = 0.71) (Fig. S4E). However, itshould be noted that most of these apparent ages have large ageuncertainties due to the fact that the young population consistsmostly of small grains that yielded a small argon beam signal.The age of 11.49 Ma is interpreted as the age of the tuff erup-tion. All three tuff ages are within 0.5 Ma of each other, sug-gesting that explosive volcanism occurred in this region over awide area at ∼11 Ma. Since the three ages are not concordant(MSWD = 4.2), this suggests that the volcanic activity lasted fora duration of at least several hundred thousands years.

U–Pb Dating of Zircon from Volcanic AshSample YC12-13-5 was taken from Kekeya section. It was theonly sample in this study that contained suitable zircon grainsfor U–Pb dating. Zircon grains were separated using traditionalflotation method, and then carefully handpicked under a bin-ocular microscope to obtain optically clear, colorless grains.Together with zircon standard Plésovice (337 Ma) (14), thezircon grains from sample YC12-13-5 were mounted onto a2.4-cm-diameter epoxy disk, ground, and then polished for anal-ysis. To reveal the internal structure, all zircons were imaged using

transmitted, reflected light, as well as cathodeluminescence. Themount was vacuum-coated with high-purity gold before analysiswith secondary ion mass spectrometry (SIMS). U, Th, and Pbwere measured using a Cameca IMS-1280 SIMS at the Instituteof Geology and Geophysics, Chinese Academy of Sciences, inBeijing. Methods used to obtain the U–Pb analytical data weresimilar to those developed by Li et al. (15). The standard zirconPlésovice (14) was used to determine U–Pb ratios and absoluteabundances. The sample was analyzed in a continuous analyticalsession with standards interspersed with every four to five un-knowns. Corrections for common Pb in this paper were madeusing measured 204Pb and an age-appropriate Pb isotopic com-position of Stacey and Kramers (16). Data processing was carriedout using the Isoplot 4.15 program (17).Twenty-five spots of 25 zircon grains from sample YC12-13-5

were measured during a single analytical session, and the dataare reported in Dataset S4, with individual errors given at 1σlevel. The analysis results showed relatively high Th/U ratios(0.391–1.903), with U contents ranging from 248 ppm to 2619ppm. Common Pb was low for the majority of analyses, with f206values (the proportion of common 206Pb in total measured 206Pb)lower than 4.23%, apart from spot 15, which yielded a value of28.89%. Since the ages of zircons are quite young, we looked for206Pb/238U age convergence of the youngest population as the tufferuption age. Twenty out of 25 zircon grains yielded a homog-enous age population with a weighted mean age of 11.18 ± 0.11 Ma(95% confidence interval, MSWD = 1.18, P = 0.27; Fig. S5A),while the other five, which could be xenocrystic zircons, pro-duced 12.4 ± 0.2 Ma, 16.3 ± 0.6 Ma, 28.7 ± 0.6 Ma, 207.0 ± 3.1 Ma,

Eolian quartz (% by weight)

Detrital contribution (%)

Eolian quartz

Detrital

0

10

20

30

40

Mio

cene

Eoc

ene

Pli.

Plt.(Ma)

Ons

et o

f C

hine

se lo

ess

50 10 15 20

0 60 80 1004020

Olig

ocen

e

Ons

et o

f des

ertif

icat

ion

ofA

sian

inte

rior

Fig. 3. Detrital contribution (blue dots) of core from Site 1215 and aeolianquartz percentage (green dots) of core LL44-GPC3 from the North Pacificsince the Eocene (22, 24). The arrows on the right side indicate the onset ofeolian loess at the Chinese Loess Plateau (20, 21) and desertification of Asianinterior obtained by this study.

Zheng et al. PNAS | June 23, 2015 | vol. 112 | no. 25 | 7665

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

Dow

nloa

ded

by g

uest

on

June

25,

202

0

and 707.6 Ma ± 10.3 Ma (spots 20, 17, 08, 10, and 07, respec-tively; Fig. S6 and Dataset S4). Alternatively, to avoid the in-fluence of common Pb, 14 ages (spots 01, 02, 03, 05, 06, 09, 11,12, 13, 16, 18, 23, 24, and 25; Fig. S6 and Dataset S4) with mea-sured 206Pb/204Pb > 1,000 were calculated, giving almost thesame weighted mean age of 11.18 ± 0.13 Ma (95% confidenceinterval, MSWD = 1.2, P = 0.26; Fig. S5B). The age of 11.18 Mais interpreted as the eruption age of the tuff layer from whichYC12-13-5 is derived.The 206Pb/238U age of 11.18 Ma of sample YC12-13-5 is very

close to the 40Ar/39Ar age of biotite (11.17 Ma) of sample YC10-19. The age difference could have resulted from different erup-tions. However, we believe that this minor discrepancy is morelikely to have been caused by the difference of closure tempera-tures of the two types of minerals or age uncertainties. We there-fore conclude that the volcanic ash was deposited soon after theeruption that occurred at ∼11 Ma in this region.

Birth of the Taklimakan Desert: When and How?Constrained by the volcanic ash and regional lithostratigraphiccorrelations, we are able to reconstruct a magnetostratigraphyfor the Cenozoic sequence based on previously published datafrom Kekeya (12) and Mazatagh (6), as well as new measure-ments from Aertashi (Fig. 2 A, D, and G). All paleomagneticsamples were progressively demagnetized from room tem-perature up to 600–695 °C in 5–100 °C steps. Remanent mag-netizations were measured using a 2G 755-R SuperconductingRock Magnetometer. The characteristic remnant magnetiza-tion (ChRM) directions were assessed on an orthogonal de-magnetization diagram and calculated by application of principalcomponent analysis to determine the direction of the best least-squares line fit (18) (Fig. S7 and SI Text).The Aertashi section comprises a relatively complete succes-

sion of Cenozoic sediments. More importantly, the section is wellconstrained by the volcanic ash in the upper part and bio-stratigraphy at the base (10), and therefore can be used as a ref-erence for regional correlations. The pattern of magnetic polarityzones of Aertashi can be correlated to the Geologic Time Scale2012 (GTS2012) (19) (Fig. S8, SI Text, and Dataset S5). Twodepositional hiatuses occurred when lithofacies changed to mas-sive sandstone and conglomerate, respectively. Geomagnetic cor-relations to GTS2012 of the Kekeya (Fig. S9, SI Text, and DatasetS6) and Mazatagh (Fig. S10 and SI Text) sections, based onstratigraphic correlations to the Aertashi section, yielded ageranges from 37.5 Ma to 10 Ma and 34 Ma to 17.5 Ma, respectively.Of greatest significance is the timing of the transition from flu-

vial deposit to diluvial fan debris flow with eolian silts at Kekeya,and red eolian dunes at Mazatagh, which is, on our updatedmagnetostratigraphy, dated to be ∼26.7–22.6 Ma (Figs. S9 and S10and SI Text). We therefore argue that, by late Oligocene to earlyMiocene time, the Tarim Basin, surrounded by a rising Tibetan–Pamir Plateau and Tian Shan, had become fully arid and deserti-fied, supplying dust to the mountain fronts, where it accumulatedas loess. Desertification of the Asian interior has had far-reachingimpacts on regional and even global scales. Provenance studies of

modern mineral dust suggest that the Taklimakan Desert is one ofthe major dust source areas to the global aerosol system (2, 3), andmight have been so since the birth of the desert. In this regard, it isworth noting that the onset of loess deposition at Qin’an in theChinese Loess Plateau (20, 21) (see Fig. 1B for location), andchange in the dust to the distal north Pacific Ocean (22–24) asshown by a significant increase in the detrital contribution andquartz content, occurred at about the same time (Fig. 3).Formation of a desert over tectonic time scales requires two

integral prerequisite conditions: an arid climate and a sufficientsupply of fine-grained sediments. During much of the Paleogene,a roughly zonal region of China between 30°N and 50°N was underan arid climate, primarily controlled by the northern westerlies(25). The Paleogene aridity of the Tarim Basin, which was part ofthis arid region, is registered by widespread accumulation of redbeds and intercalated evaporite. However, the climatic configu-ration changed dramatically during the period from late Oligoceneto early Miocene, after which the Asian monsoon set in, prevailingin southeastern China, and the arid region with greatest severityretreated to the northwest.Indo–Asia collision likely starting at ∼60 Ma resulted in the

progressive uplift of the Tibetan Plateau and other mountainranges in Central Asia (26–29). These may have grown to heightsof more than 3 km by the late Oligocene (30, 31), even if theextent of uplift has since continued to grow. Generation of suchtopography forced major climatic changes during the Cenozoic(32–35). Numerous studies suggested that significant uplift of theTibetan–Pamir Plateau and the Tian Shan occurred around thelate Oligocene–early Miocene (36–39). The uplifted plateau to-gether with the possible retreat of the Paratethys (9, 11) forcedby uplifting topography could have led to a transition from plan-etary climate system to monsoonal climate system (35, 40, 41). Atthe same time, the arid zone, with much severity, retreated to theAsian interior (25). Equally, if not more, important, is the uplift-induced erosion and weathering of bedrock, which supplied greatvolumes of sediments to the mountain fronts in the form of al-luvial and diluvial fans. These sediments would then have beensorted by fluvial and eolian processes into sand and silt fractions,with the former becoming the constituents of a dynamic desertsystem and the latter being deflated and transported as eolian dust(4). We suggest that this mechanism has been in operation sincethe late Oligocene–early Miocene time, and that the resultantformation of the Taklimakan Desert was a direct response to acombination of widespread regional aridification and increasedunroofing and erosion in the surrounding mountains, both ofwhich are closely linked to the uplift of the Tibetan–Pamir Pla-teau and the Tian Shan (36–39).

ACKNOWLEDGMENTS. The authors are grateful to the reviewers for theirconstructive comments. H.Z. is indebted to the late Prof. C. Powell and thelate Prof. X. Wu for their contributions in the field work and analysis of sedi-mentological data. This work was supported by the Strategic Priority ResearchProgram of the Chinese Academy of Sciences (XDB03020300), the NationalScience Foundation of China (NSFC 40025207, 90211019, and 41021002),and the Priority Academic Program Development of Jiangsu Higher EducationInstitutions.

1. Zhu ZD, Wu Z, Liu S, Di X (1980) An Outline of Chinese Deserts (Science Press, Beijing).2. Uno I, et al. (2009) Asian dust transported one full circuit around the globe. Nat

Geosci 2(8):557–560.3. Shao YP, et al. (2011) Dust cycle: An emerging core theme in Earth system science.

Aeolian Res 2(4):181–204.4. Zheng HB, Powell CM, Butcher K, Cao JJ (2003) Late Neogene loess deposition in

southern Tarim Basin: Tectonic and palaeoenvironmental implications. Tectonophy-sics 375(1-4):49–59.

5. Sun J, Liu T (2006) The age of the Taklimakan Desert. Science 312(5780):1621.6. Sun JM, Zhang ZQ, Zhang LY (2009) New evidence on the age of the Taklimakan

Desert. Geology 37(2):159–162.7. Sun DH, et al. (2011) Palaeomagnetic and palaeoenvironmental study of two parallel

sections of late Cenozoic strata in the central Taklimakan Desert: Implications for thedesertification of the Tarim Basin. Palaeogeogr Palaeoclimatol Palaeoecol 300(1-4):1–10.

8. Liu W, et al. (2014) Late Miocene episodic lakes in the arid Tarim Basin, western China.Proc Natl Acad Sci USA 111(46):16292–16296.

9. Bosboom R, et al. (2014) Timing, cause and impact of the late Eocene stepwise searetreat from the Tarim Basin (west China). Palaeogeogr Palaeoclimatol Palaeoecol403:101–118.

10. Bosboom R, et al. (2014) Linking Tarim Basin sea retreat (west China) and Asian ari-dification in the late Eocene. Basin Res 26(5):621–640.

11. Ramstein G, Fluteau F, Besse J, Joussaume S (1997) Effect of orogeny, plate motionand land-sea distribution on Eurasian climate change over the past 30 million years.Nature 386(6627):788–795.

12. Zheng HB, Powell CM, An ZS, Zhou J, Dong GR (2000) Pliocene uplift of the northernTibetan Plateau. Geology 28(8):715–718.

13. Ke S, et al. (2006) Petrogenesis and geochemistry of Cenozoic Taxkorgan alkaliccomplex and its geological significance. Acta Petrol Sin 22(4):905–915.

7666 | www.pnas.org/cgi/doi/10.1073/pnas.1424487112 Zheng et al.

Dow

nloa

ded

by g

uest

on

June

25,

202

0

14. Slama J, et al. (2008) Plesovice zircon - A new natural reference material for U-Pb andHf isotopic microanalysis. Chem Geol 249(1-2):1–35.

15. Li XH, Liu Y, Li QL, Guo CH, Chamberlain KR (2009) Precise determination of Phan-erozoic zircon Pb/Pb age by multicollector SIMS without external standardization.Geochem Geophys Geosyst 10(4):Q04010.

16. Stacey JS, Kramers JD (1975) Approximation of terrestrial lead isotope evolution by atwo-stage model. Earth Planet Sci Lett 26(2):207–221.

17. Ludwig K (2008) Users Manual for Isoplot 3.7 (Berkeley Geochronol Cent, Berkeley, CA).18. Kirschvink JL (1980) The least-squares line and plane and the analysis of palaeomagnetic

data. Geophys J R Astron Soc 62(3):699–718.19. Gradstein FM, Ogg G, Schmitz M (2012) The Geologic Time Scale 2012 (Elsevier, New

York).20. Guo ZT, et al. (2002) Onset of Asian desertification by 22 Myr ago inferred from loess

deposits in China. Nature 416(6877):159–163.21. Qiang X, et al. (2011) New eolian red clay sequence on the western Chinese Loess

Plateau linked to onset of Asian desertification about 25 Ma ago. Sci China Earth Sci54(1):136–144.

22. Pettke T, Halliday AN, Rea DK (2002) Cenozoic evolution of Asian climate and sourcesof Pacific seawater Pb and Nd derived from eolian dust of sediment core LL44-GPC3.Paleoceanography 17(3):3-1–3-13.

23. Rea DK, Leinen M, Janecek TR (1985) Geologic approach to the long-term history ofatmospheric circulation. Science 227(4688):721–725.

24. Ziegler CL, Murray RW, Hovan SA, Rea DK (2007) Resolving eolian, volcanogenic, andauthigenic components in pelagic sediment from the Pacific Ocean. Earth Planet SciLett 254(3-4):416–432.

25. Sun XJ, Wang PX (2005) How old is the Asian monsoon system? Palaeobotanical re-cords from China. Palaeogeogr Palaeoclimatol Palaeoecol 222(3-4):181–222.

26. Harrison TM, Copeland P, Kidd WSF, Yin A (1992) Raising Tibet. Science 255(5052):1663–1670.

27. Tapponnier P, et al. (2001) Oblique stepwise rise and growth of the Tibet plateau.Science 294(5547):1671–1677.

28. Royden LH, Burchfiel BC, van der Hilst RD (2008) The geological evolution of the Ti-betan Plateau. Science 321(5892):1054–1058.

29. Molnar P, England P (1990) Late Cenozoic uplift of mountain ranges and global cli-mate change: Chicken or egg? Nature 346(6279):29–34.

30. Sun JM, et al. (2014) Palynological evidence for the latest Oligocene–early Miocenepaleoelevation estimate in the Lunpola Basin, central Tibet. Palaeogeogr PalaeoclimatolPalaeoecol 399:21–30.

31. DeCelles PG, et al. (2007) High and dry in central Tibet during the Late Oligocene.Earth Planet Sci Lett 253(3-4):389–401.

32. Raymo ME, Ruddiman WF (1992) Tectonic forcing of late Cenozoic climate. Nature359(6391):117–122.

33. Molnar P, Boos WR, Battisti DS (2010) Orographic controls on climate and caleocli-mate of Asia: Thermal and mechanical roles for the Tibetan Plateau. Annu Rev EarthPlanet Sci 38:77–102.

34. Zhang ZS, Wang HJ, Guo ZT, Jiang DB (2007) What triggers the transition of palae-oenvironmental patterns in China, the Tibetan Plateau uplift or the Paratethys Searetreat? Palaeogeogr Palaeoclimatol Palaeoecol 245(3-4):317–331.

35. Kutzbach JE, Prell WL, Ruddiman WF (1993) Sensitivity of Eurasian climate to surface

uplift of the Tibetan Plateau. J Geol 101(2):177–190.36. Sobel ER, et al. (2013) Oceanic-style subduction controls late Cenozoic deformation of

the northern Pamir orogen. Earth Planet Sci Lett 363:204–218.37. Treloar PRP, Williams M (1991) The role of erosion and extension in unroofing the

Indian Plate thrust stack, Pakistan Himalaya. Geol Mag 128(5):465–478.38. Hendrix MS, Dumitru TA, Graham SA (1994) Late Oligocene-early Miocene unroofing

in the Chinese Tian Shan: An early effect of the India-Asia collision. Geology 22(6):

487–490.39. Liu H, et al. (2010) The detrital zircon fission-track ages constraint to tectonic process

in west Kunlun and adjacent regions. Earth Sci Front 17(3):64–78.40. Prell WL, Kutzbach JE (1992) Sensitivity of the Indian monsoon to forcing parameters

and implications for its evolution. Nature 360(6405):647–652.41. Ruddiman WF, Kutzbach JE (1990) Late Cenozoic plateau uplift and climate change.

Trans R Soc Edinburgh Earth Sci 81(04):301–314.42. Schwab M, et al. (2004) Assembly of the Pamirs: Age and origin of magmatic belts

from the southern Tien Shan to the southern Pamirs and their relation to Tibet.

Tectonics 23(4):TC4002.43. Robinson AC (2009) Geologic offsets across the northern Karakorum fault: Implica-

tions for its role and terrane correlations in the western Himalayan-Tibetan orogen.

Earth Planet Sci Lett 279(1-2):123–130.44. Coutand I, et al. (2002) Late Cenozoic tectonic development of the intramontane Alai

Valley, (Pamir-Tien Shan region, central Asia): An example of intracontinental de-

formation due to the Indo-Eurasia collision. Tectonics 21(6):1053.45. Cowgill E (2010) Cenozoic right-slip faulting along the eastern margin of the Pamir

salient, northwestern China. Geol Soc Am Bull 122(1-2):145–161.46. Schmidt J, et al. (2011) Cenozoic deep crust in the Pamir. Earth Planet Sci Lett 312(3-4):

411–421.47. Arrowsmith JR, Strecker MR (1999) Seismotectonic range-front segmentation and

mountain-belt growth in the Pamir-Alai region, Kyrgyzstan (India-Eurasia collision

zone). Geol Soc Am Bull 111(11):1665–1683.48. Zhou ZY, ed (2001) Stratigraphy of the Tarim Basin (Science Press, Beijing).49. Zheng HB, Tada R, Jia J, Lawrence C, Wang K (2010) Cenozoic sediments in the

southern Tarim Basin: Implications for the uplift of northern Tibet and evolution of

the Taklimakan Desert. Spec Publ Geol Soc London 342:67–78.50. Tada R, et al. (2010) Desertification and dust emission history of the Tarim Basin and

its relation to the uplift of northern Tibet. Spec Publ Geol Soc London 342:45–65.51. Renne PR, Balco G, Ludwig KR, Mundil R, Min K (2011) Response to the comment by

W. H. Schwarz et al. on “Joint determination of 40K decay constants and 40Ar*/40K

for the Fish Canyon sanidine standard, and improved accuracy for 40Ar/39Ar geo-

chronology” by P. R. Renne et al. (2010). Geochim Cosmochim Acta 75(17):5097–5100.52. Lee JY, et al. (2006) A redetermination of the isotopic abundances of atmospheric Ar.

Geochim Cosmochim Acta 70(17):4507–4512.53. Koppers AAP (2002) ArArCALC - Software for Ar-40/Ar-39 age calculations. Comput

Geosci 28(5):605–619.

Zheng et al. PNAS | June 23, 2015 | vol. 112 | no. 25 | 7667

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

Dow

nloa

ded

by g

uest

on

June

25,

202

0