The initial occurrence of desert landscape or eolian sand dunes is thought to have occurred long before the Pleistocene, and desertification was subsequently enhanced under cold, dusty glacial conditions. However, when and how the desert landscape persisted during both glacial and interglacial periods, defined as “permanent” desert here, remain elusive. Here, we present carbonate carbon isotope and grain-size records from the Tarim Basin, western China, revealing a detailed desertification history for the Taklimakan Desert. Our records demonstrate that after desiccation of episodic lakes at ca. 4.9 Ma, alternations of eolian sand dunes and fluvial and playa-like conditions persisted for a long period until 0.7 Ma in the Tarim Basin. The onset of permanent desert landscape around 0.7–0.5 Ma occurred concurrently with the climatic reorganization across the mid-Pleistocene transition. The occurrence of mountain glaciers on the Tibetan Plateau and atmospheric circulation changes may have controlled the formation and extreme aridification of the permanent desert in inland Asia since the mid-Pleistocene transition.

Long, continuous terrestrial records documenting inland aridity changes are extremely limited, preventing our understanding of the development of aridification and desertification in continent inland areas and their association with Pliocene–Pleistocene global climatic changes. The Taklimakan Desert in the Tarim Basin, western China (Fig. 1), the second largest shifting-sand desert in the world, is part of the prominent arid zone in Central Asia. Today, extremely dry climate prevails in the basin (annual precipitation < 50 mm at its center), and shifting sand dunes occupy most of the basin, with little vegetation (Zhu et al., 1980). Summer subtropical anticyclone and air subsidence there are thought to be remotely forced by the rising activity in monsoon regions (Rodwell and Hoskins, 1996; Sato and Kimura, 2005). A detailed aridification history over the Pliocene–Pleistocene period from this region, which is perhaps most representative of inland aridity changes around the globe, would help us to understand monsoon-desert dynamics in response to global climatic changes.

Although it has been extensively studied over the past decades, the timing and mechanisms of Taklimakan Desert formation still remain elusive. The proposed timing, mostly based on the initial occurrence of eolian deposits, ranges from the Pleistocene to Miocene and late Oligocene (Zhu et al., 1980; Fang et al., 2002a; Sun et al., 2009; Zheng et al., 2015; Heermance et al., 2018). Proposed possible mechanisms for Asian inland aridification, and by inference, desertification, include tectonic uplift (Fang et al., 2002a), land-sea distributions (Bosboom et al., 2014), and Cenozoic global cooling (Miao et al., 2012). It is generally believed that the initial desert landscape occurred well before the Pleistocene, and desertification was enhanced under cold, dusty Pleistocene glacial conditions. However, little is known about when and how the desert ceased being replaced by other landscapes during relatively warm, humid interglacial periods, such as the present Holocene period, defined as “permanent” desert, which is addressed here.

We used a 1050-m-long, continuous borehole (Ls2) that we drilled at Lop Nur (39°46′0″N, 88°23′19″E) in the eastern Tarim Basin (Fig. 1). Previous studies have established a robust chronology (Fig. S1A in the Supplemental Material1; Chang et al., 2012), and, through study of the lower section of the core (651–1050 m), inferred the occurrence of late Miocene episodic lakes, based on carbonate oxygen and boron isotopes (Liu et al., 2014). We here addressed how the Taklimakan Desert evolved into its present form after the episodic lakes dried up at ca. 4.9 Ma (Liu et al., 2014; Sun et al., 2017). Existing records from exposed Mazartag sections in the central basin (Fig. 1; Pei et al., 2011; Sun et al., 2017), where previously controversial age assignments have been reconciled with a newly reported mammal fossil (Fig. S1C; Sun et al., 2017), were also integrated to infer basinwide paleoenvironmental changes. Lop Nur currently lies at the lowest elevation in the eastern basin, ∼800 m above sea level, and is terminal for receiving river water in the basin, and the Mazartag sections are at ∼1150 m in the central basin (Fig. 1), so the borehole at Lop Nur would likely record the last stage of aridity changes in the basin, while approximately synchronous signals at both locations could indicate major, basinwide paleoenvironmental changes.

We generated carbonate carbon isotope (δ13Ccarb) and grain-size records at very high resolution (<2 k.y. on average; Table S1). The δ13Ccarb values from eolian and fluvial deposits can become progressively negative with the input of soil organic carbon during soil pedogenesis (see the Supplemental Material, and Fig. S2; Liu et al., 2011; Sun et al., 2015; Caves et al., 2016). Hence, δ13Ccarb can effectively reflect regional vegetation cover and biological productivity in arid and semiarid regions. Grain size, due to its complicated nature, is used here as a supporting indicator (see the Supplemental Material). However, when desert landscape occurs, the sand fraction should dominate the size distribution. Using the two indicators, we defined the onset of “permanent” desert formation as the point at which δ13Ccarb consistently fell within 0‰ ± 1‰ and modal size, the mostly frequent occurring particle size, was >∼100 μm, both similar to those in modern Taklimakan sand. Analytical procedures for δ18Ocarb, δ13Ccarb, and grain-size analyses can be found in the Supplemental Material.

Our δ13Ccarb record over the past 7.1 m.y. shows a long-term trend toward more positive values, with large variability between ∼1‰ and −5‰ (Fig. 2B). After the episodic lacustrine phase, as evidenced by the high variability in δ18Ocarb between 7.1 and 4.9 Ma (Fig. 2D; Liu et al., 2014), δ13Ccarb values ranged from ∼0‰ to −5‰ between 4.9 and 3 Ma, and from ∼1‰ to −3‰ with a trend becoming progressively positive between 3 and 0.7 Ma. Since ca. 0.7 Ma, δ13Ccarb values stayed around 0‰ ±1‰, close to the background signal in detrital carbonates (Liu et al., 2014). The grain-size record (Fig. 2C) also shows large variability in modal size, from 5–10 μm to 100–150 μm before ca. 0.5 Ma. An abrupt increase occurred at ca. 0.5 Ma, and since then, the modal size has varied between ∼100 and 200 μm (Fig. 2C; Fig. S3). Notably, after the mid-Pleistocene transition (MPT) between ca. 1.25 and 0.7 Ma (Clark et al., 2006), both δ13Ccarb and modal size show little glacial-interglacial variability, which indicates persistent desert landscape during both glacial and interglacial periods.

We also closely examined lithological changes (Figs. S3 and S4). Prior to ca. 4.9 Ma (651 m), the occurrence of gray and bluish gray argillaceous limestone suggests lacustrine environments. After that, the sediment gradually changed from gray and bluish-gray to yellow/brown color, with an increasing silt component. Since ca. 0.7 Ma (34 m), sediments mainly consist of coarse sand, with little soil pedogenesis developed. Thus, the δ13Ccarb and grain-size records (Figs. 2B and 2C), together with lithological changes (Figs. S3 and S4), indicate three desertification stages in the Tarim Basin: episodic lakes before ca. 4.9 Ma (Liu et al., 2014), alternations of eolian sand dunes, fluvial deposits, and playas (Fig. 2D) between ca. 4.9 Ma and 0.7 Ma, and “permanent” desert landscape after 0.7 Ma.

The δ13Ccarb records from exposed Mazartag sections in the central basin (Pei et al., 2011; Sun et al., 2017), 700 km away from our Lop Nur borehole, confirm that the Lop Nur profiles represent basinwide paleoenvironmental changes. In the lower Mazartag section, substantially increased δ13Ccarb and decreased δ18Ocarb values (Figs. 2B and 2D) indicate lake desiccation at ca. 5.3 Ma at this locality (Sun et al., 2017). Between 4.9 and 2.6 Ma, the two profiles display similar δ13Ccarb variability. Although δ13Ccarb values are more negative in the upper Mazartag section (1.6–0.1 Ma; Fig. 2B; Fig. S5), the pattern of δ13Ccarb variations in the two profiles is quite similar, with both gradually becoming more positive and an abrupt increase in δ13Ccarb values at ca. 0.7 or 0.4 Ma (Fig. S5). Further, both lithological columns show coarsening-upward trends around the MPT (Figs. S1A and S1B). The similar δ13Ccarb values and lithological changes in the two profiles allow us to place the onset of modern-form permanent Taklimakan Desert at 0.7–0.5 Ma.

The high variability before ca. 0.7 Ma in both δ13Ccarb and grain-size records appears to have been associated with glacial-interglacial changes. Spectral analysis confirms the presence of variability at orbital bands, particularly at ∼400 k.y. (Fig. S6). More negative δ13Ccarb values and smaller grain size occurred during interglacial periods, while the opposite took place during glacial periods (Figs. S7A and S7B). The extremely negative δ13Ccarb values (−3‰ to −5‰) seem to have occurred, within chronological uncertainty, ∼400 k.y. apart. Hence, as indicated by the glacial δ13Ccarb value (0‰ ± 1‰) and large modal size (∼100–200 μm; Figs. 2B and 2C), brief desert landscapes could have occurred much earlier, not inconsistent with the various timing inferred previously (Zhu et al., 1980; Fang et al., 2002a; Sun et al., 2009, 2017; Liu et al., 2014; Zheng et al., 2015; Heermance et al., 2018). However, all these brief desert occurrences were replaced by substantial interglacial vegetation cover before 0.7 Ma, as shown by interglacial δ13Ccarb values (−2‰ to −5‰) and much smaller modal sizes (5–10 μm). The gradual deterioration of vegetation during the MPT (1.25–0.7 Ma) but subsequent abrupt increase in modal size at ca. 0.5 Ma (Figs. 3F and 3G; Fig. S7C) may suggest that the permanent desert landscape could have fully formed only when a certain vegetation cover threshold was crossed through a positive vegetation (albedo) feedback in desert dynamics (Abell et al., 2020). It remains to be verified whether “permanent” desert landscapes in other Asian inland deserts developed around the same time; however, substantially enhanced aridification around the MPT time has also been reported in the Qaidam Basin (Cai et al., 2012), Gurbantunggut Desert (Fang et al., 2002b), and the west margin of the Tarim Basin (Zhang et al., 2019), largely consistent with the permanent Taklimakan Desert formation at 0.7–0.5 Ma inferred here.

Our high-resolution Lop Nur profiles show close correspondence to Pliocene–Pleistocene global climatic changes before 0.7 Ma (Fig. 2; Fig. S7). During glacial/cool periods, minimal vegetation cover in the basin occurred while Asian summer monsoons also weakened (Sun et al., 2010). During interglacial periods at ca. 1.2, 2.0, 2.4, and 2.8 Ma, marked increase in vegetation cover in the basin coincided with enhanced East Asian summer monsoons (Fig. S7A). Supported by model simulations (An et al., 2001), glacial conditions or cool global temperatures tend to enhance aridity in both Asian inland and monsoonal regions. The close correspondence between synchronous aridity changes in both monsoon and desert regions and large global climatic changes after ca. 3 Ma (Fig. 2; Fig. S7) leaves the role of tectonic uplift dispensable. In addition, the high elevations (>3000 m) of the Tibetan Plateau were attained at least by the latest Miocene (Molnar, 2005), and no firm evidence exists for marked uplift after ca. 3 Ma (Clift et al., 2017). We thus suggest that global climatic changes, the long-term cooling trend in particular, were largely responsible for the deteriorating vegetation cover in the Tarim Basin between ca. 3 and 0.7 Ma.

However, the onset of the permanent desert landscape, around 0.7–0.5 Ma (Figs. 2B and 2C), is difficult to explain by the global cooling effect. The MPT was characterized by further cooling and ice-sheet development during glacial periods (Fig. 3A; Clark et al., 2006), yet interglacial temperatures around the globe (Snyder, 2016) and in western tropical Pacific Ocean (Medina-Elizalde and Lea, 2005) remained about the same (Figs. 3B and 3C). One of the prominent regional features during the MPT was that mountain glaciers started to occur on the Tibetan Plateau and surrounding mountains (Heyman, 2014; Owen and Dortch, 2014), approximately coincident with the permanent desert formation (Fig. 3E). Mountain glaciation, through enhanced frost weathering and/or glacial grinding, would intensify physical weathering and rock breaks, thus generating more sand-sized materials (Smalley, 1995) for sustaining a permanent large desert. We suggest that the recurrent advance and retreat of mountain glaciers, corresponding to glacial-interglacial cycles, provided tremendous volumes of sand for the formation of the permanent desert in the Tarim Basin since ca. 0.5 Ma.

Intriguingly, there exists some indication that the East Asian (Sun et al., 2010; Meng et al., 2018) and Indian summer monsoons (An et al., 2001; Huang et al., 2007) were more enhanced during interglacial periods after the MPT than before (Fig. 3H), while in the Tarim Basin, interglacial aridity increased after the MPT as compared to before the MPT (Figs. 3F and 3G). Mountain glaciation could have increased aridity in both desert and monsoon regions (An et al., 2001), but this could not explain the opposite aridity behavior for interglacial periods after ca. 0.7 Ma. Instead, because air subsidence and thus suppressed rainfall in desert regions are remotely enhanced by diabatic heating over Asian monsoon regions (Rodwell and Hoskins, 1996; Sato and Kimura, 2005), this monsoon-desert mechanism could well explain the opposite aridity behavior for interglacial periods after the MPT in the two regions (Figs. 3F–3H). Previously, the enhanced interglacial summer monsoon across the MPT (Meng et al., 2018) has been linked to a greater tropical zonal temperature gradient and strengthened Walker Circulation (Wara et al., 2005), as interglacial temperatures continued to cool in the eastern equatorial Pacific (Herbert et al., 2010) but remained stable in the western Pacific (Figs. 3C and 3D; Medina-Elizalde and Lea, 2005). Accordingly, we suggest that in response to strengthened monsoon circulation across the MPT, air subsidence was enhanced in desert regions through the monsoon-desert dynamics (Rodwell and Hoskins, 1996). Hence, further reorganization of atmospheric circulation around the MPT may have also contributed to the onset of the permanent Taklimakan Desert, which requires further investigation with climate models.

In summary, our high-resolution δ13Ccarb and grain-size records over the past 7.1 m.y. from the borehole at Lop Nur provide a detailed desertification history in the Tarim Basin, western China. Our results show the onset of the “permanent” Taklimakan Desert at 0.7–0.5 Ma, when desert landscape persisted during both glacial and interglacial periods. The overall deteriorating vegetation cover between 3 Ma and 0.7 Ma in the Tarim Basin corresponded to global cooling, but the onset of permanent desert at 0.7–0.5 Ma appears to have been linked to the mountain glaciation in nearby regions and atmospheric circulation changes across the MPT.

We thank the editor and four anonymous reviewers for their constructive reviews. This research was supported by the Chinese Academy of Sciences (grants XDB40010100, XDA20070202, and QYZDY-SSW-DQC001), the National Natural Science Foundation of China (grant 41420104008), and the National Key Research and Development Program of China (grant 2016YFE0109500). The Lop Nur Drilling Project was supported by the Chinese Academy of Sciences.

1Supplemental Material. Materials, methods, and interpretations, Figure S1 (stratigraphic columns and age assignments), Figure S2 (δ13C correlated with rainfall and temperature), Figure S3 (detailed view of the onset of desertification), Figure S4 (representative core photos), Figure S5 (Lop Nur and Mazartag δ13C), Figure S6 (spectral results), Figure S7 (close-up views of records), and Table S1 (δ18O, δ13C, and grain-size data). Please visit to access the supplemental material, and contact with any questions.
Gold Open Access: This paper is published under the terms of the CC-BY license.