Mid-Miocene sea level altitude of the Qaidam Basin, northern Tibetan Plateau

A detailed uplift history of the Tibetan Plateau is essential for disentangling the proposed geodynamical models and assessing its impacts on climate and biodiversity. However, when and how the plateau formed remains highly controversial. Here, we present unusual geochemical indicators of marine signatures in the Cenozoic terrestrial strata of the Qaidam Basin, northern Tibetan Plateau, with strong implications for the basin altitude. Our investigations across the basin reveal typical marine alkenones and anomalously high carbonate carbon isotopic values during the mid-Miocene, but not at earlier stages, which are accompanied by a divergent trend in the paired carbonate oxygen and leaf wax hydrogen isotopic records. We infer an incursion of seawater into the Qaidam Basin, thus constraining the mid-Miocene basin altitude close to sea level. Hence, much of the substantial northern plateau uplift afterwards appears to be associated with the outward growth of the Tibetan Plateau. The identification of marine alkenones and anomalously high carbonate carbon isotopic values in sediments suggest the Qaidam Basin in the northern Tibetan Plateau was uplifted from sea level to its present altitude since the Mid-Miocene.

T he collision of Indian and Eurasian plates, since~55 ± 10 million years ago (Ma) 1 , generated the most prominent elevated landmass on Earth, the Himalaya-Tibetan Plateau, thought to have reshaped Asian and even global atmospheric circulation and ecosystems [2][3][4] . Hypotheses evolved from earlier proposal of plateau uplift as a whole block 5 to diachronous models, the latter type including northward propagation of stepwise growth 6 and outward growth of a "proto-Tibetan Plateau" at its present center 1 . The recent discovery of a central deep valley in the Lunpola Basin by the early Miocene 7 , possibly bounded by pre-existing mountain ranges before the collision 4 , implies that the relatively flat modern plateau topography is only a recent geological phenomenon. During the mid-Miocene, approximately around 15 Ma, a profound geodynamic event seems to have occurred, as evidenced by the outward growth of the plateau and widespread normal faulting in the plateau interior (summarized in ref. [8][9][10], indicating a change in style of plateau deformation, perhaps from previous crustal shortening and thickening to extension and thinning [8][9][10][11] . Tectonic end-member models predict competing temporalspatial evolution patterns for the northern Tibetan Plateau (NTP, north of the Kunlun Mountains) across a distance of at least 1500-2000 km from the collision zone in the south (Fig. 1a): the early surface rise since the terminal collision 5 , versus delayed orographic development in the context of stepwise growth 6 . Determining the detailed topographic history of the NTP would provide vital insights into underlying mechanisms of plateau uplift. Despite active research, when and how the NTP was uplifted remains contentious 1,[4][5][6][7][8][9][10][11][12][13][14][15][16][17] . Particularly, the mid-Miocene southern Tarim Basin was suggested to be at sea level and connected to open marine waters, inferred from planktonic foraminifers discovered at the Miran River section 17 (Fig. 1b). The topographic evolution of the Qaidam Basin (QB) on the NTP, with present altitudes of~3,000 m above sea level (Fig. 1b), also remains controversial, ranging from substantial uplift since the mid-Miocene [18][19][20] to near-present elevation by the early Oligocene 21 . These discrepancies are rooted in uncertainties in quantitative constraints on surface uplift from the isotope paleoaltimetry and plant fossils used 4 , representative with assumptions regarding climatic tolerance of fossil species 21 , preservation conditions of fossil leaves 21 , temporally and spatially varying isotopic values 18,19 , and complex moisture transport pathways 18,19 . We here report unusual geochemical indicators, bearing marine signatures, discovered in the thick Cenozoic QB strata (Methods). Our new results, corroborated with the independent constraint for the southern Tarim Basin 17 (Fig. 1b), offer a new perspective on the QB altitude during the mid-Miocene [18][19][20][21] .

Results and discussion
Unusual geochemical signatures. Marine biomarker Group III alkenones, anomalously high carbonate carbon isotopic (δ 13 C carb ) values, and paired carbonate oxygen isotopic (δ 18 O carb ) and leaf wax hydrogen isotopic (δD alk ) values, were identified from the Cenozoic Dahonggou (DHG) and Honggou (HG) sections in the northern QB (Methods, Fig. 1, Supplementary Fig. 1, Supplementary Data). Alkenones were detected exclusively in the interval from the uppermost Xia Youshashan and lowermost Shang Youshashan formations (Fig. 1c), the ages of which are constrained to the mid-Miocene (~15-11 Ma) 16,22 or the late Miocene (~9-7.5 Ma) 23 (Supplementary Figs. 2,3), with the former being our preferred age model (Methods, Fig. 2). They were captured by 45 of 450 samples in the DHG and 13 of 177 in the HG section. Within the mid-Miocene interval, alkenones were detected in relatively deep lake facies 24 (Supplementary Figs. 4, 5). Meanwhile, in the older Ganchaigou Formation of the Hongliugou Section ( Supplementary Fig. 1), with relatively deep lacustrine facies developed 25 , alkenones were not detected in 216 samples analyzed. Identified alkenones in most DHG and HG samples belong to Group III Isochrysidales 26 , typically living in marine settings, and Group II alkenones, found in lacustrine and coastal settings, appeared briefly at~11 Ma (with one exception at 15 Ma at HG) ( Fig. 2a, b, Supplementary Fig. 6, Supplementary Note 1). Similarly, at the depo-center KC-1 well in the western QB (Fig. 1b), saturated ketones, diagenetic products of alkenones, also occurred during~15-6 Ma, but not before (~18-15 Ma) 27 (Fig. 2c). Within the interval of~15-6 Ma, alkenones occurred in relatively shallow lake facies 27 , in contrast with the DHG and HG sections ( Supplementary Fig. 5). Most of them were also Alkenone contents from the DHG and HG sections, respectively. c Saturated ketone contents from the KC-1 core 27 . d Surface water temperatures estimated from marine alkenones in the DHG and HG sections. e Carbonate δ 13 C records from the DHG and HG sections. The insets show representative gas chromatographs of unsaturated/saturated alkenones produced by Group III and II Isochrysidales. In (a-c), green/light green/purple and orange filled circles indicate Group III and II Isochrysidales, respectively, while gray empty circles indicate analyzed samples with alkenones not detected. The yellow bar marks the mid-Miocene interval with unusual geochemical signatures identified across the basin, with the dashed line in (e) showing elevated δ 13 C values during this interval, based on the conventional chronology 22 (Fig. 2c). Group III alkenones hence appear to have occurred basinwide around 15 Ma. C 37 alkenone concentrations range from trace to~450 and~2900 ng g −1 in the DHG and HG sections respectively (Fig. 2a, b) and C 37 saturated ketones in KC-1 could be up to 120 ng-g −1 (Fig. 2c), comparable to or slightly higher than those reported from mid-Miocene marine sediments 28 (mostly within 100 ng g −1 and up to 400 ng g −1 ). Surface water temperatures estimated from Group III alkenones range from~24 to 29°C in the DHG and HG sections (Methods) and display roughly a cooling trend (Fig. 2d), perhaps corresponding to global cooling during the mid-Miocene Climate Transition 28,29 .
Cenozoic δ 13 C carb records in both sections display substantial fluctuations, varying mostly between −7‰ and −2‰. Except for the mid-Miocene interval (~15-11 Ma), all δ 13 C carb values are < −1.3‰. During the mid-Miocene interval when the basinwide occurrence of Group III alkenones was observed, δ 13 C carb values reach 0.3-0.8‰ for the DHG and −0.5‰ for the lesssampled HG section (Fig. 2e). The δD alk values (Methods) vary from −224‰ to −132‰ and −210‰ to −125‰ for the HG and DHG sections respectively, showing considerable similarity in the temporal trends throughout the Cenozoic. Before~15 Ma, δD alk records resemble δ 18 O carb from the same DHG section 24 , including both the long-term trend and short-term fluctuations (Fig. 3b). However, after~15 Ma, δD alk values became substantially more negative,~45‰ less than the mean value beforẽ 15 Ma and even lower than late Eocene values, while δ 18 O carb became progressively more positive as in many lacustrine and pedogenic carbonate records 30 . The paired isotopic records thus display a strong divergent trend after~15 Ma (Fig. 3b, c).
Interpretations of geochemical indicators. Alkenones produced by Group III Isochrysidales are ubiquitous in marine settings, but not in modern inland lakes 31 . Hence, as the QB became a terrestrial sedimentary basin after the early Cenozoic 32 , the basinwide occurrence of Group III alkenones during the mid-Miocene needs to be critically assessed. Sediment reworking of nearby pre-Cenozoic marine strata can confidently be excluded based on alkenone concentrations, the shift from Group III to Group II alkenones, variable temperatures estimated from marine alkenones (Fig. 2), and alkenone occurrence in contrasting lake facies at the three sections (Supplementary Figs. 4,5). These observations thus strongly indicate in situ production of alkenones (Supplementary Note 2). Regarding how marine Isochrysidales might have arrived in the mid-Miocene QB initially that served as the "seed" for alkenone production, a proposal of the remnant species associated with the last major marine transgression during the late Eocene 33 could be refuted due to no alkenones detected in a wider range of lake facies before~15 Ma (Fig. 2, Supplementary Fig. 1). Possibly remnant Isochrysidales could not survive in a highly unstable lacustrine system for an extended period 34,35 . Less plausibly or even implausibly, marine Isochrysidales were carried by birds or winds and then survived in the mid-Miocene QB by chance. Alternatively, marine Isochrysidales could evolve themselves in a lacustrine system with water chemistry similar to oceans. For instance, thick-boned fish fossil reported from the arid Pliocene Qaidam Basin is similar to the one found in the Mediterranean Sea during the Messinian desiccation period 36 . However, as marine Isochrysidales are not observed in any modern inland lakes with very diverse lake water chemistry including those similar to oceans 31 , we consider either scenario unlikely. Instead, Group III alkenones are common in coastal oceans 31,37 (i.e., Black Sea) (Supplementary Note 2). The unusual marine signature found in the QB strata thus suggests possible seawater incursion during the mid-Miocene.
The abnormally high δ 13 C carb values within the mid-Miocene differ substantially from those (<−2‰) in Cenozoic pedogenic 30 and lacustrine carbonates on the Tibetan Plateau 38,39 . Cenozoic lacustrine δ 13 C carb values of > 0‰ have been reported from the plateau, but are usually accompanied by elevated δ 18 O carb values, indicating a strong lake evaporative effect 38,39 , although paired δ 13 C carb /δ 18 O carb values in Oligocene-Miocene lacustrine limestones in southern Nevada 40 , also interpreted to be deposited in restricted, high-salinity lake basins, are similar to those in the mid-Miocene DHG section. We consider that the evaporative effect could not account for the abnormally high δ 13 C carb values for the mid-Miocene DHG section as multiple proxies suggest not particularly arid environment at the time 24 (Fig. 3, Supplementary Fig. 5), which can also be refuted by the lower δ 13 C carb values with enhanced regional aridification after~11 Ma 24 (Fig. 2e). Elevated δ 13 C carb values could also be found in special settings, such as geothermal springs, volcanic and anoxic environment, however, the mid-Miocene DHG strata appear to be normal lacustrine/deltaic settings ( Supplementary Fig. 1). Lastly, the paired δ 13 C carb /δ 18 O carb values resemble those of marine carbonates 39,41 . Meanwhile, corroborated by the concurrence of marine alkenones, the marine signal originating from ancient marine carbonates can be excluded, as those values can only be preserved without being altered/diluted by soil and lake processes, like the desert environment 39 . This leaves mid-Miocene seawater incursion as the most viable explanation. The Z value 42 (Methods) was proposed to differentiate marine vs. terrestrial environment earlier. With the confounding lake salinization effect excluded ( Supplementary Fig. 5), Z values >120 (in the range of 120-125) within the mid-Miocene QB strata (Supplementary Data) may also indicate a marine signal 42 . Hence, the identified Group III alkenones and abnormal δ 13 C carb values in QB strata, together with the reported planktonic foraminifers from the southern Tarim Basin 17 (Figs. 2, 3), all suggest a possible mid-Miocene seawater incursion.
Mid-Miocene QB at sea level altitude. Following the previous proposal that the southern Tarim Basin was connected to open marine waters 17 , we further suggest that mid-Miocene Paratethyan seawater entered the QB through the Pamir salient, southern Tarim Basin, and along the Altyn Tagn Fault (ATF) (Fig. 4a). The mid-Miocene QB may have resembled the recent Black Sea, and been impacted by seawater incursion 37 . The northward propagation of Pamir salient by the mid-Miocene 43,44 might still allow the Tarim Basin to be connected to the Paratethyan Sea 17 . A pre-Quaternary east-west paleoriver along the southern Tarim Basin existed, carrying detritus from the arcuate Pamir belt to the QB through the ATF 34 . At~24-15 Ma, the rapid left-lateral offset of the reactivated ATF [12][13][14] (Fig. 4b) might have further deepened channels to allow a connection. In the QB, the sedimentary basin experienced earlier flexural subsidence and perhaps remained at or near sea level, with isolation and partition after the mid-Miocene 16 .
This interpretation is supported by our paired δ 18 O carb and δD alk records from the DHG section (Fig. 3). The divergent trend since the mid-Miocene may have resulted from the altitudinal effect of the uplifting basin and surrounding mountains, reflected in the δD alk trend, while δ 18 O carb , similar to other lacustrine and pedogenic δ 18 O carb records 24,30 , is more affected by regional aridification, overprinting the altitudinal effect (Supplementary Note 3). Alkenone temperature (Fig. 2) is not used as independent evidence to infer QB altitude here due to large uncertainty in estimates of mid-Miocene terrestrial temperature and altitudinal temperature gradient. Nonetheless, those warm temperatures, similar to mid-latitude ocean temperatures 28,29 , do not contradict a sea level altitude. Moreover, higher global sea level, rising by~40-70 m during the mid-Miocene 45 , might have made connection of the QB to Paratethys possible or enhanced the connection (further discussion in Supplementary Note 4). We note that the interval of alkenone occurrence at all three localities, 15-11 Ma, does not exactly match the period of higher global sea level 45 (Fig. 3). The exact reason remains unclear. Apart from chronological uncertainty, we speculate that marine Isochrysidales might need time to adapt new environment in a lacustrine system or the seawater pathway was disconnected during the early part of the mid-Miocene. It also remains unclear whether multiple or persistent seawater incursions occurred during the whole interval of~15-11 Ma, as marine Isochrysidales, once entering the QB, might have existed for many generations if lake conditions did not change dramatically. Hence, with details to be explored (Supplementary Note 4), we consider seawater incursion as an accidental event, coincident with mid-Miocene sea level rise and reactivated offset along the ATF serving to deepen connecting channels.
Implications for Tibetan Plateau uplift. The unusual marine signatures identified in QB strata effectively constrain the basin altitude to sea level by the mid-Miocene. Thus, uplift to the present basin altitude,~3000 m, mostly occurred subsequently. Although our findings do not provide a direct constraint on the altitude of surrounding mountain ranges, numerous lines of evidence accumulated over the past decades, based on analyses of low-temperature thermochronology, sedimentation, and stratigraphy, indicate accelerated uplift of various parts of the NTP since the mid-Miocene 8-16 (Fig. 4b, Supplementary Note 5), while only minor pre-Miocene relief existed surrounding the QB 16 . With isostatic balance~1100 m rise of the QB since the Miocene can be attributed to sediment infill 20 . With the new constraint at sea level, about 2/3 of the present QB altitude may result from crustal shortening and thickening 9-11 , implying a similar magnitude of uplift for surrounding mountain ranges. Hence, collective evidence suggests the NTP as a whole uplifted substantially since~15 Ma.
As the volcanic chain of an Andean-style convergent plate margin, the Gangdese Mountains already attained high altitude prior to India-Asia collision 46 , the Qiangtang (Tanggula) Mountains reached >5000 m altitude by late Eocene 47 , while the Himalayas reached their present altitude by mid-Miocene 48 (Fig. 5a, additional data summarized in ref. [49][50][51]. With the NTP uplifted substantially post mid-Miocene, the available evidence supports diachronous models of the plateau uplift 4 , predicting a late rise of the NTP. However, none of the current geodynamical models place the mid-Miocene QB at sea level, with most at~1000-2000 m altitude instead. Meanwhile, the recently discovered deep central valley in the Lunpola Basin 7 (<2300 m), lower altitudes of the Qiabulin Basin 52 (<2500 m) and the Hoh Xil Basin 53 (~3000 m), and sea level altitude for the QB, together with high relief on the central plateau constrained by isotope paleoaltimetry [46][47][48][49][50][51] , suggest the existence of a series of east-west orientated deep valleys by the early Miocene (Fig. 5b), providing an ecological niche for Asian biodiversity 4 . The relatively flat modern topography on the higher plateau appears to have been achieved around~15 Ma, suggesting prior continuous plateau growth before reaching its highest elevation 9,10 . The change in style of plateau deformation at~15 Ma indicates the higher plateau transitioned from previous crustal shortening and thickening to extension and thinning [8][9][10][11] . This geodynamic event across the plateau can be potentially explained by natural outward growth of the higher plateau, removal of some mantle lithosphere, and outward flow of weakened middle to lower crust 8 . Considering the magnitude of QB uplift (~3000 m) inferred here, removal of some mantle lithosphere beneath the plateau 9-11 provides a sensible, but perhaps not unique, solution for the mid-Miocene geodynamic event.
In summary, the unusual marine signatures identified in mid-Miocene QB strata provide new insights into the topographic evolution of the NTP. With careful considerations, the marine signal is best explained by an incursion of Paratethyan seawater into the QB during the mid-Miocene, via a paleochannel (turning into a paleo-seaway temporarily) along southern margin of the Tarim Basin 34 , analogous to the recent Black Sea. We consider the incursion as an accidental event coincident with global sea level highstand 45 and reactivation of tectonic activity along the ATF [12][13][14] . If correct, our results may constrain the mid-Miocene QB to sea level, postdating all but one 17 estimate of NTP paleoaltitude. Both isotopes and plant pollen/fossils, the two main paleoaltimeters used currently to constrain the altitude of bounding mountain ranges and any altitude between the basin and ranges respectively 4 , could potentially overestimate basin altitudes. The discrepancies of mid-Miocene QB altitude estimates between ours and those from plant fossils 21 and isotopes 18,19 , support this view, although the latter also suggests accelerated uplift post mid-Miocene (Fig. 5). The mid-Miocene QB sea level altitude, attesting the widely reported change in style of plateau deformation 8-10 , can be incorporated into geodynamical models in order to better understand plateau uplift history and mechanisms, with important implications for climatic and ecological impacts.

Methods
Materials and chronologies. The northern Tibetan Plateau is conventionally separated from adjacent regions by three major faults, the Altyn Tagh Fault to the northwest, the Haiyuan Fault to the northeast, and the Kunlun Fault to the south 5 . The Cenozoic Qaidam Basin (QB), currently at~3000 m altitude, is the largest basin on the northern Tibetan Plateau, surrounded by the Kunlun Mountains to the south and west, the Altyn Mountains to the north, and the Qilian/Nan Mountains to the north and east (Fig. 1b). It received sedimentation up to 14 km thick in the middle of the basin 54 . The Dahonggou (DHG, 37.49°N, 95.21°E),  Honggou (HG, 37.54°N, 95.17°E), and Hongliugou (HLG, 38.13°N, 94.67°E) sections are well exposed at the eastern side of the QB (Fig. 1b). Specifically, both DHG and HG sections are located in the southern limb of the same Dahonggou anticline, being~10 km apart from each other (Fig. 1c). The two sections contain >5000 m thick lacustrine, deltaic to braided river deposits and can be divided into six basin-scale lithostratigraphic units, in a geological sequence, including the Lulehe, Xia Ganchaigou, Shang Ganchaigou, Xia Youshashan, Shang Youshashan, and Shizigou Formations (Fm) (Supplementary Fig. 1). Detailed descriptions of lithologies and sediment facies have been provided previously [21][22][23]27,32,54 .
At the DHG and HG sections, we collected lacustrine to deltaic mudstones and silty mudstones from the Xia Ganchaigou to Shang Youshashan Fms. For the intervals of fluvial and alluvial deposits, i.e., the Lulehe and Shizigou Fm, mudstone intercalations were sampled in order to achieve sufficient organic materials. In total, 450 and 177 samples have been attained from the DHG and HG sections, respectively. We first took samples at every~15 m and~30 m for most of the two sections respectively, and then re-sampled the layers from the upper Xia Youshashan to lower Shang Youshashan Fms at the DHG section, where alkenones were detected. At the HLG section, we collected 216 mudstone samples from the upper Xia Ganchaigou to lower Shang Ganchaigou Fms (~395 m), where sediment facies of this interval is suggested to be deep lacustrine facies 25 .
A high-resolution magnetostratigraphic study suggests DHG and HG sedimentary successions from~52 Ma (the base of Lulehe Fm) to~7 Ma (the lower Shizigou Fm), based on constraints from fossil mammals, ostracods, leaves and pollen 22 (Supplementary Fig. 2). The HLG chronology follows a previous highresolution magnetostratigraphic study which constrains the entire sequence to bẽ 53.4 Ma (the base of Lulehe Fm) to 2.4 Ma (the upper Shizigou Fm) 25,55 . Accordingly, our sampled interval at the HLG section is assigned from~36 to 28 Ma. Recently, the conventional age (~52-7 Ma) of the DHG and HG sections is challenged by a magnetostratigraphic study, which, by applying the newly discovered Honggou fauna as age control, assigned a much younger age of   (Supplementary Fig. 2), or~30-5 Ma based on magnetostratigraphy combined with detrital apatite fission-track ages from another section in the region 56 . In these age models, linear sedimentation rates were assumed between datum ages. Consequently, the interval with alkenone occurrence at the DHG and HG sections would be shifted from~15-11 Ma with the conventional chronology 22 to~9-7.5 Ma with the revised chronology 23 .
A recent review 16 concludes that the conventional chronology 22 fits reasonably well with growth-strata observations and thermochronology data across the basin. The conventional chronology at the DHG and HG sections is also consistent with the one at the HLG section 55 , anchored by two mammal fossil beds in the Shang Ganchaigou Fm. Paleoclimate studies from the HLG 55 and DHG 24 sections show good correspondence to global climatic changes if the conventional chronology is used. Further, the chronology at the KC-1 well in the western basin 57 is not affected by the controversies in the northern basin 22,23 , indicating the alkenone occurrence starting from~15 Ma at this locality 27 . Hence, to some degree, our main conclusion here, mid-Miocene seawater incursion (Figs. 2, 3), is not significantly impacted by the chronological uncertainty in the northern basin, unless the KC-1 well chronology is also subject to revision (but the mid-Miocene warmer and wetter conditions, indicated by multiple proxies 27,57 , seem difficult to be shifted to an alternative time interval). With the revised chronology in the northern basin 23 , it might be inferred that marine Isochrysidales expanded from the western basin to DHG and HG localities at a later time,~9-7.5 Ma. However, this scenario seems to be less likely considering the regional aridification history 24,57 . We thus here present our results based on the conventional chronology 22 , while the revised chronology 23 is provided in Supplementary Fig. 2 and our DHG and HG results based on the revised chronology are provided in Supplementary Fig. 3.
Alkenone analysis. Weathered sediments of the studied outcrops were removed before sample collection in the field. In the lab, all samples (from DHG, HG and HLG sections,~30 g each) were freeze-dried, ground and extracted with CH 2 Cl 2 :MeOH (9:1, v/v) using Dionex Accelerated Solvent Extractor (ASE300). Then, we separated the total lipid extraction eluting on a silica gel column with three eluents of n-hexane, CH 2 Cl 2 and MeOH, respectively. Long-chain alkenones and n-alkanes fall in CH 2 Cl 2 and n-hexane fractions, respectively.
Subsequently, the alkenone-containing fraction was analyzed on an Agilent 7890B gas chromatography (GC) with an Agilent J&W column (DB-1, 60 m × 0.25 mm i.d. × 0.1 μm film thickness) and H 2 (1.5 ml min −1 ) as the carrier gas 58 . The oven temperature program was: 60°C (2 min) to 270°C at 15°C min −1 and then to 310°C (held 40 min) at 5°C min −1 . n-C 36 alkane of a known concentration was used as an internal standard for quantification. Group III and II alkenones from sediments of the South China Sea and Lake Gahai respectively are used for alkenone identification via retention time reference ( Supplementary  Fig. 6). Additional identification of alkenones was then undertaken by gas chromatography-mass spectrometer (GC-MS), Agilent 7890 A GC coupled to 5975 mass selective detector (MSD). Chromatographic separation was achieved with the aforementioned DB-1 column and its accompanying oven temperature program. Helium was used as carrier gas with a flow rate of 1.0 ml min −1 . The MS conditions were: electron impact (EI) mode, 70 eV and 280°C, scan range m/z 50-650, cycle time 0.65 ms. Detailed peak assignments were conducted by comparing them with previously reported diagnostic fragments in the mass spectra 59 ( Supplementary  Fig. 6).
Finally, for Group III alkenones, the alkenone unsaturation index (U K′ 37 ) ratio and surface water temperature were calculated via the equations (Eqs. 1 and 2): Temperature ¼ ðU K 0 37 À 0:044Þ=0:033 ð2Þ here C 37:2 and C 37:3 are the concentrations of diunsaturated and triunsaturated C 37 alkenones 60,61 . Group II alkenones were not applied for temperature reconstruction due to the uncertain temperature response induced by phylogenetic diversity 62 . Analytical uncertainties for lab standards are typically less than 5% for C 37 alkenone concentration, and 0.005 units for the U K′ 37 ratio (equivalent tõ 0.2°C).
Carbonate carbon isotope (δ 13 C carb ) analysis. δ 13 C carb analysis for the DHG and HG sections followed the procedure described in ref. 39 , with δ 18 O carb from the same analysis at the DHG section reported previously 24 . In brief, samples (~5 g each) were first grounded in an agate mortar and sieved through a 100 mesh (150 μm) screen. Isotope analysis was achieved using an isotope ratio mass spectrometer (MAT-252, Finnigan) with an automated carbonate preparation device (Kiel II) at the Institute of Earth Environment, Chinese Academy of Sciences. Laboratory standard carbonates (TTB1) with known δ 13 C carb were analyzed daily to ensure instrumental accuracy. International NBS18 and NBS19 standards were used to monitor the long-term instrumental stability. All δ 13 C carb values are reported as per mil deviation relative to the Vienna PeeDee Belemnite (VPDB) standard. Analytical precision for repeated analyses of laboratory standards is generally within ± 0.1‰. The Z value 42 is calculated from carbonate δ 13 C and δ 18 O values (Eq. 3): Leaf wax hydrogen isotope (δD alk ) analysis. Long-chain n-alkane δD alk for the DHG and HG sections were analyzed using a coupled GC-Delta-V isotope ratio mass spectrometer (IRMS) system (Thermo Finnigan) at the Institute of Earth Environment, Chinese Academy of Sciences. Diagenetic effects and aquatic influences on long-chain n-alkanes are assessed to be minor 24 . To further minimize their combined effects, samples with low carbon preferential index (CPI < 2) and high proportion of aquatic plants (Paq > 0.3) values were excluded for analysis, mostly occurring in the early interval of the sections 63 . Detailed hydrogen isotope analysis procedure can be found in ref. 64 . Briefly, approximately 300 ng of targeted long-chain n-alkane (C 27 , C 29 and C 31 ) was injected in splitless mode, with the flow rate of carrier gas (He) at 1.0 ml min −1 . n-Alkanes were separated by GC and then pyrolyzed into H 2 at 1430°C before transporting into the IRMS for δD alk analysis. The variability of H 3 + factor was <0.1 ppm nA −1 day −1 during our sample analysis, based on daily measurement using H 2 reference gas. The in-house standard nalkane mixture was measured at every four samples to ensure that the mass spectrometer was stable. δD values of n-alkane standards are close to their supposed/actual values with the ion current intensity at 4000-6000 mV in our lab. We thus reported measured δD values with the current intensity within this range. All δD alk values are expressed in the delta (δ) notation relative to the Vienna Standard Mean Ocean Water (VSMOW) and typical precision for repeated standard analyses is generally better than ± 3‰. δD alk reported here is the average of δD values of C 27 , C 29 and C 31 n-alkanes.

Data availability
The new data reported in this study (i.