Based on the results of two CONISS ordinations combined with the taxonomic and quantitative composition of the palynological assemblage (see Sect. 2; Figs. 3, S1, S2), the succession was divided into three zones (I, II, and III) of which Zone I was further divided into three subzones (a, b, and c), all of which demonstrate unique vegetation dynamics within that zone. Important trends for each zone and subzone are described below. The zone boundaries are positioned at the upper limit of the samples that mark each boundary. A complete overview of the raw counts, percentages, and arithmetic means are given in the Supplement.

Age constraints for the RZ section are provided by the K–Ar ages from shoshonitic lavas and felsic and porphyry intrusions that are either interbedded with, or unconformably overlie, the lacustrine to alluvial Nangqian strata. Emplacement ages across the Nangqian Basin vary between 32.04 and 36.5 Ma (Deng et al., 1999); 37.0±0.2 Ma and 38.2±0.1 Ma (Spurlin et al., 2005); 37.1 and 37.8 Ma (Zhu et al., 2006); and 35.6±0.3 and 39.5±0.3 Ma (Xu et al., 2016). In the latter study, zircon U–Pb age data were derived from felsic intrusions sampled at two localities in the Nangqian Basin (Boza and Nangqian). The syenite porphyries from the Boza area (further south of the RZ section) show an emplacement age of 35.58±0.33 Ma, whereas the monzonite porphyries from the Nangqian area (just south-east of the RZ section) have older magmatic emplacement ages, ranging from 39.5±0.3 Ma to 37.4±0.3 Ma. As this age range is broadly coeval with the age of the mafic volcanic rocks in the Nangqian Basin (37.0–38.2 Ma; Spurlin et al., 2005) as well as the age range obtained by Zhu et al. (2006), here we consider ∼37–38 Ma to represent a minimum age for the RZ section. This is also congruent with palynological evidence for the overall age of the sampled strata (Fig. 4), which is discussed in more detail below.

The assemblage from the RZ section is very similar to those from the Yang Ala section in the Nangqian Basin, dated as late Eocene (Yuan et al., 2017), the Eocene Wuqia assemblage (site 98) from the west Tarim Basin (Wang et al., 1990a, b), the late-middle Eocene to late Eocene assemblage from the upper Niubao Formation, the Lunpola Basin (Song and Liu, 1982; J. G. Li et al., 2019), and the Bartonian (41.2–37.8 Ma) part of the palynological record in the Xining Basin (Dupont-Nivet et al., 2008; Hoorn et al., 2012; Han et al., 2016). Specifically, the absence of Classopollis, Exesipollenites, and Cycadopites combined with the predominance of Nitrariadites/Nitraripollis and Ephedripites pollen, and the presence of the middle Eocene–Neogene genus Fupinggopollenites (Liu, 1985), indicates that the RZ section cannot be older than middle Eocene (Fig. 4). It is also unlikely to be of latest Eocene age or younger due to the lack of significant conifers that become more common approaching the Eocene–Oligocene Transition (Hoorn et al., 2012; Page et al., 2019; Fig. 4). Specific ranges and abundance patterns of these and other key taxa within Eocene Tibetan basins (Figs. 4 and 5) enable the age of the section to be better constrained, which is explored in greater detail below.

Ephedra is a gymnosperm shrub with the oldest macrofossils from the Early Cretaceous (Bolinder et al., 2016; Han et al., 2016), but the genus is probably older, dating to the Triassic (Yang, 2002; Sun and Wang, 2005) or even the Permian (Wang, 2004) based on the ephedroid pollen record. Its current distribution is limited primarily to arid and semi-arid regions of the world (Stanley et al., 2001), and the fossil pollen representative, Ephedripites, is widespread in Cenozoic evaporates, indicating the xerophytic nature of this genus (Sun and Wang, 2005). The Xining Basin in northern Tibet records a particularly time-extensive section with good age control (Dupont-Nivet et al., 2008; Hoorn et al., 2012; Meijer et al., 2019) that reveals a detailed pattern of changes in Ephedripites pollen during the middle–late Eocene. After 38.8 Ma, Ephedripites comprised ca. 20 %–60 % of the total palynological composition in the Xining Basin, with a predominance of the derived type, Ephedripites (Distachyapites) (Han et al., 2016). Prior to this (ca. 41–38.8 Ma), the record comprised a mix of the derived type; the ancestral type, Ephedripites (Ephedripites); and another ephedroid genus, Steevesipollenites (Han et al., 2016; Bolinder et al., 2016). A similar pattern is observed in the Nangqian Basin, with a spike of the ancestral type of Ephedra only recorded in Zone II and not observed in the rest of the RZ section or elsewhere in the Nangqian Basin (Yuan et al., 2017). This suggests a correlation between Zone I of the Xining Basin and Zone II of the RZ section (Fig. 5). As it is possible that the change in Ephedripites diversity may not have occurred across Tibet simultaneously (i.e. at ∼39 Ma), we suggest that this most likely constrains the age of the RZ section to late Eocene (Bartonian; 41.2–37.8 Ma).

In addition to the proportions of the ancestral vs. derived type of Ephedripites, a significant spike in tropical forest pollen in one sample at this time, combined with a large decrease in steppe–desert pollen, suggests that Zone II of the RZ section reflects a temporary warming interval in the Eocene. Although the increase in tropical forest taxa in one sample from this zone does not indicate an actual biome shift in the Nangqian region from “steppe” to “tropical forest”, it suggests a change in regional climate through increased input of regional tropical taxa. This could possibly be concurrent with the MECO (∼40 Ma), a transient warming event that preceded rapid aridification in Central Asia (driven primarily by proto-Paratethys Sea retreat; Kaya et al., 2019). This interval is followed by a change in lithofacies (decreasing thickness of gypsum beds) and an increase in steppe–desert pollen records in north-western China (Bosboom et al., 2014). Similar trends are also observed in the Nangqian Basin (Fig. 3), suggesting a possible correlation. However, it must be considered that the upper zones of the RZ section yielded a low number of samples (zones II and III each comprise only two samples), and the tropical forest spike is only present in one of these samples. This places statistical limitations on the interpretations that can be drawn; therefore, further investigations should be made in Nangqian and other parts of Tibet to corroborate this finding. Accordingly, for the moment we do not date the RZ section on the basis of a tentative correlation to the MECO at ∼40 Ma; however, available evidence does suggest that the spike of tropical forest represents a temporary shift in regional climate. The palynomorphs from these samples were not degraded nor compressed to a greater degree than palynomorphs from the rest of the section, and they were of a similar colour and appearance, suggesting it is unlikely that the pollen in Zone II represents reworking or contamination. Furthermore, the increase in tropical forest taxa is accompanied by a large decrease in steppe–desert pollen which is not observed in the other zones of this section (average 9 % steppe–desert pollen in Zone II vs. 38 % in Zone I and 32 % in Zone III), nor later in the Eocene in the Nangqian Basin (Yuan et al., 2017). This further indicates a temporary shift in the regional climate to warmer and wetter at this time.

In northern Tibet, Pinaceae (conifers) abruptly increased in the palynological record at 36.55 Ma (Page et al., 2019), which is not observed in the RZ section. The rare conifers in this latter assemblage are in accordance with the minimum depositional age constraints of ∼ 37–38 Ma from overlying volcanic rocks. In conjunction with the palynostratigraphic correlations from across Tibet (Fig. 4), as well as the change in the proportions of the ancestral vs. derived type of Ephedripites (Fig. 5), the age of the complete section is proposed to be Bartonian (41.2–37.8 Ma; Figs. 4, 5).

The RZ section records three distinct palaeofloras in east-central Tibet that evolved in response to changing climate in the Eocene (Fig. 6). During deposition of Zone I, the climate was warm, and vegetation was characterized by steppe–desert shrubs, diverse ferns, and a lesser component of temperate and warm broad-leaved forest. Interestingly, prominent vegetation groups with very different moisture requirements existed within a limited distance of each another in the Nangqian area. A very diverse and abundant pteridophyte (fern) community as well as conifers such as Taxodiacites and Tsugaepollenites would have required higher humidity (e.g. Kotthoff et al., 2014), but the abundant halophytic and xerophytic steppe–desert vegetation would likely only have been competitive in arid environments. The dominant plants belonging to these salt- and drought-tolerant groups (Nitraria and Ephedra) grow today in Central Asian regions with a mean annual precipitation (MAP) of 100 mm or less, and they are also associated with arid palaeoenvironments through the Cenozoic (Sun and Wang, 2005). Although the conifers (produced by cypress and Tsuga) could have been windblown from further distances, the coexistence of such diverse and abundant ferns and steppe–desert vegetation in the landscape (PFTs with opposing moisture requirements for competitiveness) has not been observed in other Tibetan basins to date (Miao et al., 2016, Table 1); therefore, it does not seem to reflect conventional spatial patterning of less water-dependant vegetation growing upland. Rather, it may suggest an environment with strongly seasonal precipitation that would favour lush vegetation growth for a restricted interval and, alternately, xerophytic vegetation during the dry season.

Based on a comparison of existing palynofloral records with our new section, the northern regions of the plateau (Tarim, Qaidam, Hoh Xil, and Xining basins) were already significantly more arid than the central TP in the middle Eocene, having hosted greater proportions of xerophytic plants (Fig. 4). Therefore, precipitation in the greater Nangqian region would have been unlikely to derive from the westerlies, which served as the dominant moisture source northward of the central TP since at least the early Eocene (Caves et al., 2015). This suggests that the central TP could have instead been influenced by a southern monsoon system similar to the modern I-AM in the middle–late Eocene, although not to the degree experienced by southern Tibet, which hosted greater proportions of forest and was likely more humid (e.g. Jianchuan Basin; Fig. 4). However, it should be borne in mind that rainfall seasonality is not always a proxy for the existence of monsoons; although leaf form is the preferred method for detecting monsoons in deep time climates (Spicer, 2017), the absence of well-preserved fossil leaf assemblages from the Nangqian Basin to date prevents this comparison. Furthermore, palynological records alone are not sufficient for detecting whether the nature of monsoons in the Eocene was more similar to the present I-AM or South Asian monsoon (SAM), which contributes mostly to the moisture in the Nangqian region today (L. Li et al., 2019).

Our results indicate that the brief warming interval recorded in Zone II prompted a considerable change in the vegetation in east-central Tibet, encouraging the temporary spread of (dry) forests in the region, while steppe–desert vegetation contracted. Warming is reflected by an atypical spike in tropical forest, while a warm broad-leaved forest spike in north-eastern Tibet is coincident with the MECO (Hoorn et al., 2012; tropical forest is exceedingly rare in the latter area during the middle–late Eocene). In order to estimate relative humidity in arid environments such as these, the Nitraria / Ephedra (N/E) ratio can be used to distinguish between desert/semi-desert (<1) and steppe–desert (>1; Li et al., 2005; Hoorn et al., 2012). Although both genera occupy arid environments today, Ephedra is currently distributed primarily throughout deserts, semi-deserts, and grasslands globally (Stanley et al., 2001), while Nitraria is a relatively more humid steppe–desert taxon (Cour et al., 1999; Sun and Wang, 2005; Jiang and Ding, 2008; Li et al., 2009; Zhao and Herzschuh, 2009).

In the RZ section, the proportion of temperate broad-leaved forest in relation to warm broad-leaf and tropical forest became much greater in the upper part (Fig. 3), indicating a cooler climate in the late Eocene, which matches cooling trends recorded by clumped isotopes both in the Nangqian Basin (L. Li et al., 2019) and in the Xining Basin (Page et al., 2019). Importantly, the N/E ratio in the RZ section is lowest immediately following the warming interval in Zone II (Fig. 3) and persists for an extended period, indicating rapid, prolonged aridification. An overall expansion of steppe–desert vegetation is observed in Zone III, corresponding to patterns observed on the north-eastern TP in the late Eocene (Hoorn et al., 2012; Bosboom et al., 2014). Accordingly, our vegetation results have implications for understanding the importance and extent of aridification across Central Asia in the late Eocene, which was primarily driven by proto-Paratethys Sea regression (Kaya et al., 2019). Ecosystem responses to this event on both the north-eastern and east-central parts of the TP demonstrates that aridification across the Asian continental interior in the late Eocene could have been further-reaching than previously thought. Our findings show that after sea regression, westerly moisture supply carried from the proto-Paratethys Sea was reduced as far as central Tibet. This provides further support for the argument that this sea was a major source of moisture for the Asian interior and, thus, a primary driver of Central Asian climate during the Eocene (Bosboom et al., 2014; Bougeois et al., 2018; Kaya et al., 2019; Meijer et al., 2019).

Long-term aridification in the late Eocene exerted further influence on the vegetational composition in east-central Tibet with regards to the proportions of the ancestral vs. derived types of Ephedripites. In modern and Quaternary settings, this has been developed as a ratio to distinguish between desert and steppe–desert environments, termed the Ephedra fragilis-type sensu lato/Ephedra distachya-type (Ef / Ed) ratio (whereby E. fragilis represents the ancestral type and E. distachya represents the derived type; Fig. 5). Tarasov et al. (1998) found the E. fragilis-type sensu lato to be common in arid climates with mean temperatures of the warmest month above 22 ∘C. Herzschuh et al. (2004) applied the Ef / Ed ratio to Holocene pollen spectra from the Alashan Plateau and tested its reliability with a regional modern pollen dataset, finding Ef / Ed ratios >10 in most samples from desert sites, and values <5 in most samples from the sites with more favourable climates (e.g. forest-steppe, steppe, and alpine meadow).

In the middle–late Eocene of Central Asia, the ancestral type of Ephedripites never comprises more than 25 % of the ephedroid pollen sum in north-eastern Tibet, whereas the derived type makes up at least 60 % (Xining Basin; Han et al., 2016 and Qaidam Basin; Zhu et al., 1985; Miao et al., 2013a; Jiuquan Basin; Miao et al., 2008); this also appears true for north-western Tibet (Tarim Basin; Wang et al., 1990b; Hoh Xil Basin; Miao et al., 2016) and east-central Tibet (Yuan et al., 2017; this study). Therefore, Ef / Ed ratios >10 (supposedly indicative of desert ecosystems) are never observed, despite the N/E ratio indicating the regular existence of deserts or semi-deserts in northern Tibet (Zhu et al., 1985; Hoorn et al., 2012; Miao et al., 2016), and central Tibet (Yuan et al., 2017; this study) in the Paleogene. Sedimentological evidence suggests the N/E ratio to be more reliable for these deep time environments, with Nitraria and Ephedra pollen being widely distributed in evaporites and red beds, indicating deposition in arid or semi-arid climates (Sun and Wang, 2005). Therefore, while pollen ratios appear to reflect reliable functions of climate and landscape change for modern and Holocene settings (Li et al., 2010), our results identify possible contradictions between the N/E and Ef / Ed pollen ratios. This indicates that further verification of these pollen ratios in modern settings and across larger spatial scales is necessary for reliable palaeoenvironmental reconstructions in deep time.

A comparison of palynological assemblages across the Qinghai–Tibetan region indicates that vegetation has changed markedly from the Paleogene to the present (Fig. 4). While the Nangqian region was dominated by steppe–desert shrubs in the past, it now hosts primarily alpine biomes, as do the Hoh Xil and Xining basins. In contrast, the Tarim and Qaidam basins are now significantly more arid than in the Eocene, and forest- and shrub-steppe have been replaced with desert vegetation (Fig. 4). The Jianchuan Basin to the south was dominated by mixed tropical–subtropical coniferous and broad-leaved forest (Wu et al., 2018), and it is also forested today (but with species of a less thermophilic nature). Similarly, the Markam and Gonjo basins host alpine meadow and forest today; although detailed palynological records have not yet been recovered, macrobotanical fossils suggest these areas were dominated by mixed broad-leaved and coniferous forest in the late Eocene–early Oligocene (Su et al., 2018; Studnicki-Gizbert et al., 2008). The above changes indicate that late Paleogene and Neogene topographic growth (creating new high-elevation biomes; Fig. 1a, b), the aridification of inner Asia (Caves et al., 2014, 2016), and global cooling (Zachos et al., 2001; DeConto and Pollard, 2003; Pagani et al., 2011) were all drivers of Cenozoic vegetation shifts across the TP.

High-altitude conifers are rare in this particular record, although the high-elevation genus Tsugaepollenites (Fauquette et al., 2006) is present. This could be driven by four possible factors: (1) taphonomy, i.e. the assemblage has a high proportion of autochthonous spores and pollen with little input from the peripheral mountains; (2) the elevation of this region was relatively low in the middle–late Eocene (<3000 m as proposed by Botsyun et al., 2019; also see Wei et al., 2016); (3) due to the generally wetter climate in relation to the north-eastern plateau basins, conifers are not competitive and surrounding mountains are instead forested by temperate angiosperms; and (4) central Tibet recorded regional pollen transported by different atmospheric circulation systems.

Regarding the first possibility, conifers are windblown and can be transported long distances (Lu et al., 2008; Ma et al., 2008; Zhou and Li, 2011); as the region already likely experienced a monsoonal climate (Spicer, 2017; Licht et al., 2014; Caves Rugenstein and Chamberlain, 2018; this study), we consider it unlikely that our assemblages record little to no regional vegetation. The second factor, the elevation history of the TP, is a controversial topic of discussion, and palynological evidence from the RZ section does not provide strong support either for or against a relatively low middle–late Eocene palaeoaltitude in the region. Although the upper part of the RZ section in the Nangqian Basin likely just predates the high-elevation signal further to the north from 37 Ma onwards (Dupont-Nivet et al., 2008; Hoorn et al., 2012; Page et al., 2019), an expanding body of data indicates that a proto-Tibetan highland with complex topography was already in place during the Paleogene (Xu et al., 2013; Ding et al., 2014; Wang et al., 2014; Valdes et al., 2019).

Isotopic evidence suggests moderate to high elevations for the Nangqian Basin in the late Eocene (valley floor 2.7 (+0.6/-0.4) km above sea level; surrounding mountains 3.0±1.1 km above sea level; Li L. et al., 2019). In the adjacent Gonjo Basin, stable isotope data suggest the basin had already attained 2100–2500 m palaeoelevation by the early Eocene (Tang et al., 2017). Some of the broad-leaved angiosperms trees present in the new Nangqian assemblage could have grown at maximum elevations of 3600–4000 m during the Eocene (Ilex, Quercus: Song et al., 2010); therefore their presence in lieu of abundant conifers is not in contradiction with an elevated topography in parts of east-central Tibet at this time. This has significance for other Asian palynological studies that infer regional palaeoaltitudes and uplift history of Tibet based solely on palynological records from a single locality: a multi-proxy approach is clearly necessary to address the complex history of Tibetan uplift in future research.

Palynological data from the RZ assemblage supports climate (the third possibility) rather than altitude as a primary driving factor of vegetational composition: locally wetter conditions in the east-central region of the TP (see Sect. 5.2) would likely have promoted angiosperm tree growth over cold-temperate conifers that can withstand drought better, and utilize a winter wet growing season unlike deciduous angiosperms (Dupont-Nivet et al., 2008; Hoorn et al., 2012; Page et al., 2019). The last possibility is also supported, with the palynology of this study suggesting that central Tibet was influenced by two atmospheric circulation systems: predominantly the westerlies from the north (Caves Rugenstein and Chamberlain, 2018), and (to a limited degree) by a southern monsoon, which could conceivably also have transported windblown pollen from subtropical and warm-temperate broad-leaved forests in the south (Su et al., 2018). Today, the Nangqian region receives nearly 70 % of its moisture from the SAM, with the westerlies from the north making up the remainder (Li L. et al., 2019). This indicates that atmospheric circulation systems have changed considerably in east-central Tibet from the Paleogene to Neogene, despite the existence of monsoons in this region since at least the Eocene (Licht et al., 2014; Caves Rugenstein and Chamberlain, 2018; Spicer, 2017). Based on the above, we propose that both local climatic conditions and the influence of different regional atmospheric circulation systems contributed to the development of a unique floral ecosystem in east-central Tibet during the late Eocene.