The Kobresia pygmaea ecosystem of the Tibetan highlands – origin, functioning and degradation of the world’s largest pastoral alpine ecosystem

Kobresia pastures in the eastern Tibetan highlands occupy 450000 km2 and form the world’s largest pastoral alpine ecosystem. The main constituent is an endemic dwarf sedge, Kobresia pygmaea, which forms a lawn with a durable turf cover anchored by a felty root mat, and occurs from 3000 m to nearly 6000 m a.s.l. The existence and functioning of this unique ecosystem and its turf cover have not yet been explained against a backdrop of natural and anthropogenic factors, and thus its origin, drivers, vulnerability or resilience remain largely unknown. Here we present a review on ecosystem diversity, reproduction and ecology of the key species, pasture health, cycles of carbon (C), water and nutrients, and on the paleo-environment. The methods employed include molecular analysis, grazing exclusion, measurements with micro-lysimeters and gas exchange chambers, 13C and 15N labelling, eddy-covariance flux measurements, remote sensing and atmospheric modelling. The following combination of traits makes Kobresia pygmaea resilient and highly competitive: dwarf habit, predominantly below-ground allocation of photo assimilates, mixed reproduction strategy with both seed production and clonal growth, and high genetic diversity. Growth of Kobresia pastures is co-limited by low rainfall during the short growing season and livestock-mediated nutrient withdrawal. Overstocking has caused pasture degradation and soil deterioration, yet the extent remains debated. In addition, we newly describe natural autocyclic processes of turf erosion initiated through polygonal cracking of the turf cover, and accelerated by soil-dwelling endemic small mammals. The major consequences of the deterioration of the vegetation cover and its turf include: (1) the release of large amounts of C and nutrients and (2) earlier diurnal formation of clouds resulting in (3) decreased surface temperatures with (4) likely consequences for atmospheric circulation on large regional and, possibly global, scales. Paleo-environmental reconstruction, in conjunction with grazing experiments, suggests that the present grazing lawns of Kobresia pygmaea are synanthropic and may have existed since the onset of pastoralism. The traditional migratory rangeland management was sustainable over millennia and possibly still offers the best strategy to conserve, and possibly increase, the C stocks in the Kobresia turf, as well as its importance for climate regulation.


INTRODUCTION
The Tibetan highlands encompass 90% of the Earth's terrain above 4000 m and host the world's largest pastoral alpine ecosystem: the Kobresia pastures of the southeastern highlands, with an area of 450000 km² (Fig. 1). This ecosystem is globally unique as it is: (1) dominated by a single endemic sedge species of 1 to 4 cm in height; (2) forms a golf-course like lawn, with a very durable turf cover anchored by a felty root mat; (3) occurs over an elevational extent of 3000 m, stretching between 3000 m (in the north-eastern highlands) to nearly 6000 m (on the north slope of Mt. Everest ;Miehe 1989, Miehe et al. 2008b).
The evolution and recent changes of the Kobresia ecosystem, as well as its future development, are of great importance because surface properties of the highlands have an undisputed effect on the global climate (Cui and Graf 2009, Babel et al. 2014. The Tibetan plateau is significant in terms of carbon (C) turnover and CO2 fluxes from regional to global scales ). Its soil stores huge C amounts, mounting up to 2.5% to the global C pool (Ni 2002, Hafner et al. 2012. Of these, the Kobresia ecosystem in particular contributes about 1% (Batjes 1996, though covering only about 0.3% of the global land surface (equivalent to one-quarter of the whole Tibetan plateau).
The livelihoods of 5 million pastoralists depend on forage resources from the rangelands, which sustain about 13 million yak, and 30 million goats and sheep (Wiener et al. 2003, Suttie et al. 2005. One quarter of the world's population living in the surrounding lowlands ultimately depend on ecosystem functions of the Kobresia mats, which represent the upper catchment areas of the Huang He, Yangtze, Salween, Mekong, and partly of the Brahmaputra, rivers.
FIG. 1. Kobresia pygmaea pastures of the Tibetan highlands and forest relics. After Atlas Tibet Plateau 1990, Miehe et al. 2008b, Babel et al. 2014 The aim of this review is therefore to summarize recent findings relating to: (1) the diversity and the distribution of plant species of the ecosystem and its paleoecological background; (2) the ecology and reproduction of the dominant species, Kobresia pygmaea (C.B.Clarke) C.B.Clarke; (3) the ecosystem's water budget and hydrological fluxes; (4) fluxes in the carbon cycle; (5) soil properties and functions, including productivity and nutrients stocks; (6) the main causes of current rangeland degradation and its consequences; (7) the human impact shaping this ecosystem; (8) the current understanding of the age of this human impact.
In addition, a new concept of natural autocyclic processes of turf erosion, initiated through polygonal cracking of the turf cover increased by overgrazing and facilitated by soil-dwelling endemic small mammals, will be presented.

Species diversity and distribution
The Tibetan highlands are a center of Cyperaceae diversity (Global Carex Group 2015), with more than 30 Kobresia species (Zhang and Noltie 2010). Recent  Two species, K. pygmaea and K. yadongensis Y.C. Yang reach hardly 4 cm in height. The latter has been described recently and its distribution is poorly known Noltie 2010, Miehe et al. 2011a). By far the most important species of the Tibetan rangelands is K. pygmaea. It was first collected in 1847 by Thomas Thomson in Rupshu (NW India), and is endemic to Tibet and the Inner Himalayas, ranging from the Deosai Plains of northern Pakistan to the Yulong Shan in northern Yunnan (Dickoré 1995, Zhang andNoltie 2010). The interspecific relationships between species as currently delimited remain unknown, because standard DNA for phylogenetic analysis (nuclear ITS and chloroplast DNA regions, including matk, trnl-F and trnC-D) show low mutation rates and little divergence within Kobresia (J. Liu [personal communication]).
The weak genetic differentiation suggests that these morphologically defined species may have evolved rapidly within the recent past. Further calibration, based on genomic data and fossils are needed to clarify evolutionary relationships among the Kobresia species as currently defined.
Vascular plant α-diversity of alpine Kobresia pastures (measured in plots of 100 m 2 ) varies between 10 species in closed lawns with a Kobresia pygmaea cover of 98% (Miehe et al. 2008b, E. Seeber [personal communication]), and, in the eastern part of the plateau, more than 40 species in communities with mosaics of Kobresia patches and grasses, other sedges and perennial forbs growing as rosettes and cushions . Similar levels of richness were recorded in montane grazing lawns hosting a set of grazing weeds (Miehe et al. 2011c). The inter-annual variability of species richness is potentially high, depending on the variability of annual herbs in response to interannual changes in precipitation (E. Seeber [personal communication]).
In general, the Tibetan highlands, and specifically the eastern plateau, are poor in endemic plant genera and rich in endemic plant species (Wu et al. 1981). In contrast, and peculiar to the alpine Kobresia mosaics, is the high number of endemic monotypic genera of rosette plants, which colonize open soils around small mammal burrows (e.g., Microcaryum pygmaeum I.M. Johnst., Przewalskia tangutica Maxim., Pomatosace filicula Maxim.; Miehe and Miehe 2000).
The fauna also comprises several endemics, which are unevenly distributed among taxonomic and ecological groups. While most large predators are not endemic, a number of herbivorous mammals are, including wild yak, chiru and kiang, but also small mammals such as the plateau pika (Schaller 1998). In contrast, the beetle soil-fauna, which generally has a very high diversity in the Tibetan highlands (for example along wet gullies), is very poor and endemics are absent from the grazing lawns. Apparently the poor aeration of the turf, and the soil compaction due to trampling effects, are not suitable for the strictly edaphic ground beetle larvae (J. Schmidt [personal communication]). Important insects of the Kobresia pastures are Lepidoptera caterpillars of the genus Gynaephora (Erebidae, Lymantriinae; e.g., Yuan et al. 2015) which are known to cause severe damage (http://www.fao.org/ag/agp/agpc/doc/counprof/china/china2.htm, Yan et al. 1995, Xi et al. 2013, Zhang and Yuan 2013, Yuan et al. 2015

The paleo-ecological background
The development of biodiversity and endemism in relation to the highlands' uplift and climate history is still under debate, hampering our understanding of the present diversity patterns. It is likely, however, that today's evolutionary histories differ strongly between taxa and their traits such as climatic niches and dispersal ability. These two traits are especially relevant with respect Pleistocene climatic changes and likelihood of extinction. Table 1 reviews the available data on independent climate proxies for the Last Glacial Maximum (LGM; see Table 1). Most probably the world's most elevated highlands were even more hostile for life during the cold periods of the Quaternary than today. The former idea of a complete ice cover across the highlands (Kuhle 2001) has, however, been soundly rejected on the basis of biotic and abiotic proxies. It was replaced by a concept of fragmented but locally extensive mountain glaciations, at least in the more humid eastern highlands (Shi 2006, Heyman et al. 2009). LGM  Kathmandu, Nepal Pinus-Pollen 3 (annual) LGM  Current paleo-scenarios for humidity and temperature still offer a wide range of possibilities, depending on the proxies on which they rely. The two most divergent paleo-scenariosecological stability or complete extinction and Holocene re-migration can be assessed against the presence or absence of local endemics or populations with private haplotypes , because 10000 years since the LGM is too short a period for in-situ speciation.
In the case of the Tibetan highlands summer temperatures are particularly important.
If these drop below minimum requirements for a given species, then local and endemic populations are losta phenomenon most important in closed-basin systems. The huge number of tiny, wingless, locally endemic ground beetles in the southern highlands is important in this respect, because it testifies to the persistence of alpine habitats throughout the Pleistocene and the absence of a glacial 'tabula rasa' (Schmidt 2011). The mean altitudinal lapse rate for summer temperatures across 85 climate stations (records between 1950 and 1980) in the highlands is 0.57 K per 100 m . Biogeographical data of the current distribution of these wingless ground beetles in the mountainous topography of southern Tibet point to a decline in LGM summer temperature of only 3-4 K as compared with present conditions , which is much lower than earlier estimates (Table 1). This is supported by

LIFE HISTORY TRAITS AND REPRODUCTION OF KOBRESIA PYGMAEA
Kobresia pygmaea is one of the smallest alpine sedges, yet dominates the largest alpine pastoral ecosystem, stretching over an altitudinal range of 3000 m. Its lawns have an estimated Leaf Area Index (LAI) of only ~1 ) and a roughness length of only about 3 mm. They represent a vegetation cover with a small transpiring surface and low aerodynamic resistance to the atmosphere (Babel et al. 2014). Due to high solar radiation input and night-time long-wave radiation, Kobresia pygmaea has to cope with steep soil-to-air temperature gradients and high leaf over-temperatures on sunny days. However, the species shows low sensitivity to low soil temperatures, as leaf gas exchange was found to be negatively affected only by soil temperatures below freezing point, i.e., when soil-water availability approaches zero (Coners et al. 2016).
In fully sun-exposed sites, Kobresia pygmaea grows mostly in lawns of 2 cm in height, yet lawns of only 1 cm or up to 4 cm can occur locally, regardless of being grazed or ungrazed, and whether growing in water-surplus sites such as swamps or on steep, Although the lamps used were chosen for their spectral similarity to sunlight, after some weeks Kobresia pygmaea plants reached up to 20 cm in height. Artificial lighting cannot reach the peak intensities in radiation pertaining on the plateau (especially in the UV-B part of the spectrum), which are among the highest occurring on earth (Beckmann et al. 2014). Whether permanently high radiation loads represent an inhibitor with a photo-toxic effect that reduces the growth of Kobresia, remains unclear; the same is true for effects of wind and desiccation. No doubt, the dwarf habit could also have evolved as an effective strategy to avoid plant biomass loss and damage by grazing.
A microsatellite-based survey of alpine Kobresia pygmaea populations revealed the presence of large (>2 m 2 ) clones, and an overall high genetic diversity (with more than 10 genets/m 2 ). This probably results from consecutive events of sexual recruitment under favorable conditions, coupled with extensive periods of vegetative persistence . Thus, in contrast to many alpine species that largely abandon sexual reproduction (Steinger et al. 1996, Bauert et al. 1998, K. pygmaea benefits from a mixed reproduction strategy with clonal growth ensuring long-term persistence and competitiveness, while intermittent reseeding facilitates colonization and genetic recombination. A diaspore bank supports short-term persistence. According to microsatellite studies, populations are hardly separated between montane and alpine altitudes, indicating high gene flow within the species' distribution range (Seeber 2015).
The genetic diversity in the species is also promoted by polyploidy; K. pygmaea mostly has 2n = 4x = 64 chromosomes and is tetraploid. Ploidy of congeners range from dito hexa-or even heptaploidy (Seeber et al. 2014). An assessment of ploidy levels along an elevational gradient in Qinghai also revealed one diploid, one triploid, two octoploid and one dodecaploid individual, indicating ongoing chromosomal genetic evolution in dispersal. Germination rates differ greatly between seeds from alpine and montane environments (e.g., Li et al. 1996, Deng et al. 2002, Miao et al. 2008, Huang et al. 2009) but are generally low, both in laboratory experiments as well as in situ. Huang et al. (2009) reported 13% germination of untreated diaspores, while most studies obtained no seedlings at all (Li et al. 1996, Miao et al. 2008). The water-impermeable pericarp hinders germination, and its removal by chemical or mechanical interventions increases germination rates. Under natural conditions, microbial activity or digestion by herbivores may have the same effect and, consequently, increase germination. Nonetheless, the variability in germination was high between different studies, which may reflect population responses to different abiotic conditions, such as nutrient availability or elevation (Amen 1966, Seeber 2015.

WATER BUDGET AND HYDROLOGICAL FLUXES OF ECOSYSTEMS
Kobresia pastures exist over a wide range of precipitation from 300 to 1000 mm/yr (Miehe et al. 2008b), which falls nearly exclusively during summer. Apart from nutrient limitation (see below), the growth of Kobresia pygmaea is co-limited by low summer rainfall, at least under the conditions of the ecosystem's alpine core range (Kema; Coners et al. 2016) with 430 mm/yr and mean maximum temperature of the warmest month of 9.0 °C (1971Babel et al. 2014). Onset of the growing season is controlled by rainfall amount in early summer and under low-temperature control in autumn (September and October, depending on latitude and altitude). The leaf growth of K. pygmaea is not temperature-driven (unlike that of Androsace tapete Maxim., a cushion plant of the alpine steppe), but depends strictly on water availability (Li et al. 2016). Greening of the Kobresia pastures after the onset of the summer rains is well known; this usually occurs between mid-May and mid-June (Fig. 2). However, onset of the summer monsoon can be delayed by up to six weeks, sometimes starting as late as early August, with critical effects for livestock (Miehe and Miehe 2000). Despite its low leaf area index, alpine Kobresia pastures can reach high transpiration rates (up to 5 mm/day) in moist summer periods even at elevations >4000 m: a consequence of specific microclimatic conditions on the plateau which enhance evaporation (Coners et al. 2016). While dry periods, due to delayed onset of the summer rains, visibly hamper K. pygmaea growth (browning of the pastures), a constant daily irrigation with 2.5 mm/day or even 5 mm /day did not increase the total above-ground biomass production after 40-70 days (corresponding to 100-350 mm of were found to be 2.5 K higher than under a closed mat of 93% ). The recent  increase in surface soil temperatures in the Huang He headwaters of 0.6 K per decade has led to a drastic increase of the depth of the thawing layer (Xue et al. 2009), and to a deterioration of Kobresia tibetica swamps. In summary, overgrazing induces degradation of root mats and leads to changes in water cycle and balance at both local and regional levels; this may decrease the recovery of damaged Kobresia pastures.

THE CARBON CYCLE
Kobresia turf is a key component of the C stocks and cycle in these pastures. The root : shoot ratio ranges from 20 (Li et al. 2008, Unteregelsbacher et al. 2012, Schleuss et al. 2015 up to 90 (Ingrisch et al. 2015), depending on season, grazing intensity, and degradation stage. Soil organic carbon (SOC) storage within the root mat reaches up to 10 kg C/m 2 (Li et al. 2008, Unteregelsbacher et al. 2012, Schleuss et al. 2015, making up roughly 50% of the overall C stocks. Representing a highly significant C stock, the turf is also a highly active component of the C cycle, which receives the largest fraction of the photo assimilated C. For alpine Kobresia pastures at 4450 m, Ingrisch et al. (2015) showed that a large fraction of the assimilates is used for the build-up of new fine roots with a high turnover rate. The measured fluxes into belowground pools, mainly associated with the root and released later as soil CO2 efflux, were roughly twice as high as reported for pastures of K. humilis (C.A.Mey. ex Trautv.) Serg. (Wu et al. 2010) and montane K. pygmaea pastures at lower elevation (3440 m; Hafner et al. 2012). This emphasizes the importance of below-ground C allocation and cycling in alpine K. pygmaea pastures.
Net ecosystem exchange (NEE) measurements, derived by the eddy-covariance method, identified the alpine pastures at Kema as a weak C sink during the summer of 2010 with an ecosystem assimilation of 1.3 g C m -2 d -1 (Ingrisch et al. 2015), which is roughly 50% smaller than recorded in pastures at lower altitudes (e.g., Kato et al. 2004, Hirota et al. 2009), but agrees well with the results of a study by Fu et al. (2009) at a similar altitude. By combining the NEE measurements with 13 C -pulse labelling, Ingrisch et al. (2015) were able to estimate absolute C fluxes into the different ecosystem compartments during the main growing season. With a magnitude of 1 g C m -2 d -1 , the flux into below-ground pools was twice as high as the CO2 efflux from soil, and 10 times larger than the C flux into the above-ground biomass.
The key role of grazing for C sequestration and C cycling was demonstrated in two grazing exclusion experiments combined with 13 CO2 pulse labelling studies. The effect of grazing on the C cycle, specifically on differences in below-ground C stocks and C allocation, was shown for: (1) a montane Kobresia winter pasture of yaks, with moderate grazing regime, Short-term grazing exclusion in the alpine pasture affected only the phytomass of above-ground shoots, while neither C stocks nor assimilate allocation were altered (Ingrisch et al. 2015). In this system, roots and soil were of equal importance to C storage. By contrast, 7 years of grazing exclosure revealed that grazing is a major driver for below-ground C allocation and C sequestration in soils of montane Kobresia pastures (Hafner et al. 2012). Under a grazing regime, a higher fraction of assimilated C was allocated to below-ground pools; moreover, a larger amount was incorporated into roots and SOC. Fencing in contrast, led to a significant reduction of C sequestration in the soil and fostered turf degradation, emphasizing the key role of grazing for the biogeochemistry of these ecosystems.
Based on the long-term grazing exclosure experiments in montane pastures, we conclude that the larger below-ground C allocation of plants, the larger amount of recently assimilated C remaining in the soil, and the lower soil organic-matter derived CO2 efflux create a positive effect of moderate grazing on soil C input and C sequestration in the whole ecosystem. Due to the large size of the below-ground C stocks and the low productivity of the ecosystem, changes in the C stocks after cessation of grazing can be expected to take at least several years to become apparent.
However, the roots in the turf mat are a highly dynamic component of the C cycle, which might have implications for the interannual variability of the C budget even on the landscape scale. The C cycle appears to be largely driven by grazing, supporting the hypothesis of the pastoral origin of the Kobresia ecosystem.
At Kema research station, synchronous measurements with micro-lysimeters, gas exchange chambers, 13 C labelling, and eddy-covariance towers were combined with land-surface and atmospheric models, adapted to the highland conditions. This allowed analysis of how surface properties, notably the disintegration of the Kobresia sward (i.e., degradation stages), affect the water and C cycle of pastures on the landscape scale within this core region. The removal of the Kobresia turf fundamentally alters the C cycling in this alpine ecosystem and its capacity of acting as a C sink (Babel et al. 2014).

SOILS, PRODUCTIVITY, AND PLANT NUTRITION
Kobresia lawns typically produce a felty root mat (Afe, 'rhizomull'; Kaiser et al. 2008) up to 30 cm thick, which is situated on top of the predominant soils of Tibet's pasture , Baumann et al. 2009). The root mat consists of mineral particles, humified organic matter and large amounts of dead and living roots as well as rhizomes (Schleuss et al. 2015) The mat has typically formed in a loess layer of Holocene age (Lehmkuhl et al. 2000). The question still remains as to whether the are not directly plant-available. Prevailing low temperatures and moisture hamper the mineralization of root residues and slow down nutrient release to plant-available forms (Hobbie et al. 2002, Luo et al. 2004, Vitousek et al. 2010. From this perspective, the relatively close soil C/N and C/P ratios (Fig. 5) do not necessarily indicate an adequate N and P supply. Further, P can be precipitated in the form of calcium-phosphates due to the high abundance of exchangeable Ca.
Indeed multiple limitations of N and P constrain pasture productivity, which is shown by increased Kobresia growth following single and combined applications of N and P fertilizers. Even though single applications of either N or P favor above-ground biomass on most sites throughout the Kobresia ecosystem, the productivity strongly increases after combined N and P application (Fig. 6). This finding was also supported by three years of single and combined application of potassium (K), N and P in alpine Kobresia pastures at the Kema station. Nitrogen fertilization increased above-ground productivity about 1.2-1.6 times, while NP addition resulted in 1.5-2.4 times higher values, whereas no effect was found for the below-ground biomass (Seeber 2015).
Furthermore, fertilization increased the tissue content of N, P and K in K. pygmaea and in accompanying herbs. Overall, fertilization experiments clearly indicate that colimitations of N and P prevail in the Kobresia ecosystem. We conclude that Kobresia pygmaea has developed a dense root network to cope with nutrient limitations enabling productivity and competitive ability. The high below-ground biomass on the one hand ensures an efficient uptake of nutrients (shown by 15 N; Schleuss et al. 2015) at depths and times when nutrients are released via decomposition of SOM and dead roots; and on the other hand makes K. pygmaea highly competitive for mineral N acquisition in comparison with other plant species (Song et al. 2007) and microorganisms (Xu et al. 2011, Kuzyakov andXu 2013). Further, the enormous root biomass stores nutrients below-ground, protecting them from removal via grazing, which ensures fast regrowth following grazing events to cover the high below-ground C costs.
FIG. 6. Effects of single and combined fertilization with N and P at varying rates on aboveground biomass (AGB) extracted from 35 studies from all over the Kobresia ecosystem. Shown are Whisker-Box-Plots with outliers (white circles) and median (black line in the box) for low, moderate and high application rates (for N: low = 0-25 kg/ha*yr, moderate = 25-50 kg/ha*yr, high >50 kg/ha*yr; for P: moderate = 0-50 kg/ha*yr, high >50 kg/ha*yr; for N+P: moderate = 0-50 + 0-50 kg/ha*yr, high >50+50 kg/ha*yr). The dashed line indicates no effects, with negative effects on the left and positive effects following fertilization on the right.
That considerable amounts of resources were allocated and stored below-ground was

PASTURE HEALTH AND DEGRADATION
The term 'degradation' can refer not only to widespread negative effects of rangeland management, but also to natural processes of ecosystem disturbance that are often poorly understood. On the Tibetan highlands, degradation is by no means equally distributed; it is more severe (1)  Snowstorms introduce another form of climate variability and may prevent livestock from increasing beyond the carrying capacity. The question, therefore, is whether the alpine grazing lawns are as vulnerable as other equilibrium pasture systems. Indeed, the specific traits of the prevailing species described above suggest that degradation threat may be limited.
Estimates of grazing-induced degradation vary: the most frequently quoted value for the Tibetan highlands is that 30% of the grasslands are degraded (Harris 2010, Wang and. The loss in ecosystem services caused by the C emission and N export associated with pasture degradation has been calculated to amount to $8033/ha and $13315/ha respectively (Wen et al. 2013)  Difficulties in properly selecting and interpreting spectral data may also render largescale remote-sensing-based assessments and models questionable (Yang et al. 2005).
Fine-scale changes in vegetation and soils represent appropriate indicators for local, site-based assessments, but they are not easily detected by spectral data, and are not representative for the entire highland region. Thus pasture degradation is clearly a phenomenon with diverging regional gradients, depending on climate, soils and the regionally different impacts of rangeland management change.
Traditional nomadic systems cope with environmental heterogeneity and variability in resource availability by conducting seasonal migratory and other movements. Since Both extreme events cause changes in the volume of the sods. As soon as the root mat has reached a certain thickness with a large portion of dead roots, tensions resulting from the volume changes lead to the formation of polygonal cracks.
Overgrazing and trampling may play an additional role in weakening the stability of the root mat. On the prevailing steep inclination, polygons are progressively separated while drifting downhill with gravity, frequently above a wet and frozen soil layer (Fig.   7D). The widening of the cracks is accompanied by high SOC losses (~5 kg C/m 2 ).
Moreover, SOC loss is aggravated by decreasing root C input following root decay and SOC mineralization indicated by decreasing SOC contents with intensified degradation.
A negative δ 13 C shift of SOC caused by the relative enrichment of 13 C-depleted lignin confirmed this mineralization-derived SOC loss (~2.5 kg C/m 2 ). Overall up, to 70% of the SOC stock (0 cm to 30 cm) was lost in comparison with intact swards of alpine Kobresia pastures in the Kema region (Schleuss et al. [personal communication]).
Here, a degradation survey revealed that about 20% of the surface area has lost its Kobresia turf with bare soil patches remaining (Babel et al. 2014). Assuming that the whole Kobresia ecosystem has suffered from this type of degradation to a similar extent, and that the soil conditions in Kema (SOC stock ca. 10 kg m -2 ) ) are representative across the highlands, this would imply a total SOC loss of 0.6 Pg C for the whole ecosystem with its 450000 km². Consequently a high amount of C is released back to the atmosphere as CO2, or is deposited in depressions and riversleading to a decline of water quality at both landscape and regional levels. The widening space between the crack margins is frequently, but not exclusively, used by pikas to dig their burrows, and they also undermine the 'cliffs' for deposition of their faeces. Excavated soil covers the lawns in front of their burrows (white arrows in Figs. 7C, J), which leads to dieback and decomposition of the felty root mat. Throw-off is also subject to erosion by wind and water. Through their burrowing activity, pikas may increase the ecosystem's net emission of C (Qin et al. 2015), although Peng et al. open soils with pancake-like mats (Fig. 7G, H) is less common than the destruction of the turf, and is restricted to the eastern part of its distribution range with >300 mm/yr precipitation (Miehe and Miehe 2000).
Another common pattern of pasture degradation is found mostly on south-facing slopes, where the lower parts lack any root mat, whereas the upper slope and the ridges are covered with lawns and intact root mat. The mats form a steep cliff towards the slope with sods drifting downhill, probably along with gelifluction processes. The pattern suggests that the lower slopes had been deprived from the lawns by erosion, and the opening of the root mat may have been initiated by yak when chafing and wallowing.
Patches of dead roots covered by crustose lichens or algae (Fig. 8)  clone, but this remains to be confirmed. Comparing the C cycle of closed lawns and crust-covered root mats by 13 C-labeled amino acids revealed that more 13 C remained in soil under crusts, reflecting less complete decomposition of exudates and lower root uptake (Unteregelsbacher et al. 2012). The crust patches decrease the rates of medium-term C turnover in response to the much lower C input. Very high 13 C amounts recovered in plants from non-crust areas, and a two times lower uptake by roots under crusts indicate that very dense roots are efficient competitors with microorganisms for soluble organics. In conclusion, the altered C cycle of the crust-covered root layer is associated with strongly decreased C input and reduced medium-term C turnover. at highest densities of 200-300 animal/ha, these caused a decrease in species richness, vegetation cover, plant height and seasonal biomass. This pattern does, however, not seem to be the rule in the highlands because the most common disturbance indicators are forbs, and it is generally stated that pikas' presence increases habitat diversity and plant species richness (Smith and Foggin 1999, Miehe and Miehe 2000, Smith et al. 2006, and better water infiltration and reduced erosional effects on slopes during torrential summer rains (Wilson and Smith 2015). Herders explain high pika densities as a consequence of overstocking, and not as the cause of pasture degradation (Pech et al. 2007). Pikas have been regarded as "pests" and poisoned; meanwhile the negative and long-lasting negative effects of poisoning on natural predators have been recognized and eradication programs stopped (Pech et al. 2007).

HAVE GRAZING LAWNS FORMED AS A CONSEQUENCE OF PASTORALISM?
The ecosystem's high share of endemics (Wu et al. 1994-2011, Miehe et al. 2011c may indicate naturalness, and indeed Kobresia mats have been described as natural (e.g., Ni 2000, Song et al. 2004, Herzschuh and Birks 2010. With their tiny leaves, a root : shoot ratio >20, very low shoot biomass but a very large root system  and associated large C-storage, the Kobresia pygmaea lawns are one the world's most extensive ecosystems with a very high below-ground share of biomass. Similar root : shoot ratios are known from other cold-adapted ecosystems including arctic tundra (Bazilevich and Tishkov 1997) and high alpine communities (Körner 2003), or from vegetation types exposed to extreme nutrient shortage like the Kwongan of western Australia (Lambers et al. 2010). Climatic parameters (Lehnert et al. 2015(Lehnert et al. , 2016 including soil temperatures , and the nutrient status (see above), can, however, not fully explain the prevailing structures, at least for the montane pastures. The most likely explanation of the allocation patterns found in Kobresia lawns is nutrient shortage in combination with intensive grazing.
The Kobresia-dominated pastures are commonly known as 'alpine meadows', which is misleading in two ways. (1) 'meadow' in a European sense is an agriculturally managed grassland regularly mown for livestock forage (UNESCO classification; Ellenberg andMueller-Dombois 1965-1966). The term 'meadow steppe' is widely used in, for example, Mongolia (Hilbig 1995) and also refers to rangelands, yet of very different structure. In the Tibetan case the designation 'pasture' would be in most areas be correct (even where animal husbandry has no major impact).
(2) whereas 'alpine' is defined strictly as a mountain climate not warm enough to allow for tree growth (Körner 2012), many 'alpine meadows' occur on the same slope together with isolated tree-groves (Miehe et al. 2008c; Fig. 9 Changes from forest to grassland during the Holocene are well-documented (e.g., , Ren 2007, Miehe et al. 2014), yet the explanations are debated.
As the forest decline in various areas of the highlands took place during the mid-Holocene climatic optimum , a climatic driver is not plausible. Given the huge climatic niche covered by current Kobresia pastures, grazing offers a more parsimonious explanation. With a total height of 2-4 cm and hardly any biomass within reach of grazers (Miehe et al. 2008b), grazing-adapted plants like Kobresia pygmaea and associated species will spread at the cost of taller plants. The high root : shoot ratios and the formation of a felty turf can also be viewed in this context.
Grazing selection may also have promoted the expansion of grazing weeds.
Phylogeographic studies on Stellera chamaejasme L. , the prevalent grazing weed of Kobresia pastures (Fig. 10B), revealed a single haplotype over the whole of the highlands. Its presumably rapid expansion should also be explained in the context of livestock expansion.
Indeed, the Tibetan highlands have been grazed over evolutionary timescales by large herbivores (as testified by their richness, including endemic taxa; Schaller 1998).
Former distribution, and natural densities, of these herbivores are largely unknown in Tibet, as they are in any other rangelands of the world. Moreover, limitations in the identification of pollen types (Miehe et al. 2009) render mapping of the historical extent of natural rangeland systems impossible. In any case animal husbandry has progressively replaced wild grazing, and there is anthracological and palynological evidence that herders extended the rangelands well into the montane forest belt (Kaiser et al. 2007, Miehe et al. 2008a, 2008c. Similar patterns have been described for the European Alps (Kral 1979) and the Himalayas . This would explain why forests disappeared in spite of relatively favorable conditions, and also why montane pastures of Kobresia pygmaea change quickly after grazing exclusion.
With increasing grazing pressure, photo assimilates are increasingly allocated belowground. Most probably the grazing lawns, and the felty root layer, can be interpreted as an adaptation to high grazing pressure including trampling. It remains unknown if the absence of a specific endemic fauna of soil beetles can be seen as an effect of an ecosystem under stress. Due to the specific traits of the prevailing species (especially growth habit), this equilibrium system is less threatened by overstocking than is the case in other equilibrium grasslands.

THIS ENVIRONMENT?
The age and intensity of the human impact on Tibetan ecosystems is debated, and estimates based on evidence from various disciplines diverge by more than 20000 years. The earliest migrations and adaptation of Tibetans to high altitude hypoxia was dated to 30 ka BP with a second migration between 10 and 7 ka BP (Qi et al. 2013), to pre-LGM and 15 ka BP (Qin et al. 2010), or to 25 ka BP ), yet it remains uncertain where this mutation occurred (Madsen 2016 (Brantingham and Gao 2006, Bellezza 2008, but the dating of scattered surface remains difficult. While archeology-based estimates of the time for the first intrusion of hunters range between 30 and 8 ka BP (Aldenderfer 2006, Brantingham et al. 2007), 14 C-and OSL-dated remains suggest a more recent date suggesting that hunting parties first travelled in the region between 16 and 8 ka BP. The relevance of handand footprints in tufasediments north of Lhasa (Chusang) remains obscure as data range from 26 ka (Zhang and Li 2002) to 7.4 ka BP . Obsidian tools dated between 9.9 and 6.4 ka BP have been transported over more than 900 km ). In any case, humans have most probably travelled within the region during the Last Glacial Maximum (LGM). Long-term residential groups of hunters, or perhaps early pastoralists, may have settled in the area between 8 and 5 ka BP ), a date supported by independent evidence from the genomic signature of yak domestication .
So far direct proof of early human impact on vegetation structures has not emerged.
The presence of humans is commonly associated with the use of fire, wherever fuel loads in the dry season are high enough for lighting (Bond and Keeley 2005). Fire traces observed in Tibetan sediments may relate to human action (see Fig. 11). As highland plants lack obvious adaptations to fire (e.g., no pyrophytes as present in the Boreal Forest, the South African Fynbos, or the Australian Kwongan) it seems unlikely that fire had been present over evolutionary time-scales prior to human arrival.
Lightning occurs nearly exclusively during the rainy season, followed by torrential rains.
The seasonality of the highland climate would favors the use of fire as a tool to modify vegetation structures. This is especially true for the mid-Holocene climatic optimum between 10 and 5 ka BP, when summer growing conditions were wetter and warmer than present , resulting in presumably high fuel load available for burning during the dry cold winter. Fires lit by hunters were most probably the first impact, as they are associated with the first intrusion of humans in other parts of the world (e.g. Ogden et al. 1998). Landscape-scale management by burning has probably intensified with the introduction of livestock grazing: forests were burnt to provide better pastures, and to destroy places of concealment for predators. Charcoal of Picea (P.
crassifolia Komarov) and Juniperus (J. przewalskii Komarov) has been found in the north-eastern highland pastures where current precipitation is double the minimum amount of rainfall necessary for tree growth. Moreover, these sites are situated at altitudes 200-400 m below the upper tree line (Miehe et al. 2008c). Dating of the charcoal implies ages between 10.0 and 7.4 ka BP (Kaiser et al. 2007 , Meng et al. 2007, that have been opened up and fragmented more recently by pasturing. Charcoal and pollen records point to a forest decline (Picea, Betula) after between 8 and 6 ka BP (Shen et al. 2005, Cheng et al. 2010, Miehe et al. 2014, at a period that, according to the human impact-independent proxy of Ostracod assemblages (Mischke et al. 2005), was the most favorable climatic period of the Chenopodiaceae, similar to a pollen-sequence typical for the 'landnam' in Europe (Frenzel 1994 (Miller and Schaller 1996), the increase of grazing weeds could well be a result of increased numbers of wild herbivores.
Phylogeographical analysis of Panthelops hodgsonii Abel (chiru, Tibet antelope), however, shows a clear decrease in herd-size during this period (Du et al. 2010), probably synchronous with the domestication of the yak (7.3 ka BP; .
The archeo-zoological record of early animal husbandry and agriculture is still very limited; earliest fossil records of domestic yak reveal ages of only 3.8 ka BP (Flad et al. 2007), which are similar to the oldest records of cereals (barley in the northeastern highlands: 4 ka BP, in the Yarlung Zhangbo valley: 3.4 ka BP; Setaria italica (L.) P.Beauv. on the eastern slope of the highlands: 4.6 ka BP; Chen et al. 2015, d'Alpoim Guedes 2015. The permanent human occupation of the highlands is thought to have depended on mutual exchange between cereal farmers (trading barley as the key staple food of Tibetans) and pastoralists (giving animal products and salt), and has thus been dated as younger than 3.6 ka BP . Recent genomic evidence, however, implies a much older date for the domestication of the yak (7.3 ka BP; . One of the most important prehistoric excavations in China, near Xian on the Loess Plateau, is possibly also relevant for the question of the onset of pastoralism in the Tibetan highlands: In settlements of the Yangshao Culture (7-5 ka BP; Parzinger 2014), spindle whorls, found in large quantities, most probably testify to the weaving of sheep wool. Whether similarly aged bones of Caprini found at those sites are of domestic origin or not, is unclear (Flad et al. 2007). Given that sheep must have been introduced from their center of domestication in the mountains of the Middle East (Zeder and Hesse 2000), and that Tibet is 1600 km closer to this center of origin than is Xian, pastoralism may have started earlier in the former than the latter. If we hypothetically apply the same rate of diffusion of the Neolithic Package from the center of domestication towards Europe (3 km/yr from the Zagros Mountains in western Iran to the south-east European Vojvodina; Roberts 1998), domestic sheep and goats may have reached the Tibetan highlands around 8.6 ka BP.
There thus is evidence from various sources indicating that human land use has established shortly after the end of the glacial period. The question of when increased livestock numbers first had an impact on plant cover, and on the below-ground C allocation that forms the felty root mats, is still, however, unanswered.

CONCLUSIONS
(1) The Kobresia pastures of the Tibetan highlands are a nutrient-and water-limited high-altitude ecosystem. They form equilibrium rangeland systems, yet their vulnerability to grazing degradation is limited due to the prevalence of a dwarf sedge with its main above-ground phytomass below the grazing reach of livestock.
(2) Kobresia mats are characterized by a felty root layer that represents very large carbon (C) store. Natural degradation phenomena with polygonal crack patterns and the drifting apart of polygonal sods are, however, widespread. The widening of the cracks is accompanied by high SOC losses (~5 kg C/m 2 ). Overall up to 70% of the SOC stock was lost in comparison with intact swards of alpine Kobresia pastures. Assuming that the whole Kobresia ecosystem has suffered to a similar extent, this would imply a total SOC loss of 0.6 Pg C for the whole southeastern highlands. Consequently high amounts of C are released back to the atmosphere as CO2, or are deposited in depressions and rivers.
(3) Pastoralism may have promoted dominance of Kobresia pygmaea and is a major driver for below-ground C allocation and C sequestration, stabilizing these root mats with their distinctive C allocation patterns. Grazing exclosure experiments show that the larger below-ground C allocation of plants, the larger amount of recently assimilated C remaining in the soil, and the lower soil organic-matter derived CO2 efflux create a positive effect of moderate grazing on soil C input and C sequestration in the whole ecosystem.
(4) Due to the highlands' relevance for atmospheric circulation patterns, surface properties of these pastures have an impact on large and possibly global spatial scales. The removal of the lawns, caused by climatic stress as well as excessive human impact leads to a shift from transpiration to evaporation in the water budget, followed by an earlier onset of precipitation and decreasing incoming solar radiation, resulting in changes in surface temperature, which feedback on changes in atmospheric circulations on a local to regional scale.
(5) The age of the world's largest alpine ecosystem, and its set of endemic plants and animals, remains a matter of considerable dispute, though the degree of this uncertainty is rarely admitted (reviews in Liu et al. 2014, Favre et al. 2015, Schmidt et al. 2015, Renner 2016. Further calibrations based on genomic data and fossils (where available) are needed to clarify evolutionary relationships and divergences between the currently recognized species of Kobresia.
(6) The paleo-environmental evidence, as well as simulations, suggests that the present grazing lawns of Kobresia pygmaea are a synanthropic ecosystem that developed through selective free-range grazing of livestock. The age of the present grazing lawns, however, is not yet known. The presence of humans using fire and replacing forests by grassland may date back as far as the LGM, while archeological evidence for such an early onset of pastoralism is missing.
A multi-proxy approach, however, suggests a mid-Holocene climatic optimum age.
(7) The traditional migratory, and obviously sustainable, rangeland management system conserved and increased the C stocks in the turf and its functioning in the regional and global C cycles. However, rangeland management decisions within the past 50 years have caused widespread overgrazing leading to erosion and reducing the C sink strength. Considering the large area of the grasslands, even small reductions in C sequestration rate would affect the regional C balance, with possible impacts on the future climate of China and beyond.