CSIRO PUBLISHING The Rangeland Journal, 2016, 38,1–15 http://dx.doi.org/10.1071/RJ15040

Rangeland responses to pastoralists’ grazing management on a Tibetan steppe grassland, Province,

Richard B. Harris A,G,I, Leah H. Samberg A,H, Emily T. Yeh B, Andrew T. Smith C, Wang Wenying D, Wang Junbang E, Gaerrang B,F and the late Donald J. Bedunah A

ADepartment of Ecosystem and Conservation Sciences, University of Montana, Missoula, MT 59812, USA. BDepartment of Geography, University of Colorado, Boulder, CO 80309, USA. CDepartment of Life Sciences, Arizona State University, Tempe, AZ 85287, USA. DDepartment of Biology, Qinghai Normal University, , Qinghai, China. EKey Laboratory of Ecosystem Network Observation and Modeling, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing, China. FThe Centre for Tibetan Studies of University, Chengdu, Sichuan, China. GPresent address: Washington Department of Fish and Wildlife, Olympia, WA 98501, USA. HPresent address: College of Forestry and Conservation, University of Montana, Missoula, MT 59812, USA. ICorresponding author. Email: [email protected]

Abstract. Livestock grazing is the principal land use in arid central Asia, and range degradation is considered a serious problem within much of the high-elevation region of western China termed the Qinghai-Tibetan Plateau (QTP). Rangeland degradation on the QTP is variously attributed to poor livestock management, historical-cultural factors, changing land tenure arrangements or socioeconomic systems, climate change, and damage from small mammals. Few studies have examined currently managed pastures using detailed data capable of isolating fine-scale livestock–vegetation interactions. The aim of the study was to understand how differences among livestock (primarily sheep) management strategies of pastoralists during winter affected subsequent rangeland condition and productivity. Plant species composition, annual herbage mass, and indicators of erosion were quantified during four summers (2009–2012) on winter pastures managed by 11 different pastoralists on QTP steppe rangeland in Qinghai Province, China. Data came from repeated-measurements on 317 systematically located permanent plots, as well as pastoralist interviews and the use of GPS-equipped livestock. Relationships between annual weather variation and herbage mass were modelled using an independent set of vegetation measurements obtained from livestock exclosures. Account was taken of inherent site differences among pastures. Annual variation in herbage mass was found to be best fitted by a model containing a negative function of winter-season temperature and a positive function of spring-season temperature. Accounting for annual and site effects, significant differences among pastoralists were found for most response variables, suggesting that individual heterogeneity among management approaches had consequences, even among neighbouring pastoralists. Annual herbage mass of preferred plant species was positively associated, whereas that of unpreferred species was negatively associated, with mean sheep density and intensity of use. However, the proportion of bare soil, an index of erosion, and annual herbage mass of unpreferred forbs were found to have positive relationships with sheep grazing pressure during the preceding winter, whereas live vegetation cover and annual herbage mass of preferred grasses were negatively related. Thus, on a spatial scale, pastoralists responded adaptively to the cover of preferred plant species while not responding to total annual herbage mass. Pastoralists stocked pastures more heavily, and livestock used regions within pastures more intensively, where preferred species had a higher cover. However, where sheep grazing pressure was high, downward temporal trends in the herbage mass of preferred species were exacerbated. Pastures that were stocked at a lower density did not experience the negative trends seen in those with a higher density.

Additional keywords: China, grasslands, livestock, Qinghai-Tibetan plateau, rangeland degradation, steppe vegetation, Tibetan pastoralism.

Received 18 May 2015, accepted 1 December 2015, published online 22 January 2016

Journal compilation Australian Rangeland Society 2016 www.publish.csiro.au/journals/trj 2 The Rangeland Journal R. B. Harris et al.

Introduction regardless of its precise definition, but this may have reflected In rangelands throughout the world, livestock provide food and the general Maoist conception of nature at the time (Shapiro income to the majority of the world’s poorest people (Bedunah 2001); scientific work on pikas began only in the 1980s. Research and Angerer 2012), yet degradation of rangelands is having subsequently began taking a more nuanced view of interactions direct effects on the livelihoods, food security, and way of life between pikas, livestock and vegetation, after it emerged not of pastoralists. In addition, this degradation can significantly only that pikas were critical parts of the natural environment reduce ecosystem services provided by rangelands at regional (‘keystone species’, Smith and Foggin 1999; Lai and Smith and global scales, such as carbon sequestration and hydrological 2003; Wilson and Smith 2015), but that high density of pikas regulation and provision. Effects of degradation, such as dust likely resulted from, rather than caused, the rangeland conditions storms, food or commodity shortages, and displaced communities, with which they were associated (Shi 1983; Bian et al. 1999; can be felt over a large geographic scale (Yan 2001; Wang et al. Wangdwei et al. 2013). Regardless, programs of pika reduction 2005;Liet al. 2013). Rangeland health also affects biodiversity continue to the present day (Smith et al. 2006; Delibes-Mateos directly and indirectly because all native flora and fauna have et al. 2011; Wilson and Smith 2015). adapted to the long-term evolutionary forces that have shaped Privatisation of livestock and dissolution of the collective these environments. Livestock grazing is the dominant form of system occurred during the 1980s throughout the QTP, and in land use in arid biomes worldwide, including on the rangelands our study area in 1983. During the 1990s, following the success of the Qinghai-Tibetan Plateau (QTP) in the People’s Republic of quasi-privatisation policies in eastern China’s agricultural of China. The QTP occupies 2.5 million km2, ~25% of China’s sector, similar approaches, often referred to as the Household area, and an estimated 70% is used by grazing livestock. These Responsibility System, were initiated in the pastoral sector (Yan pastures, colloquially termed ‘China’s water tower’, are located et al. 2005). Echoing widespread concerns about the ‘tragedy of upstream and upwind of upwards of an estimated 20% of the the commons’ (Hardin 1968), these programs aimed to encourage world’s human population (Xu et al. 2009; Immerzeel et al. responsible husbandry by clarifying pasture-land tenure at the 2010). household level. The primary tenets of the government initiative Awareness by Chinese scientists and policy-makers of the involved increasing the duration of pasture-lease contracts from impacts of degradation increased in the late 1990s as several 20 to 50 years; subsidising construction of permanent winter disasters occurred, including Yangtze River floods that killed homes, fences and livestock shelters; and providing plots for thousands of residents downstream and cost billions of dollars growing supplemental winter fodder (Richard et al. 2006;Wu in economic losses, the running dry increasingly et al. 2012). Government outlays for programs related to the often, and dust-storms and sand-storms originating in western Household Responsibility System (often termed the ‘set of four’, rangelands that affected the health and economic wellbeing of or sipeitao in Chinese (Wu and Yan 2002)) were substantial; millions of city-dwellers in China’s east. Although lacking clear during 2003–2006, the central government reported investing documentation and differing in specifics, scientific papers and some ¥7.1 billion (~US$1 billion at the time) for fencing alone government policy statements have generally viewed the QTP (SEPA 2007). Whether this fundamental reform alleviated or as having become increasingly degraded in recent decades exacerbated negative trends in rangeland condition remains (Harris 2010). A frequently repeated statistic is that 90% of contentious. China’s grasslands are degraded to some extent, and that A newer set of initiatives began in the early 2000s that degradation is increasing at a rate of 200 km2 year–1 (State emphasised land protection rather than tenure and responsibility Council 2002). (Yeh 2009). Increased awareness of the consequences of Causes for this degradation are generally attributed to a upstream erosion following the devastating Yangtze River flood combination of over-stocking, unscientific livestock management, of 1998 led to government subsidies encouraging reforestation historical-cultural impediments to adopting modern livestock of cultivated lands that were unsuitable for agriculture (tuigeng management concepts, global climate change, and excessive huanlin, ‘retire cultivation, restore forests’; Grant 2003; McBeath herbivory and soil disturbance from small mammals (Li 1994; and McBeath 2010). This approach was later expanded and Chen 1996; Zhou et al. 2003; Zhang et al. 2004;Liet al. 2013). adapted to encompass non-forested lands that had been In contrast, other investigators have questioned the assumption inappropriately transformed from rangeland to agriculture of wide-scale grassland degradation on the QTP and, where they (tuigeng huancao, ‘retire cultivation, restore grasslands’), in agree it has occurred, cast their analysis in terms of rapid which artificial seeding of forage plants took the role of the changes in socioeconomic systems and alteration of land-tenure reforestation projects encouraged in more mesic climates (Shen arrangements (Miller et al. 1992; Levine 1998; Holzner and et al. 2004). A closely related program required a change of only Kreichbaum 2001; Goldstein and Beall 2002; Williams 2002;Wu one Chinese character to introduce a new approach that and Yan 2002; Banks 2003; Banks et al. 2003). represented a substantial change in policy. This program (tuimu Chinese policy-makers have implemented a variety of huancao, ‘retire livestock, restore grasslands’) was nominally programs to ameliorate negative rangeland trends (Harris 2010; intended to conserve rangeland resources and increase long- Li et al. 2013; Shang et al. 2014), some more effectively than term livestock production (Yeh 2005; Foggin 2008). However, others. As early as the 1950s, efforts were begun to exterminate a central component of the ‘retire livestock’ programs was plateau pikas (Ochotona curzoniae), a small-bodied, burrowing to eliminate grazing entirely for specified durations. lagomorph, generally associated with poor vegetation cover Thus, in contrast with programs that attempted to encourage (Smith et al. 1990; Fan et al. 1999). Policy during this period responsible husbandry through a tighter linking of pastoralists blamed small mammals for causing rangeland degradation, with specific tracts of land, these new programs encouraged Rangeland responses on Tibetan steppe The Rangeland Journal 3 cessation of pastoralism. In some areas, particularly in the semi-nomadic pastoralism. Distance to the nearest concentration area of Qinghai, this approach was coupled with, of houses to our study area was ~6 km; this village was adjacent and often conflated with shengtai yimin, ‘ecological migration’ to an historic but rejuvenated Tibetan Buddhist monastery (Foggin 2008;Du2012), which encouraged pastoralists to sell (Harris et al. 2010; Yeh and Gaerrang 2011). The landscape, their livestock and resettle entirely in government-constructed part of the eastern section of the Kunlun mountain chain, was housing located in often-distant towns. This suggests a belief characterised by rolling hills at elevations <4100 m, rising to that physical relocation of pastoralists and reorientation of their moderately sloped peaks at ~4900 m. Vegetation was sparse means of livelihood were necessary components of ecological above ~4700 m. Vegetation formations were alpine steppe, restoration. dominated by Stipa purpurea Grisebach, at elevations <4300 m, There has been considerable variation as well as mutability alpine meadow, dominated by Kobresia spp. at higher in local implementation of central level policy. In many cases, elevations, and shrublands, dominated by Salix spp. on northerly township and village leaders have focussed more on the specific exposures. Annual precipitation at the study site during costs and benefits (e.g. fencing and subsidies) rather than the 2008–2013 averaged 398.0 mm (s.d. of mean, 53.4), with ~92% underlying ecological rationale (Bauer 2005; Bauer and Yonten falling from April to September. Mean annual temperature was Nyima 2010; Yonten Nyima and Yeh in press). The ‘retire approximately –1.48C, with the warmest 8-day periods annually livestock’ program does not appear to have entirely supplanted averaging 14.08C and the coldest averaging –16.38C. the earlier responsibility-system-based programs in areas The study area had been subject to international hunting outside of the Sanjiangyuan; indeed in some places such as the focussed on blue sheep (Pseudois nayar) until 2006 (Liu 1995; Tibetan Autonomous Region, ‘retire livestock’ has been seen Harris 2008), but this activity probably had little impact on as a way to further implement the Household Responsibility pastoral practices or vegetation. As had been common on the System (Bauer and Yonten Nyima 2010). In and around QTP, during January 2007 government-sponsored workers Sanjiangyuan, the two have co-existed, sometimes in close conducted a poisoning campaign targeting plateau pikas. geographic proximity. Subsequent work showed that within a few years, pikas had It is unlikely that pastoralism will completely disappear on the repopulated most areas. QTP; other forms of agricultural land use are incompatible with The Tibetan fauna includes several wild ungulate species environmental conditions found at an elevation of 3000–5000 m. that could have foraged on vegetation (Schaller 1998; Harris Thus, in this paper we focus on how individual pastoralists 2008) but only Tibetan gazelles (Procapra picticaudata) and might encourage or discourage sustainability of their rangelands blue sheep were observed in the vicinity of vegetation plots (the and herds via their own decisions, regardless of the overarching latter only at the highest elevations). In addition to the common policy environment. We were motivated by observations made plateau pika, small mammals present in the general area included during earlier fieldwork related to wildlife research in the study Himalayan marmots (Marmota himalayana), the fossorial area (Liu et al. 2007, 2010; Harris 2008) that rangeland conditions plateau zokor (Eospalax fontanierii), the Mongolian five-toed appeared to vary by pasture, even within a single village and jerboa (Allactaga sibirica), mountain voles (Neodon spp.), voles even among those with superficially similar topographic (Microtus spp. and Lasiopodomys spp.), and dwarf hamsters attributes. We wondered if we could associate differences in (Cricetulus spp.). Tibetan woolly hares (Lepus oiostolus) approaches to livestock husbandry with this variation, and occurred at slightly lower elevations, and were rarely observed ultimately, if socioeconomic factors at the pastoralist level, or in the study pastures. the ways in which pastoralists responded to the broader policy The entire study area was grazed by livestock and used environment, could explain the choices made by pastoralists. In primarily as winter pastures. Grazing generally occurred only this paper we explore only the livestock-husbandry/rangeland- after livestock returned from summer and, sometimes, autumn response dynamics. pastures, generally in mid-October, until leaving for spring– Specifically, our objectives were: (1) accounting for site summer pastures in mid-June the following year (Yeh and and annual climatic variation, test whether differences among Gaerrang 2011). Village Five consisted of relatively high species herbage mass, and erosion indicators were explained by elevation pastures within Gouli, and had been used as summer the pastoralist managing the pasture; (2) accounting for site and and transitional (spring–autumn) pastures before the prior annual variation, test whether variation in herbage mass and collective system was dismantled. Pastoral families owned long- erosion indicators was associated with variation in sheep density term leases on set pasture lands, but not all grazed their own and/or spatially explicit measures of grazing pressure; and (3) livestock on their own pastures. Rather, many pastoralists in accounting for site and annual variation, test whether changes Gouli had begun sub-leasing their pastures to other grazers, and/ in herbage mass and erosion indicators were predicted by annual or paying rental fees to graze their livestock on lands contracted changes in sheep density. to others (Yeh and Gaerrang 2011). Some winter pastures were demarcated with boundary fences and others were not; regardless, most boundaries were known to pastoralists and Materials and methods had traditions predating de-collectivisation. Summer pastures Study area remained, as historically, in common use. However, some Our study was carried out in Village Five (‘wu dui’) of Gouli pastoralists had taken advantage of subsidised material to fence Township, , Qinghai Province, China, ~35.58N, sub-sections of winter pasture in order to distribute grazing 98.78E. Village Five consisted of ~175 residents in 37 pressure, reserve forage for emergencies, or reduce the need for households, almost all of whom were engaged primarily in herding labour. 4 The Rangeland Journal R. B. Harris et al.

Weather data dates on which each plot was sampled. Thus, we incorporated We measured temperature and precipitation at hourly intervals both annual weather effects and a quadratic model of Julian date at the research station (35834045.930N, 98836034.710E) using a in all models (see Supplementary Materials as available at ’ solar-powered logger (CR800-ST-SW-NC; Campbell Scientific, journal s website). Additionally, we hypothesised that herbage fi Logan, UT, USA) connected to a temperature probe (107-L20) mass on each pasture would vary according to xed site and a tipping-bucket precipitation gauge with snowfall adaptor conditions characterising each pasture. The physical effects (Texas Electronics, Dallas, TX, USA) beginning in September of elevation, slope, and aspect were beyond the control of 2009. However, damage to the precipitation gauge rendered it pastoralists, yet likely to affect rangeland response. Thus, we inoperable after July 2010; thus, we modelled temperature and also incorporated these variables in our analyses (see precipitation on the study area (see below) and used our limited Supplementary Materials). site-specific weather data to confirm the accuracy of the modelled predictions. Vegetation data and erosion indicators To model site-specific temperature and precipitation, we Grassland vegetation and soil conditions were assessed over interpolated meteorological data from 836 weather stations in four summers (2009–2012) using a grid of permanent plots on China using ANUSPLIN software version 4.2 (Hutchinson winter pastures. In summer 2009, we established 317 permanent, 1995). This algorithm was based on the thin plate smoothing square 0.5-m2 vegetation plots (measuring 0.71 m on each side), splines of multivariates (Hutchinson 2001, Hutchinson et al. systematically located on a 250-m grid (oriented along cardinal 2009). Observations from the weather stations were interpolated directions, Fig. 2) across 15 winter pastures managed by 11 to a 1-km spatial resolution at an 8-day time step. Correlation participating pastoralists. Pasture sizes varied from 46 to 1009 ha of 8-day mean temperature of the modelled data with 71 8-day (mean = 320 ha; s.d. of mean = 309) ha. We used permanent plots measurements taken at the field station from 14 September 2009 to allow for direct year-to-year comparisons and to reduce to 11 June 2011 was 0.939 (Fig. 1). After accounting for seasonal small-scale heterogeneity. In identifying the exact area to patterns by generating residuals from regressions using Julian establish each plot, we walked to each pre-determined Universal date as a predictive variable, correlation between modelled and Transverse Mercator (UTM) coordinate on the sample grid (i.e. site-specific temperature measurements was 0.560 (P < 0.001). 250-m intervals) using a handheld GPS, establishing the plot Fluctuations in herbage mass resulting from pastoralists’ centre exactly on the intended coordinates. In addition to strategies, our main variable of interest, were likely to be elevation, we measured aspect (in degrees) and slope (in per cent) confounded with annual weather differences as well as of each plot using a hand-held compass. To document aspect, differences in phenological stage arising from the different we quantified each plot both by its absolute deviation from true

20

15

10

5

0

–5 Mean 8-day temp. ( ° C) temp. Mean 8-day –10

–15

–20

–25 0 50 100 150 200 250 300 350 Julian date

Fig.1. Measured(triangles,dashedtrendline)andmodelled (circles,solidtrendline)mean dailytemperatureat8-day periodsat Gouli field station, (35.58N, 98.78E, 4033 m); see text for model procedures. Rangeland responses on Tibetan steppe The Rangeland Journal 5

N

Fig. 2. Detail of the study area, showing the systematic grid of vegetation plots (triangles) at 250-m intervals overlaid on shaded topography. Also shown are the pasture boundaries (solid polygons) and the location of the main river draining in the area (dashed arrow). north and from true east in degrees (i.e. an aspect of 108 was locating tagged, wire loops connected to fixed, flexible anchors given a score of 10 on the north scale and 80 on the east scale; (Berkshire HD Stakes with Cables, Buckeye Trap Supply, an aspect of 3008 was given a score of 60 on the north scale and Ashland, OR, USA) that crews had inserted into the soil ~30 cm 150 on the east scale). at each plots’ diagonal corners when establishing the plots in Plots were designed to be relocated and identified by field 2009. Each anchor left a small, protruding (~3 cm diameter) loop crews in two ways: (1) via their UTM coordinates; and (2), by of ~5-mm diameter steel cable to which we affixed a numbered 6 The Rangeland Journal R. B. Harris et al. metal tag (5 cm diameter, Forestry Suppliers, Jackson, MS, rills, gulleys, and pedestalling (NRC 1994). Erosion categories USA), but otherwise left the surrounding area unaffected. When were re-coded on an ordinal scale, with 1 representing the sampling vegetation, field crews used light-weight PVC 0.5-m2 least and 6 the most evidence of erosive forces. For consistency, frames; upon locating the cables, they fixed the frame corners we have used the terminology recommended by Allen et al. to the two loops to re-create the precise plot location. (2011). Vegetation and site data were collected by crews of trilingual (Tibetan, Chinese and English) seasonal technicians. Prior to each field season, crews were trained in species identification Stocking rate and density and field protocols. Field crews assessed presence of plant Because our intent was to examine the effects of grazing species, height, cover, and herbage mass for each of the 10 main practices as actually implemented by Tibetan pastoralists, species in each plot, as well as canopy cover estimates of total responding as they wished to the geophysical, biological, vegetation, soil, rock, and litter (previous years’ dry vegetation). cultural, economic and policy environments in which they Most plots were quantified once during the growing season, found themselves, we made no attempt to control grazing. We but some were measured twice or more, depending on access. used direct counts to estimate the number of sheep and yaks on Because we used permanent plots, we could not quantify each pasture in each of the three winters included in the study herbage mass by clipping vegetation. Thus, we estimated herbage (2010, 2011, and 2012; throughout, we refer to winter grazing mass individually by plant species in a stepwise process. First, by the year beginning in January; thus, for example, we use the in the field, crew members standardised their estimations of term ‘winter 2010’ to refer to grazing that occurred during species-specific fresh weight by collecting reference samples ~October 2009 through early June 2010). In addition, we drew of known fresh weight (usually 1 g) from an adjacent, off-site from interviews and surveys carried out with the 11 pastoralists location, and moving them among locations within the plot whose pastures were included in the study area. In 2009 and where that species was growing, providing for close, visual 2010, in-depth interviews were carried out with each pasture comparison with unclipped vegetation. Second, to provide on- owner by a member of our field crew as part of a larger going calibration of fresh weight estimation, a system of random investigation of land tenure and land management (Yeh and check-plots was set up, in which crews learned whether plots Gaerrang 2011). Interview data included numbers of livestock were selected for calibration only after vegetation data had had of each species owned currently, approximate time periods been collected. If selected, a nearby 0.5-m2 location with similar within each winter during which livestock grazed on that vegetation to the plot was identified, subjected to the full particular pasture, and trends in livestock ownership over measurement protocol, and then clipped and sorted to species. past years. Pastoralists also described the extent to which they Species-specific fresh weights of the check-plot were recorded, rented or contracted pastures or livestock, their dependence and compared with the actual (non-clipped) plot. Finally, to on herding for income, and perceived changes in climate and convert fresh to dry weights, clipped samples from check-plots grassland condition. Estimated livestock numbers derived from were placed in paper bags, and either air-dried at the base-camp these two approaches were then divided by the area of the or within a light-weight solar oven (Sport Solar Oven, Solar pasture (estimated by walking the periphery of each pastures Oven Society, Apple Valley, MN, USA), until weights stabilised with a hand-held GPS), as well as percentage of the year (5–8 days). Dry-weight conversions were estimated by species livestock spent on the pasture, to estimate mean stocking rate and by month. Because we were unable to dry a sufficient for each pasture in each year. number of samples for species categorised as unpreferred During the three winters included in our study, we attached forbs, we analysed fresh herbage mass rather than dry herbage GPS units (DG-100 Data Logger, US GlobalSat Inc., Chino, CA, mass for forbs and total vegetation. USA: Qstarz BT-Q1000XT, Taipei, Taiwan; or iBT-GPS Solar Throughout, we took, as our response variables, estimates of Bluetooth 747, GandV Global Tech Co., Taipei, Taiwan) by herbage mass (fresh or dry, see above) of the following 10 species means of nylon harnesses to 1–2 individual sheep in each (or groups, listed here in alphabetical order, not order of pastoralist’s herd for periods of 2–3 days for three periods each importance): (1) Astragalus/Oxytropis spp. (forms sometimes winter. Data downloaded from the GPS units provided specific difficult to distinguish from one another, and likely to respond locations of herds at 2–6-min intervals. These data on location similarly to biotic and abiotic influences); (2) Cardamine spp.; were used to demarcate levels of grazing density within each (3) Carex spp.; (4) Heteropappus altaicus Novoprokr; (5) study pasture in each winter. All points falling within a Kobresia spp.; (6) Leymus secalinus (Georgi) Tzelev; (7) Poa pasture in a given winter, both from the owner’s flock and spp.; (8) Potentilla bifurca Linnaeus; (9) Stipa purpurea; and from neighbouring flocks, were entered into a pasture-specific 10) Thermopsis lanceolata R. Brown. In addition, we examined analysis. We used ArcGIS 9.0 (ESRI, Redlands, CA, USA) aggregated life-forms as (1) grasses, (which included minor to enumerate the number of sheep locations within 250 m of species); (2) sedges; and (3) unpreferred forbs (Liu 1986; each vegetation plot. To standardise metrics of relative use Damiran 2005). We also examined dry herbage mass of all across pastures that differed in their intensity of sampling, we grasses, and total herbage of all species. As proxies for recast stocking density in terms of pasture-specific proportional magnitude of the presence of or potential for erosion, we use. In addition, GPS collars (Log V2 livestock collars, Kedziora examined (1) litter canopy cover; (2) proportion of bare soil (i.e. Innovation Group, Mannsville, NY, USA), were attached to ground unvegetated and not covered by rocks, potentially 14 yaks belonging to four different pastoralists from autumn vegetated but bare); and (3) an index of erosion that was 2011 to spring 2012, for periods of 1 week to 1 month. These recorded categorically and reflected the relative presence of collars provided an indication of the ways in which winter Rangeland responses on Tibetan steppe The Rangeland Journal 7 pastures were grazed by yaks but we do not report or use these during the study. Thus, before conducting analyses we removed data further because we discovered that yaks primarily used from consideration all plots in which a species was absent pastures outside the study area (i.e. in summer pasture even during all sampling occasions. All statistical analyses were during winter), returning only occasionally to the study area. conducted using the software package JMP 11.1.1 (SAS Institute, Cary, NC, USA).

Statistical analyses Results Prior to statistical analyses, we removed outlier data that we Characteristics of vegetation plots suspected as representing data-recording errors. We defined outliers as entries >4 s.d. from the mean. This generally resulted Vegetation and erosion indicators from 317 plots were measured in removing ~1% of data entries. at least once yearly; pastures contained between 8 and 55 plots. To address objective 1, we asked if time-invariant measures Analysis was performed on a total of 1771 measurements of species-specific herbage mass and erosion indicators (across (Table 1). Plots varied in elevation from 3863 to 4511 m (mean all 4 years) varied by the pastoralist with user rights to pasture (s.d. of mean) = 4142 (140) m) and were located on slopes 8 8 land during the time period. We used ANOVA to test for ranging from 0 to 70 (mean (s.d. of mean) = 17.7 (13.0) ). The 8 differences and Tukey’s HSD to differentiate levels of aspects of the plots ranged from 0 to 359 ; plots on aspects 8 difference by pastoralists. To address objective 2, we asked if within 60 of true south were most common (31%), followed time-invariant measures of species-specific herbage mass and by those most nearly oriented towards the west (28%), north erosion indicators (across all 4 years) were associated with the (23%) and east (18%). Plots contained between 0 and 18 genera mean stocking rate of sheep (measured at the scale of the entire of grasses, forbs, and shrubs (mean (s.d. of mean), 6.5 (2.7) per pasture), site-specific relative stocking density (as estimated plot). Total vegetation cover in plots ranged from 0% to using the GPS-marked sheep), or by the product of these two 98.5% (mean (s.d. of mean) 33.9 (20.3)%). fi measures on the pasture during the sampling period. We obtained suf cient dry samples of the herbage mass We developed mixed linear models examining each of S. purpurea from off-site plots to quantify the relationship between fresh- and dry weights (F = 9.7, d.f. = 3, 64, P < 0.001; hypothesis. Such analyses incurred a risk of incorrect 2 interpretation because they lacked a before-versus-after time r = 0.31). We thus replaced each measurement of fresh weight element, causation could logically have run in either direction. of herbage of S. purpurea by their dry weight. In 2009, 2010, We addressed this difficulty by means of objective 3, taking 2011 and 2012 the ratios of dry weight to fresh weight were advantage of our repeated-measures on the same plots to 0.61, 0.49, 0.67 and 0.54. Because ~65% of grass herbage was examine temporal trends, and to relate those to (1) time; (2) S. purpurea, we applied these annual conversions to analyses differences in the stocking rate by sheep at the pasture scale; of aggregated grass species. and (3) sheep density within a pasture (i.e. at the plot scale) in the winter before that year’s growth (as well as interactions Effects of individual pastoralists on cover and herbage among these variables). Here, we used time to separate causality, mass of species, and erosion indicators reasoning that events occurring later in time could be effects After accounting for the effects of site heterogeneity (tables or could be unrelated, but could not be causes of events that S2–S4), we observed significant differences among individual occurred earlier. In these dynamic analyses, we examined pastoralists that evidently reflected in aggregate their current differences in measurements taken in 2010, 2011, and 2012 from and past management practices (Table 2). Leymus secalinus those taken in 2009 (the baseline year; all 2009 values were fixed to zero). Standardising all measurements by their values in Table 1. Pastoralists with long-term leases on each (coded for 2009 placed data from all plots on a common basis, effectively anonymity) and pasture codes (numbered) in the study area, Village removing any site effects. We thus omitted elevation, slope Five, Gouli township, Dulan County, Qinghai, 2009–2012 and aspect as independent variables in these dynamic models. Shown are pasture sizes (in ha) and number of plot readings (including By not including site variables in regressions, we implicitly replicate within-year readings, n) in each year assumed that subsequent changes (i.e. slope coefficients) with Pastoralist Pasture Size 2009 2010 2011 2012 Total time, sheep stocking rate and density were not themselves codes (ha) nnnnn functions of site variables. However, all models included the B 3 46 10 8 12 25 55 effects of temperature (Model 2, Supplementary Materials table A ’ D1 10 n.d. 10 10 20 17 57 S1 as available at journal s website) because we considered D2 7 467 49 68 80 82 279 these primary; only trends with time or sheep stocking rate or G 9 110 20 19 36 31 106 density that were significant, while accounting for possible H 12, 15, 16 253 39 40 64 51 194 effects of annual variation in temperature, were considered K 1 678 56 81 85 88 310 valid. We also included Julian date and (Julian date)2 in all L1 5 616 27 42 39 29 137 models, to account for differences in phenological stage arising L2 6 350 19 22 27 26 94 from the different dates on which each plot was sampled. In all N 11 136 16 12 21 18 67 cases, plot was retained in regression models as a random factor. S 2 1009 48 57 73 77 255 Because the data used to address objective 3 were differences Y 13, 14, 17 179 29 32 51 45 157 –– from 2009 values, we encountered situations in which all values Total 323 391 508 489 1711 were zero because the species was not documented in the plot An.d. = no data. 8 The Rangeland Journal R. B. Harris et al.

Table 2. Predicted order of abundance (from most, 1, to least 11) of response variables (rows) by pastoralist (columns) Values are from models accounting for the effects of pastoralist-specific topography (see Supplementary Materials table S3), with plot and the date of observation within each year as random factors. Row entries sharing letter codes were not significantly different from one another (P < 0.05; Tukey’s HSD test)

Pastoralist B D1 D2 G H K L1 L2 N S Y Astragalus/Oxytropis 1a 5a 2a 6a 7a 8a 9a 11a 10a 4a 3a Cardamine 11a 3a 7a 9a 8a 2a 5a 4a 6a 1a 10a Carex spp. 1a 7a 3a 11a 8a 5a 2a 6a 10a 4a 9a Heteropappus altaicus 9a 4a 1a 3a 6a 5a 8a 7a 10a 2a 11a Leymus secalinus 1a 9ab 2a 3a 7ab 11b 5ab 10ab 4a 8ab 6ab Poa spp. 1a 4ab 2ab 11b 9b 8ab 3ab 10b 6ab 5ab 7ab Potentilla bifurca 6a 2a 10a 11a 1a 4a 5a 7a 9a 3a 8a Stipa purpurea 2abc 6bc 10abc 7abc 8bc 11c 5abc 9bc 1a 3ab 4abc Thermopsis lanceolata 6ab 10ab 7ab 11ab 9b 2ab 5ab 3ab 4ab 1a 8ab Total grass herbage mass 1abc 9bcd 3abc 5abc 6abc 11d 8abcd 10cd 2a 7abc 4ab Unpreferred herbage mass 11ab 2a 6ab 9b 8ab 3a 1a 5ab 10b 4a 7ab Unpreferred : preferred ratio 11b 2ab 6ab 9b 7b 1a 4ab 5ab 10b 3ab 8b Erosion index 11a 8a 9a 1a 3a 2a 10a 6a 5a 4a 7a Bare soil 11a 9a 6a 1a 3a 2a 10a 7a 5a 4a 8a Litter cover 1a 9cd 4abc 2a 6ab 10d 7abcd 11d 3ab 8bcd 5abc

had higher cover in pastures controlled by pastoralists B, D2, grasses, and second-to-last in the cover of unpreferred forbs, G, and N than in pastures controlled by pastoralist K. Poa including Oxytropis spp. and H. altaicus. In contrast were spp. had higher cover in pastures controlled by pastoralist B than pastures showing evidence of stress, including those operated in those of pastoralists H and L2. The dominant and preferred by pastoralist K (ranked second in cover of bare soil and perennial S. purpurea had higher cover in pastures controlled erosion index, second-to-last in cover of litter, first in the ratio by pastoralist N than all others, and a lower cover in the pasture of unpreferred : total vegetation, and last in relative cover of of pastoralist K than all others. The disturbance-associated both S. purpurea and L. secalinus); pastoralist S (characterised legume, T. lanceolata, had higher cover in the pasture of by high proportions of unpreferred species, and first in the cover pastoralist S than others, and a lower cover in pastures controlled of the unpreferred legume T. lanceolata), and pastoralist G by pastoralist H than others. Due in part to the small samples (ranked first in cover of bare soil and erosion, and last in the sizes owing to their rarity, cover of Astragalus/Oxytropis, cover of preferred graminoids, Poa and Carex spp.). Other Cardamine, Carex spp., H. altaicus, and P. bifurca, accounting pastoralists’ areas were intermediate in these indices of stress for topography, did not differ (Tukey’s HSD text; P < 0.05) (Table 2). among pastoralists (Table 2). Total herbage mass consisting of (generally preferred) grass species was greater in the pastures of pastoralist N than in most Associations of mean stocking rate and density of sheep others, and less in the pasture of pastoralist K than most others. with herbage mass of species and erosion indicators Total herbage mass consisting of unpreferred species was greatest Stocking rate at the pasture scale varied from 0 to 5.89 sheep ha–1 among pastures controlled by pastoralists D1, K, L1 and S than annually (Table 3). We found no associations of mean stocking it was among those controlled by pastoralist G. The ratio of rate in 2009–2011 at the pasture scale with indicators of erosion unpreferred : preferred herbage mass of species was greater in at the scale of individual vegetation plots. We found positive pastures controlled by pastoralist K than those controlled by associations of mean stocking rate (during 2009–2011) at the pastoralists B, G, H, N, and Y. Litter constituted a higher pasture scale with herbage mass of litter (b = 0.176, s.e. = 0.043, proportion of pastures controlled by pastoralists B and G than t = 4.05, P < 0.001) and total herbage mass of grasses (b = 0.237, pastoralists D1, K, L2, and S, a lower proportion of pastures s.e. = 0.103, t = 2.31, P < 0.05) at the plot scale (Table 4a). We controlled by pastoralist L2 than B, D2, G, H, N, and Y found negative associations of mean stocking rate at the pasture (Table 2). Neither the cover of bare soil nor erosion index was scale with total fresh herbage mass (b = –0.316, s.e. = 0.130, significantly affected by pastoralists. t = –2.42, P < 0.05), fresh herbage mass of unpreferred species Notable pastures that appeared, on the basis of these (b = –0.532, s.e. = 0.147, t = –3.63, P < 0.001), and the proportion rankings, to be in relatively ‘healthy’ condition were those of the total herbage mass consisting of unpreferred species operated by pastoralist B, whose pastures ranked lowest in (b = –0.063, s.e. = 0.017, t = –3.80, P < 0.001; Table 4b). erosion indicators of over-use such as cover of bare soil and There were no consistent trends of any response variable erosion index, first in proportion of litter cover and of the with the proportional pasture use. When combining the metrics preferred Poa spp., L. secalinus and Carex spp., and ranked low of mean stocking rate and proportional pasture use, we found in the cover of unpreferred forbs, Cardamine spp. and negative associations of stocking density with the per cent H. altaicus; and pastoralist N, whose pastures portrayed cover of bare soil (b = –1.713, s.e. = 0.875, t = –1.96, P = 0.05), intermediate indicators of over-use, but ranked first in the cover herbage mass of unpreferred species (b = –3.662, s.e. = 1.397, of the preferred grass S. purpurea, second in the cover of total t = –2.62, P = 0.01), and the ratio of the herbage mass of Rangeland responses on Tibetan steppe The Rangeland Journal 9

– Table 3. Estimated sheep densities (individuals ha 1) on each pasture table S15). This decline was not associated with stocking rate in the study, Gouli township, Dulan County, Qinghai Province, at the pasture scale or with stocking density at the plot scale. – 2010 2012 The decline with time, however, was greater following pasture Pastoralist Pasture 2010 2011 2012 Mean use by higher than lower stocking rates of sheep (year b = –1.656, s.e. = 0.350, t = –4.72, P < 0.001; sheep density K 1 0.03 0.04 0.07 0.05 b = –0.567, s.e. = 0.213, t = –2.65, P < 0.01, interaction –0.441, S 2 0.24 0.00 0.20 0.15 s.e = 0.140, t = –3.14, P < 0.01; Fig. 3f; table S16). The herbage L1 5 0.16 0.00 0.93 0.37 mass of L. secalinus was negatively associated with stocking L2 6 0.57 2.71 0.66 1.31 rate in the previous winter, the relationship strengthening D2 7 0.00 0.00 0.00 0.00 b – G 9 4.99 1.95 2.16 3.03 throughout the study period (density = 0.326, s.e. = 0.107, N 11 1.84 1.58 1.27 1.56 t = –3.05, P < 0.01; year b = 0.453 s.e. = 0.223, t = 2.04, P < 0.05; Y 14 1.15 1.62 0.73 1.17 interaction b = –0.208, s.e. = 0.084, t = –2.49 P < 0.05; Fig. 3g; H 15 1.47 2.13 1.27 1.63 table S17). When the interaction of changes over time with H 16 1.70 1.79 1.47 1.66 changes following winter grazing by sheep were modelled Y 17 3.05 5.89 1.93 3.62 together, we observed an increasingly positive response by Poa spp. with stocking rate (density b = 0.402, s.e. = 0.155, t = 2.6, unpreferred species to total herbage mass (b = –0.392, P < 0.01; interaction b = 0.344, s.e. = 0.111, t = 3.10, P < 0.01; s.se. = 0.158, t = –2.47, P < 0.05; Table 5). Fig. 3h; table S18). No relationships with time, stocking rate at the pasture scale, or stocking density at the plot scale were Effects of temporal variation in stocking rate of sheep on observed for other species considered individually. erosion indicators and herbage mass of species Cover of bare soil increased with time (b = 5.760, s.e. = 0.743, Discussion < t = 7.759, P 0.0001; table S5), as well as with pasture-scale Site effects on vegetation stocking rate (b = 1.289, s.e. = 0.445, t = 2.89, P < 0.01; table S6, Fig. 3a). The year sheep density interaction was also significant Our focus was on responses to the management practices of and positive, with the trend on time being greater in those livestock by individual pastoralists but, because their pastures pastures with higher sheep density (interaction b = 1.645, varied in their inherent biological and site characteristics, it was fi s.e. = 0.321, t = 5.12, P < 0.0001 table S7). important to rst quantify characteristics of rangelands by site, fl The erosion index did not differ with time but was positively independently of any in uences introduced by heterogeneity related to the stocking rate of sheep on the pasture during in management. Before one can understand any effects that the preceding winter (b = 0.040, s.e. = 0.019, t = 2.08, P < 0.05; livestock management may have had on the herbage mass of Fig. 3b, table S8), although not at the plot scale with grazing a plant species, one needs to understand whether that species pressure. When viewed together, the increase in erosion with would be expected to be common or not in that pasture stocking rate at the pasture scale increased with time (interaction regardless of management. For example, a high herbage mass of stocking rate year b = 0.037, s.e. = 0.015, t = 2.55; P = 0.01; of noxious legumes of the Astragalus/Oxytropis spp. would table S9). generally be considered indicators of over-use. That pastures Cover of live vegetation declined with time (b = –7.544, managed by pastoralists L2 and N had similarly low herbage s.e. = 0.622, t = –12.14, P < 0.0001; table S10), and was not masses of these legumes (Table 2) might, therefore, suggest fl related to stocking rate in the previous winter. However, when a similarity of management in uences. These legumes were time and stocking rate in the previous winter were combined in associated with steep slopes (table S2); the pastures of pastoralist a single model, all were highly significantly negative (table S11), L2 were the steepest (table S3) whereas those of pastoralist N fl suggesting that vegetation cover declined more in areas with were relatively at. Thus, we would not expect to encounter greater than lower stocking rate (Fig. 3c). many of these legumes in the pastures of pastoralist N, whereas We documented no relationships among stocking rate at the paucity found in the pastures of pastoralist L2 is surprising the pasture scale or stocking density at the plot scale with and potentially more informative. Similarly to our analysis of total fresh herbage mass. However, herbage mass of grasses annual variation in weather, accounting for site effects allowed ’ declined with time (b = –0.956, s.e. = 0.335, t = –2.86, P < 0.01; us to isolate effects attributable to pastoralists management table S12). The decline of herbage mass of grasses with time from those beyond their immediate control. was greater in pastures with higher than lower stocking rate fl (Fig. 3d, table S13). In uence of individual pastoralists No relationships among time, stocking rate at the pasture Yeh and Gaerrang (2011) showed that, despite joint scale, or at the plot scale were observed with herbage mass of membership in a small, seemingly cohesive village, pastoralists Carex spp. However, among unpreferred forbs, we found that differed in their approach to livestock management, but these herbage mass was positively associated with density of sheep authors’ analyses did not extend to possible consequences of grazing in the previous winter (b = 1.496, s.e. = 0.741, t = 2.02, this heterogeneity as expressed by vegetation. Our analyses P < 0.05), more so towards the end than the beginning of the provided insight into the magnitude of the effects that individual study period (Fig. 3e; table S14). variation in management had on the rangelands over and above Herbage mass of S. purpurea declined with time during annual differences due to temperature and inherent biological the study (year b = –1.412, s.e. = 0.312, t = –4.52, P < 0.001; differences arising from heterogeneity in the site of the pasture. 10 The Rangeland Journal R. B. Harris et al.

Table 4. Linear models relating various response variables at the plot scale to the mean density of sheep on the scale of the individual pasture during winters 2009–2011 For each model, coefficients (b), standard errors (s.e.), Student’s t, and P-value are shown for fixed-effect explanatory variables; plot was also included in each model as a random effect. a. Models with mean sheep density positively associated with litter and total grass biomass. b. Models with mean sheep density negatively associated with total fresh biomass, total biomass of unpalatable species, and the proportion of total biomass consisting of unpalatable species. Response variables were square-root transformed

a. Litter n = 1581, Adj. R2 = 0.696 b s.e. tP Julian date 0.014 0.013 1.08 0.2785 Julian date2 –0.001 0.001 –1.29 0.1985 Elevation 0.001 0.001 0.99 0.3206 Slope –0.020 0.003 –6.05 <0.0001 North deviation 0.002 0.001 2.83 0.0049 East deviation –0.001 0.001 –1.00 0.3182 Spring cumulative temperature –1.671 0.059 –28.36 <0.0001 January temperature 0.607 0.015 40.61 <0.0001 Pasture sheep density 0.176 0.043 4.05 <0.0001 Herbage mass grass species, n = 1575, Adj. R2 = 0.876 b s.e. tP Julian date 0.086 0.012 6.90 <0.0001 Julian date2 –0.001 0.001 –7.50 <0.0001 Elevation –0.002 0.001 –2.05 0.0416 Slope –0.035 0.008 –4.37 <0.0001 North deviation 0.006 0.002 2.99 0.0030 East deviation –0.004 0.003 –1.38 0.1691 Spring cumulative temperature 0.290 0.055 5.25 <0.0001 January temperature –0.130 0.014 –9.18 <0.0001 Pasture sheep density 0.237 0.103 2.31 0.0216 b. Total herbage mass n = 1574, Adj. R2 = 0.694 b s.e. tP Julian date 0.386 0.027 14.46 <0.0000 Julian date2 –0.001 0.001 –15.06 <0.0001 Elevation 0.002 0.001 2.47 0.0139 Slope –0.036 0.010 –3.58 0.0004 North deviation 0.004 0.002 1.57 0.1175 East deviation –0.002 0.004 –0.50 0.6181 Spring cumulative temperature 0.483 0.119 4.06 <0.0001 January temperature –0.208 0.030 –6.86 <0.0001 Pasture sheep density –0.316 0.130 –2.42 0.0159 Unpreferred herbage mass, n = 1570, Adj. R2 = 0.794 b s.e. tP Julian date 0.284 0.023 12.38 <0.0000 Julian date2 –0.001 0.000 –12.79 <0.0001 Elevation 0.005 0.001 4.03 <0.0001 Slope –0.003 0.011 –0.23 0.8195 North deviation –0.001 0.003 –0.29 0.7756 East deviation 0.004 0.004 1.01 0.3153 Spring cumulative temperature 0.056 0.102 0.55 0.5816 January temperature –0.039 0.026 –1.50 0.1335 Pasture sheep density –0.532 0.147 –3.63 0.0003 Unpreferred : total herbage mass, n = 1526, Adj. R2 = 0.860 b s.e. tP Julian date 0.010 0.002 4.59 <0.0001 Julian date2 0.001 0.001 –4.67 <0.0001 Elevation 0.000 0.000 3.16 0.0018 Slope 0.003 0.001 2.18 0.0303 North deviation 0.000 0.000 –1.50 0.1348 East deviation 0.001 0.000 1.56 0.1204 Spring cumulative temperature –0.022 0.009 –2.35 0.0188 January temperature 0.008 0.002 3.23 0.0013 Pasture sheep density –0.063 0.017 –3.80 0.0002 Rangeland responses on Tibetan steppe The Rangeland Journal 11

Table 5. Linear models relating various response variables at the plot scale to the mean grazing pressure (mean sheep density/pasture ¾ pasture-specific proportional use) during winters 2009–2011 For each model, coefficients (b), their standard errors (s.e.), Student’s t, and P-value are shown for fixed-effect explanatory variables; plot was also included in each model as a random effect. Models shown indicate that mean grazing pressure was negatively associated with percent bare soil, biomass of unpalatable biomass, and the proportion of total biomass consisting of unpalatable species. Response variables were square-root transformed

Bare soil, n = 1584; Adj. R2 = 0.658 b s.e. tP Elevation –0.001 0.001 –1.72 0.0871 Slope –0.009 0.007 –1.19 0.2354 North deviation –0.002 0.002 –1.31 0.1912 East deviation –0.001 0.003 –0.33 0.7397 Julian date –0.077 0.018 –4.147 <0.0000 Julian date2 0.001 0.001 4.13 <0.0001 Spring cumulative temperature 0.979 0.083 11.73 <0.0001 January temperature –0.135 0.021 –6.35 <0.0001 Stocking rate proportional use –1.713 0.875 –1.96 0.0511 Unpreferred herbage mass, n = 1570; Adj. R2 = 0.794 b s.e. tP Elevation 0.005 0.001 4.63 <0.0001 Slope –0.002 0.012 –0.13 0.8949 North deviation –0.002 0.003 –0.73 0.4680 East deviation 0.004 0.004 0.96 0.3388 Julian date 0.283 0.023 12.32 <0.0000 Julian date2 –0.001 0.001 –12.73 <0.0001 Spring cumulative temperature 0.055 0.102 0.53 0.5933 January temperature –0.039 0.026 –1.49 0.1358 Stocking rate proportional use –3.662 1.397 –2.62 0.0092 Unpreferred : total herbage mass, n = 1562, Adj. R2 = 0.861 b s.e. tP Elevation 0.001 0.001 3.85 0.0001 Slope 0.003 0.001 3.24 0.0309 North deviation –0.001 0.001 –2.02 0.0446 East deviation 0.001 0.001 1.49 0.1386 Julian date 0.009 0.002 4.54 <0.0000 Julian date2 0.001 0.001 –4.62 <0.0001 Spring cumulative temperature –0.022 0.009 –2.37 0.0181 January temperature 0.008 0.002 3.24 0.0012 Stocking rate proportional use –0.392 0.158 –2.47 0.0139

We observed patterns suggesting that some pastures were less using livestock exclosures, suggested that most QTP steppe liable to change than others, even taking site differences into vegetation appears to be well adapted to grazing in winter, at least account. For example, pastures such as those allocated to at low stocking rates. In most plant species, increased intra- pastoralists B and N, were characterised both by a higher cover or inter-specific competition for resources in the absence of of preferred species (e.g. S. purpurea, L. secalinus and Poa herbivory evidently counteracted whatever benefits plants spp.) as well as less bare soil and erosion. In contrast, cover of enjoyed from a respite from herbivory. Thus, the possibility T. lanceolata, P. bifurca and Cardamine spp. on pastures such that grazing actually increased herbage production (Hilbert as those of pastoralists K and H was generally associated with et al. 1981; McNaughton 1983) in this system should not be greater bare soil and erosion. Analysis of factors that might dismissed. Harris et al.(2015), however, also found that grazing have caused pastoralists, who shared cultural practices and increased erosion over conditions prevailing within livestock were exposed to a similar policy environment, to differ in their exclosures. management approaches is beyond the scope of this paper. A simpler interpretation however is that pastoralists, returning to winter pasture in October, adjusted their stocking levels to their (generally correct) perceptions of the herbage mass of the Spatial associations among mean grazing levels preferred vegetation. Despite pastoralists occasionally stating and rangeland response during interviews that they lacked ability to distinguish one Our analyses relating stocking rate and density to changes species from another (sometimes referring to all vegetation in vegetation and erosion indicators (Tables 4, 5) could be simply as ‘grass’), our analyses suggest that pastoralists rationally interpreted as suggesting that stocking level was negatively adjusted herd size in winter to reflect the herbage mass of associated with cover of bare soil and unpreferred species, preferred grasses, and actually reduced grazing pressure on while being positively associated with litter and herbage mass specific pastures if herbage consisted disproportionately of of grass species. Might grazing have had such unexpected and unpreferred species. Within pastures, the proportion of time beneficial effects? Analyses conducted by Harris et al.(2015), spent near each plot by GPS-monitored sheep was lower where 12 The Rangeland Journal R. B. Harris et al.

40 (a)(0.5 b) 35 0.4 30 0.3 25 0.2 20 0.1 15 0 Bare soil

10 Erosion index –0.1 5 –0.2 0 –0.3 –5 0 (c)(0 d)

–5 –2

–10 –4 –15 –6 –20 –8 –25 –10

Vegetation cover Vegetation –30 –35 –12 –40 –14

15 (e)(0 f ) 13 –1 11 –2 –3 9 –4 7 –5 5 –6 3 –7 herbage mass herbage mass 1 purpurea dry S. –8

Unpreferred forb fresh forb Unpreferred –1 –9 –3 –10 (g)(h) 1.5 5

1.0 4 0.5 3 0 2 –0.5 1 –1.0 spp. fresh spp. Poa herbage mass herbage mass fresh herbage mass Grass fresh secalinus L. –1.5 0 –2.0 –1 012345 012345 Sheep ha–1

Fig. 3. Differences from summer 2009 levels (summer 2010 solid, summer 2011 dash, and summer 2012 dot-dash) as a function of the density of sheep on the pasture during the preceding winter. (a) Proportion of plots devoid of vegetation (i.e. bare soil). (b) Index of erosion. (c) Proportion of plots covered by live vegetation. (d) Fresh herbage mass of all grass species. (e) Fresh herbage mass of unpreferred forbs. (f) Dry biomass of Stipa purpurea.(g) Fresh biomass of L. secalinus.(h) Fresh biomass of Poa spp. unpreferred species were higher in herbage mass than elsewhere. soils, displaying patterns generally consistent with the hypothesis Thus, both stocking rate, comparing among pastures, and that, at least at sufficiently intense levels, herbivory of senescent stocking density, viewed within each pasture, responded vegetation and winter-time compaction reduced subsequent adaptively, increasing with vegetation selectively eaten by herbage masses of many preferred species. The models, sheep (Harris et al. 2015; see also Cincotta et al. 1991). summarised in Fig. 3, used as explanatory variables pasture- specific measurements of stocking rate during the winter Rangeland response to annual variation in stocking rate preceding the observed rangeland response. Thus, unlike the Importantly however, our dynamic analyses indicate that static analyses summarised in Tables 4 and 5, the two types of grazing exerted a negative effect on preferred vegetation and variables were not contemporaneous, but rather followed one Rangeland responses on Tibetan steppe The Rangeland Journal 13 another in sequence. Because our data were observational We detected annual responses in most indicators of erosion, rather than experimental, it is possible that relationship shown as well as in the herbage mass of most (albeit not all) preferred in Fig. 3 were not causal, but rather that causes resided in some species to sheep grazing pressure during the preceding dormant other, unmeasured variable. However, it could not have been season. Accounting for annual weather fluctuations and site the case that the response variables depicted in Fig. 3 (e.g. variability, both cover of bare soil and erosion varied positively herbage mass of S. purpurea) produced the stocking rates with stocking rate of sheep. Although total herbage mass was documented, because the stocking events occurred and ended not related to stocking rate of sheep, total vegetation cover, before the response variables (e.g. vegetation growth) were herbage mass of S. purpurea and of all grass species varied quantified. inversely with stocking rate of sheep in the preceding winter, Accounting for fluctuations in annual temperature and site whereas herbage mass of unpreferred forbs varied directly variability, rangeland condition, as measured, appeared to with stocking rate. Within pastures, the increase in cover of worsen during the 4 years of our study, independently of annual bare soil was more pronounced where stocking rate was higher, stocking rate of sheep. The cover of bare soil increased, and and per cent live vegetation cover and herbage mass of vegetation cover decreased during 2009–2012. Although total S. purpurea declined more strongly where stocking rate was fresh herbage mass did not change, herbage mass of S. purpurea, higher than lower. as well as total grasses, declined. We found no evidence of Our data diverge from the opinions of most of the pastoralists temporal changes in the herbage mass of L. secalinus, and Poa who, although recognising their pastures had finite capacities, and Carex spp. However, the herbage mass of unpreferred tended to view possible negative effects of heavy stocking only species, as well as the proportion of unpreferred species of the in terms of potential livestock mortality, and not in rangeland total herbage mass, increased with time during 2009–2012. Two productivity itself, although they did note a downward trend reasonable hypotheses to consider to explain these dynamics in the liveweight of sheep without attributing it to grazing. In are long-term climate changes, associated with changes in soil our study area, pastoralists had the ability to encourage the moisture content or temperature, or alternatively, a lagged summer-time herbage mass of preferred species, and discourage response to herbivory levels before our study. expansion of unpreferred species, through their stocking levels Quantifying the size of the negative effects of winter-time in winter. stocking rates of sheep on the subsequent herbage mass of the preferred grasses is not straightforward. Most effects we found were complicated by the general, downward trend in Acknowledgements herbage mass observed among most species through time. Principal funding for this work was from the USA National Science Our dynamic models, showing trends in herbage mass resulting Foundation, Dynamics of Coupled Natural and Human Systems Program, from changes in the stocking rates of sheep, considered all plots Award 0815441. Supplementary funding was provided by the Trace but the starting cover of each species varied by pasture and Foundation, and the Bridge Fund. Field data were ably collected by pastoralist. The size of effects associated with the highest Chungjyid, Dorjiejyal, Drubgyal, Gurudorjie, Hulchendorjie, Lamojia, stocking rates documented in this study, were substantial. For Pagmostso, Pemabum, Puhuadongzhi, Rinchentso, Sonam, Sonamtso, S. purpurea, we estimated the mean herbage mass in 2009 over Tseringdorjie, and Wanmananqing. Pemabum ably ran the field station. – fi all plots as ~180 kg DM ha 1. Our models suggested that by For eld assistance, we thank Zhou J. K., Ma L. L., Shi Y. H., and Qi 2012, at a stocking rate of 2 sheep ha–1 would have reduced S. F. Statistical assistance was provided by B. Steele. For administrative –1 support, we thank L. Arends, T. Baerwald, H. Bjorn, W. Bleisch, J. Burchfield, herbage mass by ~100 kg DM ha , or more than half. M. Garry, Kunchok Gelek, and E. Yang. The map was produced by Studies in experimental situations have suggested that G. Maclaurin. liveweight gains by sheep were negatively correlated with stocking rates (Zhou et al. 1995), and among yaks were positively correlated with ratios of preferred to unpreferred References plants (Dong et al. 2003). Additional studies of the relationships among stocking rates, herbage mass of preferred species, and Allen, V. G., Batello, C., Berretta, E. J., Hodgson, J., Kothmann, M., Li, X., McIvor, J., Milne, J., Morris, C., Peeters, A., and Sanderson, M. (2011). liveweights of sheep in working pastures would be useful. An international terminology for grazing lands and grazing animals. Grass and Forage Science 66,2–28. doi:10.1111/j.1365-2494.2010. 00780.x Conclusions Banks, T. J. (2003). Property rights reform in rangeland China: dilemmas Our analysis shows that pastoralists stock their pastures with on the road to the household ranch. World Development 31, 2129–2142. sheep on the basis of the herbage mass of preferred grasses, not doi:10.1016/j.worlddev.2003.06.010 on the basis of total herbage mass. We found that the stocking Banks, T. J., Richard, C., Li, P., and Yan, Z. L. (2003). Governing the rate of sheep at the pasture scale was negatively associated grasslands of Western China. Mountain Research and Development – with total herbage mass and of the herbage mass of unpreferred 23, 132 140. doi:10.1659/0276-4741(2003)023[0132:CGMIWC]2.0. forbs, but was positively related to the herbage mass of grasses. CO;2 Bauer, K. (2005). Development and the enclosure movement in pastoral Stocking rate at the pasture scale did not appear to be related to Tibet since the 1980s. Nomadic Peoples 9,53–81. doi:10.3167/08227 indicators of erosion (cover of bare soil, total vegetation cover, 9405781826119 and erosion index) but, within pastures, sheep tended to avoid Bauer, K., and Yonten Nyima, (2010). Laws and regulations impacting the areas with a relatively large cover of bare soil and relatively enclosure movement on the Tibetan Plateau of China. Himalaya 30, large proportions of unpreferred vegetation. 23–38. 14 The Rangeland Journal R. B. Harris et al.

Bedunah, D. J., and Angerer, J. P. (2012). Rangeland degradation, poverty, Hilbert, D. W., Swift, D. M., Detling, J. K., and Dyer, M. I. (1981). Relative and conflict: how can rangeland scientists contribute to effective growth rates and the grazing optimization hypothesis. Oecologia 51, responses and solutions? Rangeland Ecology and Management 65, 14–18. doi:10.1007/BF00344645 606–612. doi:10.2111/REM-D-11-00155.1 Holzner, W., and Kreichbaum, M. (2001). Pastures in south and central Bian, J. H., Jing, Z. C., and Fan, N. C. (1999). The effect of grassland Tibet (China): probable causes of pasture degradation. Die Bodenkultur fencing on the population density of plateau pikas. Acta Biologica 52,37–44. Plateau Sinica 14, 110–115. [in Chinese] Hutchinson, M. F. (1995). Interpolating mean rainfall using thin plate Chen, S. (1996). Inner Asian grassland degradation and plant transformation. smoothing splines. International Journal of Geographical Information In: ‘Cultural and Environment in Inner Asia, Volume 1: The Pastoral Science 9, 385–403. doi:10.1080/02693799508902045 Economy and the Environment’. (Eds C. Humphrey and D. Sneath.) Hutchinson, M. F. (2001). ‘ANUSPLIN version 4.2. User Guide.’ pp. 111–123. (The White Horse Press: Cambridge, UK.) (Centre for Resource and Environmental Studies, Australian National Cincotta, R. P., van Soest, P. J., Robertson, J. B., Beall, C. M., and Goldstein, University: Canberra, ACT.) M. C. (1991). Foraging ecology of livestock on the Tibetan Changtang: Hutchinson, M. F., McKenney, D. W., Lawrence, K., Pedlar, J. H., a comparison of three adjacent grazing areas. Arctic and Alpine Hopkinson, R. F., Milewska, E., and Papadopol, P. (2009). Development Research 23, 149–161. doi:10.2307/1551379 and testing of Canada-wide interpolated spatial models of daily Damiran, D. (2005). ‘Palatability of Mongolian Rangeland Plants.’ (Eastern minimum-maximum temperature and precipitation for 1961–2003. Oregon Agricultural Research Center: Union Station, OR.) Journal of Applied Meteorology and Climatology 48, 725–741. Delibes-Mateos, M., Smith, A. T., Slobodchikoff, C. N., and Swenson, J. E. doi:10.1175/2008JAMC1979.1 (2011). The paradox of keystone species persecuted as pests: a call for Immerzeel, W. W., van Beek, L. P. H., and Bierkens, M. F. P. (2010). the conservation of abundant small mammals in their native range. Climate change will affect the Asian water towers. Science 328, Biological Conservation 144, 1335–1346. doi:10.1016/j.biocon.2011. 1382–1385. doi:10.1126/science.1183188 02.012 Lai, C. H., and Smith, A. T. (2003). Keystone status of plateau pikas Dong, Q. M., Zhao, X. Q., Ma, Y. S., Li, Q. Y., Wang, Q. J., and Shi, J. J. (Ochotona curzoniae): effect of control on biodiversity of native birds. (2003). Studies on the relationship between grazing intensity for yaks Biodiversity and Conservation 12, 1901–1912. doi:10.1023/A:102416 and plant groups in Kobresia parva alpine meadow. Acta Agrestia Sinica 1409110 13, 334–343. [in Chinese, English abstract] Levine, N. E. (1998). From nomads to ranchers: managing pasture among Du, F. C. (2012). Ecological resettlement of Tibetan herders in the ethnic Tibetans in Sichuan. In: ‘Development, Society and Environment Sanjiangyuan: a case study in Madoi County of Qinghai. Nomadic in Tibet, Proceedings of the Seventh Seminar of the International Peoples 16, 116–133. doi:10.3167/np.2012.160109 Association for Tibetan Studies’. (Eds G. E. Graz and G. E. Clarke.) Fan, N. C., Zhou, W. Y., Wei, W. H., Wang, Q. Y., and Jiang, Y. J. (1999). pp. 69–119. (Verlag der Österreichischen Akademie der Wissenchasften: Rodent pest management in the Qinghai-Tibet alpine meadow Vienna, Austria.) ecosystem. In: ‘Ecologically-based Management of Rodent Pests’. (Eds Li, M. S. (1994). Characteristics and rational exploitation of Tibet’s land G. R. Singleton, L. A. Hinds, H. Leirs and Z. B. Zhang.) pp. 285–304. resources. Journal of Natural Resources 9,51–57. [in Chinese] (Australian Centre for International Agricultural Research: Canberra, Li, X. L., Gao, J., Brierley, G., Qiao, Y. M., Zhang, J., and Yang, Y. W. (2013). ACT.) Rangeland degradation on the Qinghai-Tibet plateau: implications Foggin, J. M. (2008). Depopulating the Tibetan Grasslands: national policies for rehabilitation. Land Degradation & Development 24,72–80. and perspectives for the future of Tibetan herds in Qinghai Province, doi:10.1002/ldr.1108 China. Mountain Research and Development 28,26–31. doi:10.1659/ Liu, R. T. (1986). ‘Aksai Kazak Autonomous County Grassland Resources mrd.0972 and Planning.’ (Gansu Province Aksai Kazak Autonomous County Goldstein, M. C., and Beall, C. M. (2002). Changing patterns of Tibetan Agricultural Planning Office: Gansu, China.) [in Chinese] nomadic pastoralism. In : ‘Human Biology of Pastoral Populations’. (Eds Liu, Y. S. (1995). International hunting and the involvement of local W. R. LeonardandM. H. Crawford.)pp. 131–150.(CambridgeUniversity people, Dulan, Qinghai, People’s Republic of China. In: ‘Integrating Press: Cambridge, UK.) People and Wildlife for a Sustainable Future’. (Eds J. A. Bissonette Grant, A. (2003). A study of the implementation of China’s sloping land and P. R. Krausman.) pp. 63–67. (The Wildlife Society: Bethesda, MD.) conversion policy ‘tui geng huan lin’: a case study – Hanyuan County, Liu, Q. X., Harris, R. B., Wang, X. M., and Wang, Z. H. (2007). Home range Sichuan Province. Forests. Trees and Livelihoods 13, 331–343. size and overlap of Tibetan foxes (Vulpes ferrilata) in Dulan County, doi:10.1080/14728028.2003.9752469 Qinghai Province. Acta Theriologica Sinica 27, 370–375. [in Chinese] Hardin, G. (1968). The tragedy of the commons. Science 162, 1243–1248. Liu, Q. X., Harris, R. B., and Wang, X. M. (2010). Food habits of Tibetan doi:10.1126/science.162.3859.1243 fox (Vulpes ferrilata) in the Kunlun Mountains, Qinghai Province, China. Harris, R. B. (2008). ‘Wildlife Conservation in China: Preserving the Mammalian Biology 75, 283–286. doi:10.1016/j.mambio.2009.02.002 Habitat of China’s Wild West.’ (M.E. Sharpe, Inc.: Armonk, NY.) McBeath, J. H., and McBeath, J. (2010). ‘Environmental Change and Food Harris, R. B. (2010). Rangeland degradation on the Qinghai-Tibetan Security in China.’ (Springer Dordrecht: Heidelberg, Germany.) plateau: a review of the evidence of its magnitude and causes. Journal of McNaughton, S. J. (1983). Compensatory plant growth as a response to Arid Environments 74,1–12. doi:10.1016/j.jaridenv.2009.06.014 herbivory. Oikos 40, 329–336. doi:10.2307/3544305 Harris, R. B., Bedunah, D. J., Yeh, E. T., Smith, A. T., and Anderies, J. M. Miller, D. J., Bedunah, D. J., Pletscher, D. H., and Jackson, R. M. (1992). (2010). Determinants of rangeland dynamics on the Qinghai-Tibet From open range to fences: changes in the range-livestock industry plateau, China: livestock, wildlife, and pastoralism. Pastoralism 1, on the Tibetan Plateau and implications for development planning 325–326. and wildlife conservation. In: ‘Proceedings of the 1992 International Harris, R. B., Wang, W. Y., Badinqiuying, , Smith, A. T., and Bedunah, D. J. Rangeland Development Symposium’. (Eds G. K. Perrier and C. W. Gay.) (2015). Herbivory and competition of Tibetan steppe vegetation in pp. 95–109. (Society for Range Management: Littleton, CO.) winter pasture: effects of livestock exclosure and plateau pika reduction. NRC (1994). ‘Rangeland Health: New Methods to Classify, Inventory, and PLoS One 10, e0132897. doi:10.1371/journal.pone.0132897 Monitor Rangelands.’ (National Academies Press: Washington, DC.) Rangeland responses on Tibetan steppe The Rangeland Journal 15

Richard, C., Yan, Z. L., and Du, G. Z. (2006). ‘The Paradox of the Individual Williams, D. M. (2002). ‘Beyond Great Walls: Environment, Identity, and Household Responsibility System in the Grasslands of the Tibetan Development on the Chinese Grasslands of Inner Mongolia.’ (Stanford Plateau, China.’ USDA Forest Service Proceedings RMPS-P-39. (USDA University Press: Stanford, CA.) Forest Service: Fort Collins, CO.) Wilson, M. C., and Smith, A. T. (2015). The pika and the watershed: Schaller, G. B. (1998). ‘Wildlife of the Tibetan Steppe.’ (University of the impact of small mammal poisoning on the ecohydrology of the Chicago Press: Chicago, IL.) Qinghai-Tibetan plateau. Ambio 44,16–22. doi:10.1007/s13280-014- SEPA (State Environmental Protection Agency Nature Reserve Bureau) 0568-x (2007). Report on environmental 2006: Grasslands. Available at: www. Wu, N., and Yan, Z. L. (2002). Climate variability and social vulnerability mep.gov.cn (accessed 18 June 2007)[in Chinese] on the Tibetan Plateau: dilemmas on the road to pastoral reform. Shang, Z. H., Gibb, M. J., Leiber, F., Ismail, M., Ding, L. M., Guo, X. S., and Erdkunde 56,2–14. doi:10.3112/erdkunde.2002.01.01 Long, R. J. (2014). The sustainable development of grassland-livestock Wu, N., Yan, Z. L., and Lu, T. (2012). Enclosure and resettlement in the systems on the Tibetan plateau: problems, strategies and prospects. The eastern Tibetan Plateau: dilemma of pastoral development during Rangeland Journal 36, 267–296. doi:10.1071/RJ14008 the last three decades. In: ‘Pastoral Practices in High Asia: Agency Shapiro, J. (2001). ‘Mao’s War against Nature: Politics and the of ‘Development’ Effect by Modernization, Resettlement and Environment in Revolutionary China.’ (Cambridge University Press: Transformation’. (Ed. H. Kreutzmann.) pp. 291–306. (Springer Cambridge, UK.) Dordrecht: Heidelberg, Germany.) Shen, Y. Y., Ma, Y. S., and Li, Q. Y. (2004). Grassland restoration in Xu, J., Grumbine, R. E., Shrestha, A., Eriksson, M., Yang, X., Wang, Y., and Dari County, Qinghai Province. In: ‘Implementing the Natural Forest Wilkes, A. (2009). The melting Himalayas: cascading effects of climate Protection Program and the Sloping Lands Conversion Programs: change on water, biodiversity, and livelihoods. Conservation Biology Lessons and Policy Implications’. (Eds E. Katsigris, J. Xu and 23, 520–530. doi:10.1111/j.1523-1739.2009.01237.x T. A. White.) pp. 303–40. (Beijing Forestry Publishing House: Beijing, Yan, J. P. (2001). ‘Strategies and Countermeasures of China’s Great China.) Western Development Strategy.’ (Science Press: Beijing, China.) [in Shi, Y. Z. (1983). On the influences of rangeland vegetation on the density Chinese] of plateau pika (Ochotona curzoniae). Acta Theriologica Sinica 3, Yan, Z. L., Wu, N., Dorji, Y., and Ru, J. (2005). A review of rangeland 181–187. [in Chinese] privatization and its implications in the Tibetan Plateau, China. Smith, A. T., and Foggin, J. M. (1999). The plateau pika (Ochotona Nomadic Peoples 9,31–51. doi:10.3167/082279405781826155 curzoniae) is a keystone species for biodiversity on the Tibetan Plateau. Yeh, E. T. (2005). Green governmentality and pastoralism in western Animal Conservation 2, 235–240. doi:10.1111/j.1469-1795.1999. China: ‘Converting pastures to grasslands’. Nomadic Peoples 9,9–30. tb00069.x doi:10.3167/082279405781826164 Smith, A. T., Formozov, N. A., Hoffmann, R. S., Zheng, C. L., and Erbajeva, Yeh, E. T. (2009). Greening western China: a critical view. Geoforum 40, M. A. (1990). The pikas. In: ‘Rabbits, Hares, and Pikas: Status Survey 884 –894. doi:10.1016/j.geoforum.2009.06.004 and Conservation Action Plan’. (Eds J. A. Chapman and J. E. C. Flux.) Yeh, E. T., and Gaerrang, (2011). Tibetan pastoralism in neo-liberalising pp. 14–60. (IUCN: Gland, Switzerland.) China: continuity and change in Gouli. Area 43, 165–172. doi:10.1111/ Smith, A. T., Zahler, P., and Hinds, L. A. (2006). Ineffective and j.1475-4762.2010.00976.x unsustainable poisoning of native small mammals in temperate Asia: Yonten Nyima, and Yeh, E. (in press). Environmental issues and conflict a classic case of the science-policy divide. In: ‘Conservation Biology in in Tibet. In: ‘Ethnic Conflict in Western China’. (Eds B. Hillman and Asia’. (Eds J. A. McNeely, T. M. McCarthy, A. T. Smith, L. Olsvig- G. Tuttle.) (Columbia University Press: New York.) Whittaker and E. D. Wikramanayake.) pp. 285–293. (Society for Zhang, H. F., Liu, F. G., Zhou, Q., and Duo, H. R. (2004). Degradation Conservation Biology, Asia Section and Resources Himalaya mechanism of the grass in Qinghai Plateau and its prevention and control Foundation: Kathmandu, Nepal.) countermeasures. Ziran Zaihai Xuebao 13, 115–120. [in Chinese] State Council (2002). Some suggestions regarding strengthening Zhou, L., Wang, Q. J., Zhao, J., and Wang, Q. (1995). Studies on optimum grassland protection and construction. State Council Circular 2002, 19 stocking intensity in pasturelands of alpine meadow. I. stocking intensity [Beijing, China]. to maximize production of Tibetan sheep. In: ‘Alpine Meadow Wang,Y.B.,Wang,G.X., Sheng,Y. P.,andWang,W.L. (2005).Degradation Ecosystem 4’. pp. 365–375. (Science Press: Beijing, China.) [in Chinese] of the eco-environmental system in alpine meadow on the Tibetan Zhou, H. K., Zhou, L., Zhao, X. Q., Liu, W., Yan, Z. L., and Shi, Y. (2003). plateau. Journal of Glaciology and Geocyrology 27, 634–640. [in Degradation process and integrated treatment of ‘black soil beach’ Chinese] grasslands in the source regions of the Yangtze and Yellow Rivers. Wangdwei, M., Steele, B., and Harris, R. B. (2013). Demographic responses Chinese Journal of Ecology 22,51–55. [in Chinese] of plateau pikas to vegetation cover and land use in the Tibetan Autonomous Region, China. Journal of Mammalogy 94, 1077–1086. doi:10.1644/12-MAMM-A-253.1

www.publish.csiro.au/journals/trj