Rangeland vegetation dynamics in the Altai mountain region of Mongolia, Russia, Kazakhstan and China: effects of climate, topography, and socio-political context for livestock herding practices

Discriminating between climate- and human-induced variation in rangeland quality poses a major challenge for developing policy to sustain herder livelihoods and alleviate herder poverty. We contrasted changes in rangeland vegetation cover across a region—the Altai Mountains of central Asia (China, Kazakhstan, Russia and Mongolia)—that juxtaposes strongly contrasting social, political and economic conditions across a community of herders of shared cultural background (all of Kazakh origin). Our analysis focused on a satellite-derived vegetation index (Normalized Difference Vegetation Index—NDVI) from the Advanced Very High Resolution Radiometer sensor during the period 1982–2013, which included the breakup of the Soviet Union in 1990 and heralded a transition away from pervasive state control on herding practices in many parts of the region. Grassland cover increased with decreasing elevation and increasing precipitation. Grassland also decreased under increased livestock density but was largely unresponsive to the dramatic changes that occurred in the sociopolitical context for grazing practices. Average NDVI values and duration of growing season were greater after the Soviet Union’s collapse across the region, trends that precipitation and temperature data indicate were most likely driven by a changing climate. We conclude that rangeland policy development to assure sustainability of herder livelihoods in the Altai Mountain region should focus on climate change adaptation measures rather than modifying herders’ grazing practices.


Introduction
Rangelands occupy approximately 50% of the world's land surface area and include grasslands, shrublands, woodlands, savannas, steppes, deserts, and tundra (Allen-Diaz et al 1995, Mannetje 2002). Rangelands provide multiple, often underestimated ecological functions and associated ecosystem services that benefit millions of people primarily via livestock production (Allen-Diaz et al 1995, Campbell et al 2000, Mannetje 2002, Johnson et al 2006, Endicott 2012). Yet, rangelands worldwide are subject to widespread degradation (Bedunah and Angerer 2012, Mansour et al 2012) with increasing human populations and livestock densities associated with overgrazing, conversion to croplands, improper land management practices, and climate change-associated desertification among the main causes (Han et al 2008, Harris 2010. Climate is a major determinant of rangeland vegetation dynamics (Propastin et al 2008, Kariyeva and Van Leeuwen 2011, Dubovyk et al 2016. Precipitation patterns and temperature gradients are frequently identified as the primary drivers of inter-annual variability both on a seasonal basis and on longer time frames associated with climate cycles (Propastin et al 2008, Kariyeva and Van Leeuwen 2011, Dubovyk et al 2016. Other abiotic factors such as topography, soil, and surficial hydrological characteristics also have a profound effect on grassland productivity, although unlike climate their effects are relatively constant within a time scale of a few decades (Paudel and Andersen 2010).
Different forms of land tenure and herding practices are also known to be important drivers of global land-cover changes, but their impact on pastureland sustainability are poorly known (Bedunah et al 2006, Sankey et al 2009, Squires 2009, Endicott 2012, Prishchepov et al 2012. A lack of understanding of the relative contributions of abiotic, climatic and human activities on grassland dynamics is a major impediment to decision-makers seeking to develop policy to sustain herder livelihoods and alleviate herder poverty The Altai Mountains provide a useful geographic focus to evaluate questions of interacting effects of topography, climate, and socio-political systems on rangeland dynamics. The region occurs at the border zone of Russia, Mongolia, China, and Kazakhstan and covers a vast area of mostly grassland ecosystems where grazing remains the dominant form of land use (Bedunah et al 2006, Endicott 2012, Benson and Svanberg 2016. Much of the region except in China has seen rapid transformation of the social, political, and economic system associated with collapse of the Soviet Union followed by abrupt but divergent transitions from state-command to market-driven economies in a region that is otherwise comparable ecologically and culturally in terms of herder societies (Fernandez-Gimenez 2011, Endicott 2012. Together these strong and simultaneous socioeconomic changes among four countries have created a long-term, large-scale 'quasi-experiment' to examine the role of social, economic and political systems in concert with environmental factors in driving the dynamics of rangelands. To this end, we contrasted rangeland dynamics from 1982 to 2013 (a period that bracketed the collapse of the Soviet Union in 1991) among the adjacent China, Russia, Mongolia and Kazakhstan portions of the Altai Mountains to examine how environmental factors have influenced dynamics of rangeland vegetation cover. Our specific objectives were to: (1) evaluate changes in vegetation cover over three decades; (2) examine how environmental factors (climate and topography) have influenced dynamics of rangeland vegetation cover; (3) identify the role, if any, of country-specific social and political factors in mediating these changes; and (4) inform development of policy to sustain herder livelihoods and alleviate herder poverty in the broader region, where policies for sustainable rangeland management are largely non-existent yet where herdingbased livelihoods predominate.

Study area
The study area occupied nearly 83 000 km 2 of the Altai mountain region (figure 1). Located in the middle of Eurasia, the region is characterized by a continental climate with short and cool summer (average July temperatures do not exceed +15°C) and extremely cold winters (average temperatures −15°to −35°C, minimum to −60°C) (Kokorin et al 2001, Batima 2006, Kokorin 2011. The four countries that comprise this region-Russia, China, Kazakhstan and Mongolia -provide striking contrasts over time and space in systems governing livestock grazing. The characteristic feature of traditional grazing management system in the region has been seasonal vertical-horizontal movement between pastures (Baylagasov 2011, Endicott 2012). During the Soviet era, transhumant grazing systems were curtailed and rangeland management became more centrally controlled with provision of heavy subsidies such as veterinary care, winter shelters for livestock, hay mowing equipment, hydraulic wells, state-managed delivery of emergency fodder, and improved transportation (Bedunah et al 2006, Fernandez-Gimenez 2006, Endicott 2012, Benson and Svanberg 2016, Mirzabaev et al 2016, Eddy et al 2017. After the collapse of the Soviet Union centralized planning systems in many but not all parts of the region were dismantled and government subsidies disappeared (Bedunah et al 2006, Fernandez-Gimenez 2006, Endicott 2012. Russia and Kazakhstan were constituent republics of the Soviet Union, whereas Mongolia was never a fully integrated part of the Soviet Union (Lattimore 1956, Sharma 2016. Therefore, subsequent transitions of herding systems from statecommand to market-driven varied among the three former Soviet States in the region. Livestock numbers (especially goats) increased in Mongolia (figure 2) due to high demand for cashmere wool on Chinese market . In this quasi-experimental assessment, China was used as a 'control' because the communist regime in China, distinct from the Soviet Union, did not collapse as rapidly and state-controlled rangeland management has persisted to the present.   Statistical data from the regional department of agricultural and agroindustrial management (1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991), China, 11. Statistical data from the regional department of agroindustry and animal husbandry , China). Livestock data were assembled by administrative unit within country (see figure 1) for Mongolia at the somon level (i.e. district, second level administrative subdivision) (Bugat, Nogoonnuur, Tsagaannuur, Tsengel, Ulaankhus), at the kolhoz level (a form of a collective farm) for Russia (Dzhazator settlement, Kuray and Chagan-Uzun settlements, Beltir settlement, Mukhor-Tarkhata settlement, Ortolyk settlement, Telengit-Sortogoy settlement, Tobeler settlement, Kokory settlement, Kazakh and Tashanta settlements), at the county level for China (Altay, Burchin, Burultoqay, Qaba), and for Katon-Karagay district in Kazakhstan. Year was included to account for interaction between rangeland dynamics and the transition of political systems associated with the fall of the Soviet Union (Bennington and Thayne 1994, Prishchepov et al 2012). Lastly, country was included as the proxy for socio-political context for herder livelihoods on the basis of the following predictions: (1) NDVI would increase in Russian and Kazakhstan after the collapse of the Soviet Union due to reduction in livestock numbers and elimination of pastoral nomadic herding; (2) NDVI would decrease in Mongolia because of the more intensive use of rangelands and an increasing number of livestock, and (3) NDVI would not change in China given the stability of the social-political system there or would possibly increase due to evident rises livestock numbers over the study period (figure 2, supplementary materials S1 is available online at stacks.iop.org/ erl/14/104017/mmedia).
To assess the role of livestock densities in rangeland dynamics, livestock data were assembled by adminstraive unit within each country (see figure 2 for units and data sources). Total livestock for all areas was converted to sheep units (SU) per km 2 based on the standard body weight: 1 sheep=1 SU, 1 goat=1 SU, 1 horse=7 SU, 1 cattle/yak=6 SU, 1 camel=5 SU (Kawamura et al 2005b, Lise et al 2006, Wang et al 2007, Paudel and Andersen 2010, Bhatt et al 2013. Kazakhstan was omitted from this analysis because livestock data was only available on large-scale, district level.

Data analysis
We developed linear mixed models to analyze NDVI time-series as a function of environmental and sociopolitical factors as well as the relationship between the change in NDVI values and livestock numbers. We also assessed changes in growing season phenology by contrasting Julian dates of the beginning and end of each growing season by country over the study period.

The role of environmental and social-political on rangeland dynamics
We used linear mixed models (without intercept term) with a Gaussian distribution in R ('lmer' function, 'lme4' package) (Bates et al 2015) to partition variation in NDVI in the region as a function of elevation, precipitation, temperature, interaction of country and year, and with pixel ID as a random factor (i.e. random intercept), with all other variables treated as fixed effects. Continuous variables were standardized to facilitate interpretation of parameters estimates (Schielzeth 2010). A threshold-based filter was applied to check for multicollinearity and predictors with variance inflation factor (VIF) of 5 or higher were considered problematic and were omitted from the analysis

Growing season phenology
To investigate the changes in the start, end and duration of the growing season we interpolated NDVI time series to 1 d temporal resolution using a cubic smoothing spline (Tang and Oki 2007, Cong et al 2012, Lin et al 2013 in R ('spline' function, 'stats' package) (R Core Team 2016). For each pixel we identified the beginning of the growing season defined as the date when the rate of change in NDVI between two consecutive (NDVI t+1 -NDVI t ) days reached its first local maximum value after January 30th, and the end of the growing season as the date when the rate of change in NDVI between two consecutive (NDVI t+1 -NDVI t ) days reached its first local minimum value after August 31st (Ding et al 2015). We then averaged values for the start, end and length of the growing season for Soviet Union era (1985)(1986)(1987)(1988)(1989)(1990)(1991) and post-Soviet times (1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013). Data from 1982 to 1984 and 1992 to 1994 was removed due to the stratospheric aerosol effect of El Chichon (1982) and Mt. Pinatubo (1991) volcanic eruptions (Myneni et al 1997. Paired t-tests using pixel as the replicate were used to contrast start, end and the length of the growing season between the two eras within each country.

The role of social-political and environmental factors in the dynamics of rangelands' vegetation cover
Analysis of the 32 year long (1982-2013) GIMMS NDVI3g data series using linear mixed models revealed that the full model (including elevation, precipitation, country, and year) was by far the topranked model (AICc weights>0.99, table 2, S1) (temperature was omitted from the models due to multi-collinearity, with VIF=13.02). The marginal R 2 (variance explained by fixed factors) for the full model was 0.51 and the conditional R 2 (variance explained by both fixed and random factors) was 0.97. There was no detectable spatial autocorrelation in model residuals at any lag distance. The full model indicated that, on average, NDVI was highest in Kazakhstan and lowest in Mongolia (table 2) with a drop in mean and maximum NDVI values in all four countries the year following the collapse of Soviet Union in 1991 (on average NDVI declined from 1991 to 1992 by 10.0% in Russian, 7.7% in Kazakhstan, 7.5% in Chinese, and 7.1% in Mongolian sectors of the study area) followed by an extended period of stability to 2008 in all countries and then increase in NDVI to the present (figure 3). In terms of biophysical variables, NDVI increased with elevation up to 1500 m but thereafter declined. NDVI increased with increasing precipitation during the growing season, rising by 5.58 × 10 -4 for each 1 mm increase in precipitation (table 2).

The role of livestock in rangeland dynamics
Increased precipitation and temperature was associated with higher NDVI values, whereas an increase in livestock numbers over the previous year was associated with decrease in NDVI values in the current year (table 3(A)). Our models revealed no apparent relationship between increase or decrease in NDVI the previous year and the number of animals on the pasture during the current year (table 3(B)), that is, herders did not appear to be tracking the previous year's vegetation cover in decision-making on livestock stocking in the current year.

Dynamics of growing season phenology
Differences in start, end and the length of the growing season during the Soviet Union era and post-Soviet era indicated that the growing season startearlier during the post-Soviet era in Mongolia, China, and Kazakhstan (with no difference in Russia) while ending later and lasting longer during the post-Soviet era in all four countries (table 4). Trends in averaged May-September NDVI values by country during 1985-1991, 1992-1994 and 1995-2013 indicated a positive significant effect of year and divergent effects of country, but no significant interaction between country and year (supplementary materials S3).

Discussion
Our analysis of changes in vegetation cover in the Altai Mountains in central Asia based on AVHRR NDVI time series for 1982-2013 revealed a long-term positive trend in the NDVI time series in all four countries in the study area albeit interrupted by a short episode of decline in mean and maximum NDVI values immediately following the collapse of Soviet Union in 1991. Abiotic factors (precipitation, temperature, topography) were the primary drivers of NDVI dynamics and dominated any potential effects of social, political and economic drivers of herding practices in this region. These outcomes contradicted our expectation of sharp and divergent transitions in rangeland quality and dynamics after the Soviet Union collapse amongst the four political units that comprise the region given their distinct trajectories since the end of the Soviet era. Climate clearly played a dominant role in rangeland dynamics in the Altai Mountain region, with average NDVI values higher after the Soviet Union collapse in all four countries, an effect seemingly driven by climate change. Central Asia is exhibiting one of the strongest warming signals on the planet and climate scenarios project a continued increase in temperature (Yu et al 2003, Lioubimtseva andHenebry 2009). Mongolia is a case in point. The Mongolia study area segment was at the upper end of the elevational gradient for the high elevation grassland ecosystems in the Altai region, and herders there dramatically increased the number of livestock during the study period coincident with a strong increase in NDVI values. Coupled with this increase in temperature and precipitation, we observed an expanding growing season, likely contributing to the increase observed in NDVI in this region of perennial grasslands. Our findings are consistent with other studies that suggest increase in growing season NDVI in this region driven by warmer temperatures (Myneni et al 1997, Zhou et al 2001, Propastin et al 2008, Dubovyk et al 2016. It is important to note that the sharp decline observed in NDVI values throughout all four countries from 1991 to 1992 following the collapse of the Soviet Union coincided with a short-term global-scale change in climate associated with the Mt. Pinatubo eruption. Immediate, short-term impacts of sociopolitical change on rangeland dynamics during this transitional period in this region (figure 3) may therefore have been masked by Mt Pinatubo-induced climate disruption whereas over the longer-term, environmental factors were the primary drivers in NDVI dynamics in the region ( Independent of the influence of climate, the strong upward tendency in NDVI forces reconciliation of the notion that increasing herd sizes occurring across much (but not all) of the region degrade rangeland. In both the countries (Russia and Mongolia) where relationships between NDVI and livestock were measured, NDVI has increased strongly coincident with increasing herd sizes over the last decade. Traditional theory conceptualizes rangelands as equilibrium systems regulated by animal density-dependent feedback, whereas an alternative view is of rangelands as nonequilibrium systems where abiotic factors are limiting (Ellis andSwift 1988, Fernandez-Gimenez andAllen-Diaz 1999). More recent studies suggest that grassland ecosystems can exhibit characteristics of both equilibrium and non-equilibrium systems at the same time (Fernandez-Gimenez and Allen-Diaz 1999). Our study suggests that the Altai Mountain region is governed mostly by abiotic factors with minor (modification by livestock grazing pressures and herding practices. Similarly, Hilker et al (2014) demonstrated that for the north and north east of Mongolia precipitation is one of the main drivers in NDVI dynamics although in Central and Southern Mongolia precipitation contributed less than 10% to NDVI dynamics and a negative trend in inter-annual NDVI values attributed over the last 10 years to an increase in livestock numbers. Annual precipitation also correlates positively with Leaf Area Index (LAI) with an apparent increasing trend of LAI in Chinese part of Altai Mountain region in 1981-2011(Fang et al 2013. An analysis of NDVI residual trends not explained by precipitation indicated elevated livestock density in Mongolia could depress NDVI (John et al 2016).
In the context of our study, whereas the Soviet era changed grazing philosophy for a few decades, after the collapse of the Soviet system when regional markets were lost due to high transportation cost, localized markets reemerged along with the traditional knowledge of maintaining rangeland health and resiliency (Retzer and Reudenbach 2005, Fernandez-Gimenez 2006, Johnson et al 2006, Endicott 2012. Many herders have returned to former, more transhumant herding approaches. The expanded movement among pastures among contemporary herders may have allowed vegetation to recover during the season leading to apparently increasing grassland productivity coinciding with increasing total herd biomass.

Conclusion
Much of central Asia where livestock herding represents the dominant livelihood lacks any coherent grazing policy to sustain herder livelihoods and alleviate herder poverty, with China being a notable exception (Fernandez-Gimenez 2000, 2011, Endicott 2012. Our study calls into question whether modifying grazing practices should be the primary focus of decision-makers for developing sustainable grassland use in this region. Climate is clearly changing in the region (Yu et al 2003, Lioubimtseva and Henebry 2009, Kokorin 2011. The revitalization of herding communities and former herding practices in some regions may have enabled a simultaneous increase in both livestock numbers and grassland cover, for example, in western Mongolia (figure 3) where traditional intraand inter-community relationships have reassembled following the collapse of the Soviet Union (Fernandez-Gimenez 2011, 2000, Endicott 2012. Traditional herding practices may ultimately be the most important buffer for herder livelihoods in the region as the climate continues to change and extreme weather events (especially dzuds) are expected to become more common (Fernandez-Gimenez 2000, 2011. In contrast, China is still attempting to settle herding communities based on the policy framework that urban life is superior to rural life, a policy that has led to the limitation and elimination of herder mobility and to pasture degradation (Bedunah et al 2006, Endicott 2012, Benson and Svanberg 2016. Our analysis of a continuous and field-validated (Paltsyn et al 2017) index of vegetation cover spanning a 32 year long period implies that assuring sustainability of herder livelihoods and habitat quality for rangeland-associated wild species in this globally significant ecoregion should focus more on climate change adaptation (mostly focusing on developing decision-making frameworks for responding to precipitation and vegetation growth variability, altering strategies of livestock management, modifying household financial capital, and monitoring the status of pasture degradation, Wang et al 2013) than modifying local grazing systems, of which traditional grazing practices are apparently more closely associated with rangeland resilience.

Acknowledgments
This project was supported by NASA's Land Cover Land Use Change Program (grant #NNX15AD42G).