Grassland conservation programs, vegetation rehabilitation and spatial dependency in Inner Mongolia, China

Land Use Policy 64 (2017) 429–439

Contents lists available at ScienceDirect

Land Use Policy journal homepage: www.elsevier.com/locate/landusepol

Grassland conservation programs, vegetation rehabilitation and spatial dependency in Inner Mongolia, China Haibin Chen a,b , Liqun Shao a , Minjuan Zhao a,∗ , Xing Zhang a , Daojun Zhang a a b

College of Economics and Management, Northwest A&F University, Yangling 712100, China College of Resources Science and Technology, Beijing Normal University, Beijing 100875, China

a r t i c l e

i n f o

Article history: Received 26 November 2016 Received in revised form 14 March 2017 Accepted 14 March 2017 Keywords: Grassland conservation Spatial dependency NDVI Spatial panel model Inner Mongolia

a b s t r a c t Three nationwide grassland conservation programs have been implemented in Inner Mongolia since 2000. All aim to relieve grazing pressure, hence to reverse the grassland degradation trend. The different timings and spatial configurations of these programs present an unusual setting of quasi-natural policy experiment for exploring their effectiveness and interactions with other drivers on a regional scale. In this paper, a spatial panel model was developed to examine the effects of the programs on vegetation rehabilitation, meanwhile to detect the spatial interdependent relationship among grassland management units occurring in the process of program implementation. The methodology used a panel dataset of SPOT-VEGETATION NDVI data, multi-station surface meteorological observations, and socio-economic statistics across 88 counties from 2000 to 2013. The modeling results suggested that these programs in general significantly facilitated grassland vegetation rehabilitation. Enrollment in the Beijing–Tianjin Wind/Sand Source Control Program and in the Grazing Withdrawal Program was predicted to increase the normalized difference vegetation index (NDVI) value by an amount equivalent to the effects of 136 mm and 56 mm additional annual precipitation, respectively. The positive and significant coefficient of spatial lag term indicated that there was a synergistic relationship in the vegetation variations of neighboring counties, and a unit increase in the weighted sum of all neighboring counties’ NDVI values could approximately increase a target county’s NDVI value by 0.2, after controlling for other factors’ effects. Certain spatial spillover mechanisms may function to generate this effect, such as benign competition, mutual cooperation and coordination, or sharing of successful experiences among neighboring counties in carrying out the programs. Nevertheless, the actual mechanisms need to be confirmed by field surveys in future studies. © 2017 Elsevier Ltd. All rights reserved.

1. Introduction Grassland degradation is of critical concern worldwide because grasslands occupy nearly 40% of the Earth’s land surface and support the livelihoods of more than 1 billion people, mostly in the developing world (Millennium Ecosystem Assessment, 2005). The term “grassland degradation” is often referred to as a process in which grassland productivity decreases and ecosystem conditions deteriorate, including fragmentation of grass coverage, reduction in soil fertility, soil compaction, declines in the percentage of high quality forages, and encroachments of deserts (Feng et al., 2009; Li et al., 2013). Adopting grazing pressure relief as the dominant strategy, many conservation initiatives have been established around

∗ Corresponding author. Tel.: +86 29 8708 1398; fax: +86 29 8708 1209. E-mail address: [email protected] (M. Zhao). http://dx.doi.org/10.1016/j.landusepol.2017.03.018 0264-8377/© 2017 Elsevier Ltd. All rights reserved.

the world, aiming to reverse the degradation trend, facilitate the sustainable development of vast pastoral regions and enhance their adaptive capacity to climate change (Van Andel and Aronson, 2012). However, few attempts have been made to systematically evaluate the effectiveness of these initiatives (Pullin and Bajomi, 2008; Birch et al., 2010). This study intends to fill this knowledge gap by evaluating the effects of three government-funded conservation programs implemented in Inner Mongolia since 2000 on the vegetation cover condition, thereby examining the effectiveness of grazing pressure relief and its interactions with other drivers. Grassland degradation is especially serious in Inner Mongolia, which has 78 million ha grassland, accounting for 21.7% of China’s total grassland area, and is recognized as the largest of the country’s five major pastoral regions (Ellis, 1992). As of 2000, approximately 90% of Inner Mongolia’s natural grassland was degraded to some extent (Jiang et al., 2006; Akiyama and Kawamura, 2007). Acknowledged as an important ecological network providing a variety of

430

H. Chen et al. / Land Use Policy 64 (2017) 429–439

ecosystem services for northern China, such as soil and water conservation, carbon sequestration, repository of genetic resources etc. (Xu et al., 2009; Wu et al., 2015), grassland degradation of this magnitude can be responsible for several regional, and even global, environmental problems (Sheng et al., 2000; Yang et al., 2010), of which the most prominent is the frequent occurrence of severe sand storms and dust storms across northern China in recent decades, especially around Beijing and adjacent regions (Ye et al., 2000; Han, 2004; Tan and Li, 2015). Moreover, such degradation can also directly affect the livelihood of millions of people, especially indigenous ethnic Mongolians who have lived in the region for generations (Waldron et al., 2010). Three nationwide major grassland conservation programs have been carried out in this region since 2000, namely the Beijing–Tianjin Wind/Sand Source Control Program (hereafter, BTWSSC) since 2001, the Grazing Withdrawal Program (hereafter, GW) since 2003, and the Ecological Subsidy and Award System (hereafter ESAS) since 2011. The objectives, durations, ranges and major measures taken of each program were summarized in Section 2.2. Nevertheless, the basic logic underlying the three programs was that overgrazing is the fundamental cause of grassland degradation (Li and Li, 2016). Therefore, the major counter-measures to restore the grassland vegetation condition are grazing pressure relief by year-round or seasonal grazing cessation, rotational grazing and achieving a forage-livestock balance (Miao et al., 2015). The different timings and spatial configurations of the programs present an unusual setting of quasi-natural policy experiment. It is very interesting and promising to explore the impacts of those programs and other drivers, which is what we intend to accomplish in this study. Controversy exists in previous studies about the ecological effects of the programs. Many scholars suggest that grassland conservation programs have had positive impacts on restoring grassland vegetation (Liu et al., 2003; Xing et al., 2005; Shi et al., 2009; Bao and Zhang, 2015); yet, others argue that these programs have generated negative impacts on the larger scale ecosystem (Wang, 2009; Wang and Qiao, 2011; Li and Li, 2016). Researchers who are skeptical about the three conservation programs point out that grazing pressures are shifted to non-program areas (via leakage), thus increasing degradation in non-program regions, and that illegal grazing activities commonly occur in program areas (Wang, 2009; Wang and Qiao, 2011; Li and Li, 2016;). In addition, some scientists believe that long-term grazing exclusion is harmful to vegetation regeneration and is unsustainable (Xue et al., 2010; Gu and Li, 2013; Li and Li, 2016). According to the grazing optimization hypothesis, an appropriate level of grazing intensity is needed to sustain grasslands and can contribute to enhanced biodiversity and primary productivity (McNaughton, 1979). The impact of grazing intensity control on grassland vegetation change still needs to be verified. Most of the previous studies on the effectiveness of the conservation programs in Inner Mongolia used small-scale data generated from field experiments and household surveys (Li and Li, 2015). The small scale of these studies impeded the discovery of largescale effects of the programs, and results could not be expanded to regional levels due to spatial heterogeneity. Although some research used large-scale data generated from remote sensing, the time spans of such studies were relatively short (usually 1–2 years); thus, it was difficult to find the long-term effects of the conservation programs, and specifically to find the long-term effects of grazing exclusion. Li et al. (2012) used long-term panel data to empirically analyze the natural and man-made causes to grassland degradation in Xilingol League over the last two decades. However, both fixed-effects and random-effects models of panel data imply an assumption of spatial independence; ignoring the spatial depen-

dency among neighbors could lead to biased estimation (Anselin, 2013). Spatial externalities play a central role in the recent emergence of “spatial thinking” in mainstream social sciences (Goodchild et al., 2000), driven by a rising awareness that spatial interdependency (or spatial correlation) broadly and inherently exists in social phenomena, such as interdependent behaviors of economic agents, social–economic agglomeration externalities as well as spatial knowledge spillovers. Spatial panel models were developed to take spatial dependency into account by introducing spatial lag error terms or spatial lag dependent variables into normal panel models (Anselin, 2013). These models allow cross-sectional dependence as well as state dependence, and can also enable researchers to control for unknown heterogeneity (Lee and Yu, 2010). Hence, spatial panel data models have a wide range of applications in fields such as agricultural economics (Druska and Horrace, 2004), transportation (Frazier and Kockelman, 2005), public economics (Egger et al., 2005), and demand for goods (Baltagi and Li, 2006). However, the application of such models in land-change science has been relatively slow and almost nonexistent in grassland conservation policy appraisals. Considering the limitations in previous studies, the goal of this paper is to assess the effects of government conservation programs and other factors on the grassland vegetation cover condition at a large, regional scale, meanwhile to detect the spatial interdependent relationship among grassland management units occurring in the process of program implementation. Specifically, we will investigate the program effects and spatial dependency by building a panel dataset and employing a novel spatial panel model. The panel dataset will consist of the SPOT-VEGETATION NDVI data product, multi-station surface meteorological observations, socio-economic statistical data, as well as the implementation duration and range of major grassland conservation programs across all counties in Inner Mongolia from 2000 to 2013. The spatial panel model pooled with spatial autocorrelations and temporal fixed effects will disentangle the potential spatial interactions embedded in grassland management process and quantify the effects of such factors as climate and demographic change, economic development, and agricultural expansion on the grassland vegetation condition. It is hoped that, with the empirical data and spatial modeling method, this study will improve our knowledge of the effectiveness and spatial spillover effects of government conservation policy, which may help to design and implement policy more effectively, ultimately, to achieve better resource conditions. The paper is organized as follows. The methodological framework was described in Section 2. Empirical results and detailed discussions were then presented in Sections 3 and 4, respectively. Finally, the implications for policy making and future studies were summarized in Section 5.

2. Material and methods 2.1. Study area Inner Mongolia is located in the north of China (97–126◦ E, 37–53◦ N), and covers a total land area of 1.18 million km2 , most of which belongs to the Mongolian Plateau, with an average altitude of 1000 m (Fig. 1). Geomorphology in the region varies greatly, including several plateaus (Alxa, Bayan Nur, Hulun Buir, Xilingol, and Ulan Qab), mountains (Great Khingan, Helan and Yinshan), plains (Hetao, Liaohe, Nenjiang, and Tumuochuan), and also valleys and basins. There are five major deserts (Badaim Jaran, Bayan Ondor, Qubqi, Tengger, and Ulan Buh) and five major sandlands (Hulun Beir, Hunshandak, Khorchin, Mu Us, and Ujumqin) in this region.

H. Chen et al. / Land Use Policy 64 (2017) 429–439

431

Fig. 1. Geographic location of Inner Mongolia and its administrative divisions.

million metric tons

35

Grain Output

Inner Mongolia has a characteristically temperate continental climate, with a large daily temperature difference in the summer and freezing temperatures in winter. Annual mean temperature is 0–8 ◦ C, and annual precipitation ranges between 50 mm and 500 mm, varying greatly among years and demonstrating a declining gradient from east to west. Annual evaporation ranges from 500 mm to 3000 mm, and sand-dust storms occur 5–20 days per year in most areas. Inner Mongolia is an ethnic Mongolian concentrated region; however, the population of ethnic Han people dominates as a result of the immigration since the 1970s. As of 2012, the total population was approximately 25 million, and per capita gross domestic production was approximately 64,000 Yuan (Bureau of Statistics, Inner Mongolia Autonomous Region, 2013). Inner Mongolia is comprised by 102 counties (banners, districts or municipalities), of which 33 are listed as pastoral counties, 21 are listed as semi-pastoral counties, and the remaining 48 are urban districts or crop-farming dominated counties (Fig. 1). Extensive grazing dominates the agricultural sector of pastoral counties. However, in the semi-pastoral counties, relatively intensive animal husbandry as well as crop farming exceeds the grassland use. Grassland is the dominant landscape component in Inner Mongolia, covering 78.8 million ha and accounting for 74.5% of the total land area of the region, making this region an important livestock husbandry base in China. Grassland in this area can be categorized into three major types from east to west: meadow steppe, typical steppe, and desert steppe. The productivity of meadow steppe is the highest of the three types and can reach 1650 kg ha−1 year−1 . Typical steppe is the predominant form of Inner Mongolia’s grassland, and covers a total area of 27.67 million ha. Desert steppe is located in the transition zone from grassland to desert, and has the lowest productivity of the three types. Grassland animal husbandry is the traditional agricultural industry in Inner Mongolia and is an important livelihood source for

30

y = 1.5088x - 3007.3 R² = 0.9595

25 20 15 10 5 0 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Year

Fig. 2. Total grain output of Inner Mongolia from 2000 to 2013 (Statistical Year Book of Inner Mongolia (2001–2014)).

rural residents. After the implementation of the Household Production Responsibility System, the number of livestock has increased rapidly, for instance, from 2 million sheep units (SU, one horse is 6 SU, one cattle is 5 SU, and one goat is 0.8 SU) in 1977 to 18 million SU in 2000 in Xilingol League (Wang et al., 2005). In addition to animal husbandry, the scale of crop farming has become larger and reclaimed increasing amounts of grasslands in recent decades, as indicated by the increasing grain production (Fig. 2). Approximately 46.7 million ha, or 74% of the useable natural grassland of Inner Mongolia has been degraded to some extent; light-, medium-, and severe-degradation account for 31%, 37%, and 32%, respectively, of the grassland area (Brown et al., 2008). Grassland degradation of this magnitude reduces biodiversity, biomass, carbon reserves, usable lands and the productive capacity of households, and also contributes directly to the frequency and severity of dust and sand storms in northern China.

432

H. Chen et al. / Land Use Policy 64 (2017) 429–439

2.2. Overview of conservation programs The Beijing–Tianjin Wind/Sand Source Control Program aimed to alleviate the severe problem of sand storms and dust storms around Beijing, Tianjin and adjacent regions. This program was implemented in five provinces and covered a total area of 458,000 km2 . In total, 369,000 km2 in Inner Mongolia (31.9% of this region’s total area) was enrolled in the BTWSSC, accounting for 80.6% of the program’s total implemented area. The duration of the first stage of this program was from 2001 to 2010, and the central government invested a total of 41.2 billion Yuan (1 Yuan = 0.15 USD) to financially subsidize grazing cessation and grassland enclosures, pen construction for livestock confinement, human emigration and resettlement from areas where grazing was banned, aerial seeding, and cultivated pastures establishment (The State Council, 2002). The second stage (with a budget of 87.8 billion Yuan) was initiated in 2013 and planned to expire in 2022 (The State Council, 2012). The second program, the Grazing Withdrawal Program, began in 2003 and aimed to reduce overall grazing pressure because decision-makers believed that overgrazing is the fundamental cause of grassland degradation. The major measures in GW include grassland enclosures, year-round grazing bans in ecologically fragile regions and severely degraded grasslands, seasonal grazing pauses within medium- or lightly degraded grasslands, and rotational grazing in regions having relatively benign vegetation conditions. Furthermore, intensive animal husbandry, pen-feeding systems and cultivated pasture establishment, sedentary lifestyle and other livelihood opportunities were encouraged. Forage grains were provided as in-kind subsidies to households according to their registered grassland areas for both year-round grazing bans and seasonal grazing pauses, and cash subsidies were provided for barbed wire fencing (Ministry of Agriculture, 2005). From 2003 to 2011, the central government invested a total of 15.5 billion Yuan in the GW program (Ministry of Agriculture, 2012). Within g, 33 counties (banners) participated in this program, and the subsidy standards were 82.5 kg forage grains ha−1 year−1 for year-round grazing cessation, 20.625 kg ha−1 year−1 for seasonal enclosure, and a one-time cash payment of 50 Yuan ha−1 for barbed wire fencing (Development and Reform Committee, Inner Mongolia Autonomous Region, 2003). By the end of 2011, grazing was banned from 7.23 million ha, seasonal grazing resting was applied on 8.63 million ha, rotational grazing was used on 410,000 ha, and 725,000 ha of cultivated pastures were established (Grassland Supervision and Management Bureau, Inner Mongolia Autonomous Region, 2012). To continue pursuing the above objectives, a follow-up program, the Ecological Subsidy and Award System, was initiated in 2011 and targeted more extensively managed grasslands than did BTWSSC and GW. The ESAS dramatically increased investment in grassland rehabilitation with an annual budget of 13.4 billion Yuan, which was nearly the total expenditure of the previous 8 years in GW. In fact, the actual expenditure by the ESAS program was 14.9 billion Yuan in 2012 (Miao and Liu, 2013), exceeding the budget arrangement. In addition to subsidizing grazing withdrawals (90 Yuan ha−1 year−1 ), the ESAS also rewards sustainable grazing practices that are based on grassland carrying capacity (the so-called “forage-livestock balance”, 22.5 Yuan ha−1 year−1 ), and subsidizes purchasing improved livestock breeds (800 Yuan head−1 year−1 for breeding rams, and 50 Yuan head−1 year−1 for breeding cows in Inner Mongolia). The ESAS also subsidizes improved perennial grass cultivation (1050 Yuan ha−1 for three years in Inner Mongolia) and other comprehensive production materials (500 Yuan household−1 year−1 in Inner Mongolia), and provides funds for vocational education and training of herders. In Inner Mongolia, 56.6 million ha of grasslands were registered in the ESAS program; including 18.6 million ha banned from grazing and 38 million ha

Fig. 3. Participation in three major grassland conservation programs (Beijing–Tianjin Wind/Sand Source Control Program, BTWSSC; the Grazing Withdrawal Program, GW; and the Ecological Subsidy and Award System, ESAS) by counties in Inner Mongolia.

managed according to the forage-livestock balance. In addition, 6 million ha of unexpired GW program grassland were transferred into the ESAS program (Department of Agriculture and Animal Husbandry, Inner Mongolia Autonomous Region, 2011). In summary, 31 counties from four leagues (cities) participated in the first stage of the BTWSSC program since 2001, 33 counties from six leagues (cities) enrolled in the GW program since 2003, and all 54 pastoral or semi-pastoral counties enrolled in the ESAS program. In total, 31 counties enrolled in both the GW and ESAS programs and 20 counties enrolled in both the BTWSSC and ESAS programs. Some counties only enrolled in one of the conservation programs (BTWSSC, 11; ESAS, 3; and GW, 2) and 21 counties did not participate in any programs (Fig. 3). 2.3. Research methods Before the spatial econometric models could be built, the spatial autocorrelations, or spatial aggregation degree of NDVI values, of all county units had to be tested. The commonly used test statistics include Moran’s I, Geary’s C and the Getis-Ord Gi*. 2.3.1. Moran’s I Moran’s I (Moran, 1950) uses the geographical locations and attribute values of spatial units simultaneously to measure the spatial autocorrelation. This statistical test was applied in this study. It is calculated as the following equation:

n n Moran’s I =

i=1

j=1

S2



Wij (Yi − Y )(Yj − Y )

n n i=1

 n

j=1

(1)

Wij



 n

where S 2 = 1/n (Y − Y ); Y = 1/n Y ; Yi denotes i=1 i i=1 i the NDVI value of county i; n is the number of counties; and Wij is the spatial weight matrix, reflecting the spatial correlations of counties. The range of Moran’s I values is [−1, 1]. Values closer to −1 denote greater differences or less concentrated distributions; conversely, values closer to 1 denote greater similarities or higher aggregations (aggregations of high attribute values or low attribute values), while values equal to 0 denote random distribution.

H. Chen et al. / Land Use Policy 64 (2017) 429–439

The Monte Carlo simulation was used to test the statistical significance. The Z value was calculated using the following equation: ZI =

I − E[I]



(2)

V [I]

where E[I] = −1/(n − 1) and V [I] = E[I 2 ] − (E[I])2 . 2.3.2. Specifications of the spatial panel model In the current study, a spatial panel model of spatial autocorrelations pooled with temporal fixed effects was set up to estimate the effects of the three major conservation programs and to reveal the spatial spillover effects of grassland conservation investments, using the software written by Elhorst (2014). Fixed effects assumptions are proper in our study, given that the unobserved heterogeneity of all counties in Inner Mongolia are not fully independent of the explanatory variables; in particular, the geomorphological variables are usually believed to be temporally constant but play an important role in shaping surface vegetation (Wooldridge, 2015). A spatial lag model instead of a spatial error model was selected based on Lagrange multiplier (LM) tests for significances of spatial lag effects and spatial error effects (see results in Section 3.3). The vegetation condition is a good indicator of grassland degradation, and the NDVI is widely used to indicate the coverage and productivity of aboveground vegetation (Geerken and Ilaiwi, 2004; Tong et al., 2004; Liu et al., 2013), hence we used it as the dependent variable in the model. The NDVI is defined as ˛ − ˛vis NDVI = nir (3) ˛nir + ˛vis where ˛nir and ˛vis represent surface reflectances averaged over ranges of wavelengths in the near infrared and visible regions of the spectrum, respectively. It is clear from its definition that the NDVI is not an intrinsic physical quantity, but it is indeed correlated with certain physical properties of the vegetation canopy: leaf area index, fractional vegetation cover, vegetation condition, and biomass. Larger positive NDVI values indicate increasing amounts of green vegetation. NDVI values near zero and decreasing negative values indicate lower vegetation coverage or non-vegetated features such as barren surfaces (rock and soil) and water, snow, ice, and clouds. Numerous causes contribute to grassland degradation in Inner Mongolia. However, there is a consensus that over-grazing is the primary cause of the degradation (Waldron et al., 2010), and the major measures taken by the programs is to relieve grazing pressure (Miao et al., 2015). Besides, Rapid population growth (Briske et al., 2015; Hua and Squires, 2015), intensified and improper anthropogenic activities including industrial development and agricultural reclamation (Brown et al., 2008; Waldron et al., 2010), and adverse effects of droughts exacerbated by climate change (Li et al., 2012; Wang et al., 2013) are acknowledged as three major driving forces of grassland degradation. Hence we used grazing intensity, annual precipitation, human population density, crop farming intensity, and industrial activity intensity as independent variables. First, a pooled model without dummy variables of the programs was defined as the following equation: NDVIit

=

n 

Wij NDVIjt + ˇ1 grazeit + ˇ2 precit + ˇ3 popit

j=1

(4)

+ ˇ4 gdpit + ˇ5 farmit + ai + it where i denotes a specific county, j denotes a specific county neighboring county i based on the Queen strategy (see Section 2.3.3) and t denotes the years from 2000 to 2013. Although there are 102 counties in the study area, we grouped neighboring urban districts into

433

a common observation unit while keeping pastoral counties, semipastoral counties and farming counties in their original categories. This process resulted in a total of 88 units (n) in the study. In Eq. (4), Wij is the spatial weight matrix; NDVIit is the mean of all NDVI values for all pixels in county i at year t; grazeit denotes the grazing intensity, which is a county i’s livestock number (head, including both captive and free-ranged livestock) at year t divided by its land area (km2 ); precit is county i’s annual precipitation (mm) at year t; popit denotes the human population density, which is county i’s population (person) at year t divided by its land area (km2 ); gdpit denotes the economic activity intensity, which is a county i’s consumer price index-adjusted gross domestic production (thousand Yuan) at year t divided by its land area (km2 ); farmit denotes the crop farming intensity, which is a county’s grain output (metric tons) divided by its land area (km2 ); and ai is the fixed effect of a county i, which represents all unobserved factors affecting a county i’s NDVI value that do not change over time. These might include certain demographic features of a county’s population (age, race, and education structure), or geomorphological variables. For historical reasons, counties may have very different vegetation status, and historical factors are also effectively captured by the fixed effects constant ai . The variable it is often called the idiosyncratic error or time-varying error, and represents unobserved factors that change over time and affect NDVIit .  and ˇ1 · · ·ˇ5 were the respective coefficients of each independent variables. A fixed effects transformation was applied to eliminate ai (Wooldridge, 2015), yielding a general time-demeaned equation for NDVI (Eq. (5)): NDVI it

=

n 

Wij NDVI jt + ˇ1 graze it + ˇ2 prec it + ˇ3 pop it

j=1

+ ˇ4

gdp

it

+ ˇ5

farm

it

+ 

(5)

it

the solution of which we estimated using the maximum likelihood method. In Eq. (5), NDVI it = NDVIit − NDVIi , and similarly for all explanatory variables and  it . To clarify the effects of individual programs, we revised the pooled model represented by Eqs. (4) and (5) and replaced grazeit with dummy variables representing implementation of the three conservation programs, in consideration of the fact that the major implementation objective of the programs is to control grazing intensity. Thus, the spatial panel model with programs was defined as the following equation:

NDVIit

=

n 

Wij NDVIjt + ˇ1 precit + ˇ2 popit + ˇ3 gdpit + ˇ4 farmit

(6)

j=1

+ ˇ5 BTWSSCit + ˇ6 GWit + ˇ7 ESASit + ai + it

where BTWSSCit , GWit , and ESASit are three dummy variables and assigned to 1 when a specific program was implemented in county i at year t, otherwise 0. The definitions of all variables in the spatial panel model (Eqs. (4) and (6)) are summarized in Table 1.

2.3.3. Generation of spatial weight matrix According to the first law of geography, everything is related to everything else, but near things are more related than distant things (Tobler, 1970). In this study, a spatial weighted matrix was generated using the contiguity-based Queen strategy, in which polygon features that share a boundary or share a node (or both) are counted as neighbors and the spatial weights of neighbors are assigned the value of 1; otherwise the value is 0. The spatial weight matrix was normalized using the program of Elhorst (2014) to yield row sums of unity.

434

H. Chen et al. / Land Use Policy 64 (2017) 429–439

Table 1 Definitions of variables in the pooled model. Variables

Definition

Unit

Expected sign

NDVIit grazeit precit popit gdpit

Mean of all cells’ NDVIa values in county i at year t Livestock number of county i at year t divided by its land area Annual precipitation of county i at year t Human population of county i at year t divided by its land area Consumer price index-adjusted gross domestic production of county i at year t divided by its land area Grain output of county i at year t divided by its land area Dummy variable and assigned to 1 when the BTWSSCb program was implemented in county i at year t, otherwise 0 Dummy variable and assigned to 1 when the GWc program was implemented in county i at year t, otherwise 0 Dummy variable and assigned to 1 when the ESASd program was implemented in county i at year t, otherwise 0

Null Head km−2 mm Persons km−2 Thousand Yuan km−2

Null − + − −

Metric ton km−2 Null

− +

Null

+

Null

+

farmit BTWSSCit GWit ESASit a b c d

Normalized difference vegetation index. Beijing–Tianjin Wind/Sand Source Control. Grazing withdrawal. Ecological subsidy and award system.

2.4. Data acquisition and processing

3. Results

We used 14 years of panel data for all 88 counties of Inner Mongolia from 2000 to 2013, a total of 1232 observations. The county shapefile was retrieved from the Administrative Division Map of China (scale of 1:4000,000; National Geomatics Center of China). The NDVI data were retrieved from SPOT-VEGETATION NDVI data products (Flemish Institute for Technological Research, Vito, Belgium) at a spatial resolution of 1 km and a temporal resolution of 10 days. Because the original data were stored as bytes (0–255), we transformed them into NDVI values (−1, 1) using the band math function in ENVI 4.8 (Harris Geospatial Solutions, Broomfield, Colorado, the U.S.). For our purpose, we only used the NDVI values acquired at the end of August because this is when grassland grows most vigorously. We then calculated the mean of NDVI values of all pixels in each county each year. Annual precipitation data were retrieved from the Chinese National Meteorological Information Center (http://data.cma.cn/). There are 47 meteorological stations in Inner Mongolia, so we used the Kriging method to interpolate the annual precipitation values of all pixels in Inner Mongolia, and then calculated the mean values in each county each year. The data for livestock numbers, human population, gross domestic production, grain output and total land area of each county were retrieved from the Statistical Year Book of Inner Mongolia Autonomous Region (2001–2014). The implementation durations and ranges of the BTWSSC, GW, and ESAS programs were identified by reviewing relevant policy documents from the State Council and Inner Mongolia (Department of Agriculture and Animal Husbandry, Inner Mongolia Autonomous Region, 2011; Development and Reform Committee, Inner Mongolia Autonomous Region, 2003; The State Council, 2002, 2012), and confirmed by personal communications with staff in the Department of Agriculture and Animal Husbandry of Inner Mongolia Autonomous Region.

3.1. Spatial and temporal variations of vegetation coverage

2.5. Data analysis The determination of global Moran’s I values and their significance testing, as well as the generation of the spatial weight matrix, was accomplished using the software Geoda 1.8 (Anselin et al., 2006). The MATLAB R2014a software package (Mathworks Inc., Natick, Massachusetts, the U.S.) was used to estimate the parameters of the spatial panel model. The software ENVI 4.8 was used to interpret and calculate the NDVI differences. ArcGIS 10.2 (ESRI Inc., Redlands, California, the U.S.) was used for spatial interpolation, statistics and mapping.

Spatial and temporal variations of vegetation coverage of the study area were determined based on the remote sensing data interpretation and then the calculation of the NDVI differences. Given the starting points of year and the durations of each program, we presented the vegetation variations in the whole study period (2000–2013) and three phases (2000–2002, 2002–2010, and 2010–2013), which can help to better understand the vegetation dynamics in each phase and to separate the effects of each program. As shown in Fig. 4, following implementation of the first conservation program (BTWSSC) in 2001, the grassland vegetation status across Inner Mongolia showed a general trend of improvement. From 2000 to 2013, 70.38% of grasslands improved and only 29.62% were degraded. The major degradation concentrated in the northeastern part of the study area, mainly in Hulun Buir League, which has a large area dominated by non-pastoral counties and did not enroll in any programs. Scattered degradations were also found in Hinggan, Tongliao, Baotou and Xilingol League (Fig. 4A). In the first phase (2000–2002) of program implementation, 53.29% of study area vegetation was degraded and 46.71% was improved, reflecting that after one years of BTWSSC implementation, the overall vegetation status had not improved and grassland degradation was broadly distributed across the whole study area (Fig. 4B). However, after the GW program was initiated in 2003, the grassland status showed a general trend of improvement. In the second phase (2002–2010), approximately 70% of vegetation coverage improved, although by different degrees (Fig. 4C), indicating that the joint implementation of the BTWSSC and GW programs effectively reversed the degradation trend of Inner Mongolia’s grassland. In the third phase (2010–2013), following the initiation of the ESAS program and its replacement for the GW program in 2011 the grassland vegetation status demonstrated a slight improvement. From 2010 to 2013, 54.75% of study area vegetation improved and 45.25% deteriorated. The major degradation concentrated in Hulun Buir League, and scattered areas of degradation were found throughout the autonomous region (Fig. 4D). In summary, grassland vegetation conditions as a whole generally improved in Inner Mongolia since 2000, following implementation of the first of three conservation programs in 2001. However, because the improvements may have been caused by factors other than the effects of the conservation programs, such as precipitation variations, further examination of the vegetation

H. Chen et al. / Land Use Policy 64 (2017) 429–439

435

Fig. 4. Normalized difference vegetation index (NDVI) differences in Inner Mongolia during various periods: (A) between 2000 and 2013; (B) between 2000 and 2002; (C) between 2002 and 2010; and (D) between 2010 and 2013. 0.87 0.86

y = -0.0036x + 7.9622 R² = 0.377

Global Moran's I

0.85 0.84

that grassland vegetation in Inner Mongolia exhibits significant and positive spatial autocorrelations, making necessary the inclusion of spatial autocorrelations in econometric models that are designed to derive unbiased estimates of relevant factor effects on vegetation coverage variations.

0.83 0.82 0.81

3.3. Lagrange multiplier tests for spatial lag and spatial error effects

0.8 0.79 0.78 0.77 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Year

Fig. 5. Temporal variations of global Moran’s I of normalized difference vegetation index (NDVI) in Inner Mongolia from 2000 to 2013.

To decide whether spatial lag or spatial error dominates, LM tests were conducted on the fixed effects model without a term of spatial autocorrelation (Eq. (6)), and the results are shown in Table 2. Both LM statistics of spatial lag and spatial error were highly significant at the 1% level; however, the robust LM statistic of spatial lag was more significant (p = 0.000) than that of spatial error (p = 0.968). Hence a spatial lag model was selected. NDVIit

responses was needed to identify the precise causes of the vegetation changes.

= ˇ1 precit + ˇ2 popit + ˇ3 gdpit + ˇ4 farmit + ˇ5 BTWSSCit + ˇ6 GWit + ˇ7 ESASit + ai + it

(6)

3.4. Results of spatial panel data models 3.2. Tests for spatial autocorrelations of NDVI The global Moran’s I values of NDVI in Inner Mongolia from 2000 to 2013 were determined to demonstrate the spatial autocorrelation degree of vegetation coverage. Although showing a decreasing trend, all values of global Moran’s I for NDVI from 2000 to 2013 were relatively high, in the range [0.7818, 0.8570], and all values were significant at the 1% level (Fig. 5). These test results indicate

Spatial panel data models without programs and with programs were tested, and their parameter estimates and corresponding statistical significance levels are summarized in Table 3. The goodness of fit of both models (adjusted R2 was 0.2777 and 0.3340 for the “without programs” and “with programs” models, respectively) was higher than that of the non-spatial panel data model (R2 = 0.2704, Table 2), reflecting that the consideration of spatial

436

H. Chen et al. / Land Use Policy 64 (2017) 429–439

Table 2 Results of Lagrange multiplier (LM) tests on variables in the fixed effects model. Dependent variable: NDVI Independent variables

Parameter estimates (levels of significance)

graze

0.000065** (0.040857) 0.000346*** (0.000000) −0.000101 (0.174413) 0.000000 (0.409078) 0.000073 0.184258 0.2704 2031.1 1232 0.000 0.000 0.000 0.968

prec pop gdp farm Adjusted R2 log-likelihood N LM test no spatial lag, probability Robust LM test no spatial lag, probability LM test no spatial error, probability Robust LM test no spatial error, probability ** ***

p < 0.05. p < 0.01.

Table 3 Parameter estimates and their significance levels for two spatial panel data models. Dependent variable: NDVI Independent variables

Without programs

With programs

graze

–

BTWSSC

0.000050* (0.059493) 0.000272*** (0.000000) −0.000066 (0.290186) 0.000000 (0.542982) 0.000064 (0.165745) –

GW

–

ESAS

–

prec pop gdp farm

n i=1

Wij NDVIj 2

Adjusted R Log-likelihood N * ** ***

0.230958*** (0.000000) 0.2777 2239.2593 1232

0.000266*** (0.000000) −0.000099 (0.103644) 0.000000 (0.715956) 0.000084** (0.036945) 0.036173*** (0.000000) 0.014778*** (0.000102) −0.010434* (0.070161) 0.222981*** (0.000000) 0.3340 2271.982 1232

p < 0.10. p < 0.05. p < 0.01.

dependency can better explain the effects of relevant factors on grassland vegetation coverage variations. n Both coefficients of lag dependent variables W NDVIjt in j=1 ij the two models were statistically significant and positive at the 1% significance level, further showing the benefit of including spatial autocorrelations in the models. The parameter estimate of the model without programs was slightly higher than that of the model that included programs, and after controlling for other factors’ impacts, the unit increase in the weighted sum of neighboring counties’ NDVI values was predicted to increase a target county’s NDVI value by 0.23 in the model without programs, and 0.22 in the model that included programs, respectively. The parameter estimates of annual precipitation in both models were statistically significant and positive at the 1% significance level. Accordingly, 100 mm additional annual precipitation is pre-

dicted to increase the NDVI value by 0.027. Comparing with the mean of the NDVI values across the study area (0.3838), this effect is rather small. Given the aridity of the study region, 100 mm higher annual precipitation means a significant improvement in climatic condition, which may result in a remarkable improvement in vegetation growth. This seemly abnormal result may be due to the nonlinear relationship of the NDVI value against real vegetation biomass. The NDVI has been criticized because its sensitivity to surface biomass becomes increasingly weak with increasing biomass beyond a loosely defined threshold (Carlson and Ripley, 1997). The other possibility has to do with the distribution of the dependent variable—NDVI, which has a mean of 0.3838 and a standard deviation of 0.1314. This lack of variability is expected to result in coefficients to be small and less likely to be significant. Nevertheless, the positive result meets our expectation. In the model without programs, the effect of grazing intensity on NDVI was also significant (p < 0.10); however, its positive sign was unexpected. By eliminating the variable of grazing intensity and substituting for it the three dummy variables of program implementations, the model with programs generated more reasonable results for the impacts of the grassland conservation programs. Both coefficients of BTWSSC and GW were statistically significant and positive at the 1% significance level. Keeping other explanatory variables constant, enrollment in the BTWSSC or GW program is predicted to increase a county’s NDVI value by 0.0362 or 0.0148, respectively. Although the ESAS program’s effect on NDVI was weakly significant (p < 0.10), its sign was negative, which was unexpected. This may be attributed to the relatively short implementation period and a lack of data on this program. No effects of any other independent variables were statistically significant except those of the variable farm in the model with programs (p < 0.05). However, because 1 t km−2 higher grain output is predicted to only increase NDVI value by 0.0001 based on the parameter estimation, the effect of the variable farm is small and can be neglected. Nevertheless, this result is out of expectation, which may be attributed to the technological progress in agriculture. Technological progress increases agricultural productivity, hence although the total grain output significantly increased during the study period (Fig. 2), the area of cropland may not necessarily increase equivalently.

4. Discussion 4.1. Effects of major grassland conservation programs on vegetation rehabilitation Vegetation rehabilitation was visually evident in Inner Mongolia grassland since 2000, and the temporal analysis of vegetation changes (as described in Section 3.1 and shown in Fig. 4) confirmed this conclusion. Analyses using spatial panel data models further quantified the contributions of major grassland conservation programs to grassland vegetation improvement. According to the regression results, the implementations of the BTWSSC and GW programs significantly increased the NDVI value in Inner Mongolia by 0.0362 and 0.0148, equivalent to the effects of 136 mm and 56 mm additional annual precipitation, respectively. Given the annual precipitation ranges only between 50 mm and 500 mm and water is the most important limiting factor to vegetation regeneration in this arid region, the positive effects of these two programs were practically significant. Because the ESAS program was initiated in 2011 and only 3 years of data were included in the econometric model, the effect of this program on grassland vegetation variations was unexpected. This program’s long-term ecological effects need to be further monitored. Our research findings were generally consistent with the results of previous studies

H. Chen et al. / Land Use Policy 64 (2017) 429–439

that examined the ecological effects of grassland conservation programs (sometimes called grassland construction programs). In a literature review, Li and Li (2015) found that 72% of studies conducted since 2002 showed that rangeland ecological construction programs led to effective protection and restoration of rangelands; however, most of the reviewed studies were based on local-scale field experiments or household surveys. Our study further contributes to this research field by quantifying the vegetation rehabilitation effects at a regional scale. In addition, due to the control of spatial dependence, the quantitative results of our study are more robust than those from previous studies that did not control the spatial effect. Temporal and spatial analysis of vegetation changes (described in Section 3.1) also suggested that although the grassland vegetation condition was in general improved during the study period, different degrees of degradations were also found across the study area. Degraded areas were especially prominent in the northeastern part of Inner Mongolia where relatively more crop-farming dominated counties and urban districts were located that did not participate in any grassland conservation programs. This result further confirmed the positive role of grassland conservation programs, and suggested that certain negative spatial spillover effects (i.e., leakage) may occur during the programs implementation process. Among the spillover effects, the shift of grazing pressures from program areas to non-program areas should be noted, and counter measures should be implemented to eliminate this negative externality. These measures could include such things as strengthening field monitoring and strictly prohibiting the herd flows from program areas to non-program areas, or gradually bringing non-program counties into the conservation programs. Previous studies were predicated on the belief that higher grazing intensity should lead to grassland degradation in arid Inner Mongolia (Courtois et al., 2004; Akiyama and Kawamura, 2007; Schönbach et al., 2011). However, in the current study according to the model without programs, the parameter estimate of grazing intensity was significantly positive, apparently suggesting paradoxically that grazing intensity improved grassland vegetation. The positive sign may be attributed to the extension of confined animal production (i.e., pen-raising methods) that has been promoted by the government in recent decades. According to the Statistical Yearbook of Inner Mongolia, there were 76.96 million SU reared in a total of 146.23 million m2 pens in 2000, and in 2013 these figures increased by 2 times and 1.95 times to 150.01 million SU and 291.13 million m2 pens, respectively. In contrast, the total number of livestock in Inner Mongolia, including both captive and freeranged livestock, increased by only 1.37 times during the same time period, indicating that the rearing of animals in captivity accounts for a higher percentage of animal husbandry patterns in Inner Mongolia. These more intensive livestock production methods rely more on imported forage as a feed source than do extensive production methods and lessen the pressure on local grassland. Hence, although the apparent grazing intensity increased during the study period, the grassland vegetation condition still recovered. Nevertheless, the inaccuracy in reported statistical measures of livestock numbers may also be responsible for the paradoxical effect of grazing intensity on grassland conditions. If, for example, local governments over-reported livestock numbers to acquire more compensation, published data on livestock numbers would increase instead of decrease, even though the main objective of the conservation programs was to control grazing intensity. Hence, although the main objective of the programs was to control grazing intensity, the reported livestock numbers increased instead of decreased. However, in view of a 4.3 times increase in the value of animal husbandry output in Inner Mongolia from 1999 (188 million Yuan) to 2013 (806 million Yuan) (both values are consumer price index-adjusted using year 1998 as a base), we have confidence

437

that livestock numbers actually did increase during this period, and that the trend of rearing livestock in captivity is the cause for the positive sign on the parameter estimate of grazing intensity. 4.2. Spatial dependency of grassland management units The statistics of the global Moran’s I values indicated that there were significant and positive spatial autocorrelations in grassland vegetation conditions for all counties in Inner Mongolia, and the Lagrange multiplier tests suggested that the spatial lag effect was more robust than that of spatial error effect. Spatial panel data model regressions further quantified the spatial lag effect and suggested that a unit increase in the weighted sum of neighboring counties’ NDVI values could approximately increase a target county’s NDVI value by 0.2, after controlling for other factors’ effects, which is a very large effect in practical term, bearing in mind that the mean of NDVI values of all counties is only 0.3838. This result means that the variation of vegetation conditions in neighboring counties could significantly affect those of a target county. Certain positive spatial spillover mechanisms may function to generate this spatial lag effect. Positive aggregation effects may develop during the process of implementing programs, such as benign competition among neighboring counties to surpass each other, or mutual cooperation and coordination in carrying out the programs, or sharing of successful experiences. It may also be due to the positive covariance of annual precipitation among neighboring counties, which plays an important role in deciding the vegetation conditions in this arid region. Nevertheless, if this spatial lag effect is produced by spatial cooperation or coordination of neighboring counties, it would be of an important guiding significance in improving the design and implementation of government conservation policy. Ignoring the spatial dependency among neighbors could lead to biased parameter estimates; hence, including spatial dependency in econometric models of both cross-sectional and panel data is necessary and can better explain the effects of relevant factors on grassland vegetation coverage, including a lag-dependent variable itself. Furthermore, pooled models of fixed effects and spatial lag can effectively address the temporal dependency of crosssectional units, as well as spatial autocorrelations, while enabling researchers to control for unknown heterogeneity. Thus, the inclusion of spatial autocorrelations in econometric models is a powerful tool for accurately estimating the effects of major programs and revealing and spatial spillover effects of grassland conservation investments. However, given current research methods, this study cannot pinpoint and quantify the precise contributions of all possible mechanisms to this lag effect, other methodologies such as field survey should be used in future studies to clarify the enabling mechanisms and perfect this study. Moreover, it is necessary to notice that although the effect of spatial error on NDVI values was not robust, it was significant at the 1% level, suggesting that the grassland vegetation variation in a target county was also impacted by unobserved time-varying factors in a neighboring county, which were not considered as independent variables in our models. For instance, improved management levels and efficiencies in a neighboring county government may spill over to enhance the efficiency of management in a target county. The animal husbandry production pattern (extensive grazing or intensive pens rearing) has been shown to affect grassland vegetation condition (Li et al., 2015), and the successful experience of introducing the intensive pattern in neighboring counties may be quickly shared with the target county. Institutional arrangements to facilitate a community’s collective management can also play an important role in protecting grassland in this arid region (Li and Li, 2012), and there may also have been spatial knowledge sharing about this institutional innovation. All of these factors are encom-

438

H. Chen et al. / Land Use Policy 64 (2017) 429–439

passed in the idiosyncratic error and may generate spatial spillover effects on neighboring counties.

gains and losses and facilitate achieving a goal of multi-objective synergy.

5. Conclusions

Acknowledgments

Using pooled panel data of vegetation index, surface meteorological observations and socio-economic statistics at county level in Inner Mongolia from 2000 to 2013, we applied spatial econometric models to examine the ecological effects of three major grassland conservation programs implemented in this region since 2000, specifically the BTWSSC, GW and ESAS programs. The results suggested that these programs in general played a significant and positive role in reversing the trend of grassland degradation and accelerating its rehabilitation, therein our study verified the effectiveness of grazing pressure relief. Future grassland ecological policies should follow the template established by these programs, but policy implementers should also take counter measures to eliminate the negative spillover effects of program areas on nonprogram areas. The pattern of rearing livestock in captivity, which relies on imported forage as a feed source and lessens the pressure on local grassland, should be further encouraged. In the meantime, to alleviate the economic burden of purchasing forage, cultivated pasture establishment should be further expanded. The spatial lag effect helps to enhance synergies of conservation efforts among neighboring counties; hence, benign competition and knowledge sharing, which develop during the process of program implementations, should be strengthened. Moreover, comprehensive grassland management capacity also should be enhanced to address the spatial error effects, in terms of improving local government’s management efficiency, promoting intensive animal husbandry, and facilitating self-organized community management. Nevertheless, the underlying mechanisms enabling both spatial lag and spatial error effects have not been clarified by this study, and should be further studied using other methodologies, such as field survey. Spatial econometrics provides a powerful toolset to address spatial interdependency or spatial correlation problems, and has been widely applied in social science research. However, the application of spatial econometrics in grassland conservation policy appraisals is rare. Ignoring the spatial correlations among neighboring grassland management units could lead to biased parameter estimates. This study addressed these pitfalls by using a pooled model of fixed effects and spatial lag effects. This model can better explain the effects of grassland conservation programs on vegetation rehabilitation so as to improve future policy design and implementation. Spatially referenced data and their analyses are special in the sense that the spatial arrangement of the observations provides important information that should not be ignored. Hence, spatial data analysis techniques, in particular spatial econometric modeling or spatial regression modeling, should be further applied in grassland ecological policy appraisals. Three cautions should be noted. First, the non-linear nature and lack of variability in the NDVI value may dampen the potential effects of relevant independent variables, as demonstrated by the “less-than satisfactory” precipitation effect in our modeling results. Furthermore, the NDVI is only a proxy of the coverage and productivity of aboveground vegetation, other ecological indicators should thus be brought in to assist understanding changes in structure and composition, as well as other complex ecological processes in grassland degradation or recovery in future studies. Second, because only three-year effects of the ESAS program were examined in this study, its longer effects need to be further monitored and evaluated. Third, researchers should also examine further the effects of the conservation programs on herders’ livelihoods and community culture. A tradeoff analysis could help compare comprehensive

We thank two anonymous reviewers for their constructive comments on an earlier version of the paper. This work was supported by he MOE (Ministry of Education of China) Humanities and Social Sciences Fund (Grant No. 16YJC630003), National Social Science Major Foundation of China (Grant No. 15ZDA052), the National Natural Science Foundation of China (Grant No. 71373209, and National Social Science Fund of China (Grant No. 15XJY010).

References Akiyama, T., Kawamura, K., 2007. Grassland degradation in China: methods of monitoring, management and restoration. Grassland Sci. 53, 1–17, http://dx. doi.org/10.1111/j.1744-697X.2007.00073.x. Anselin, L., Syabri, I., Kho, Y., 2006. GeoDa: an introduction to spatial data analysis. Geog Anal. 38, 5–22, http://dx.doi.org/10.1111/j.0016-7363.2005.00671.x. Anselin, L., 2013. Spatial Econometrics: Methods and Models. Springer Science & Business Media, Berlin, Heidelberg, Germany. Baltagi, B.H., Li, D., 2006. Prediction in the panel data model with spatial correlation: the case of liquor. Spat. Econ. Anal. 1, 175–185, http://dx.doi.org/ 10.1080/17421770601009817. Bao, Z., Zhang, S., 2015. Analysis on the implementation effects of the policy of compensation to and rewarding for grassland eco-protection—taking Chifeng grassland as an example. J. Anhui Agr. Sci. 43, 328–329, http://dx.doi.org/10. 13989/j.cnki.0517-6611.2015.10.116 (in Chinese). Birch, J.C., Newton, A.C., Alvarez Aquino, C., Cantarello, E., Echeverria, C., Kitzberger, T., Schiappacasse, I., Garavito, N.T., 2010. Cost-effectiveness of dryland forest restoration evaluated by spatial analysis of ecosystem services. Proc. Natl. Acad. Sci. U.S.A. 107, 21925–21930, http://dx.doi.org/10.1073/pnas. 1003369107. Briske, D.D., Zhao, M., Han, G., Xiu, C., Kemp, D.R., Willms, W., Havstad, K., Kang, L., Wang, Z., Wu, J., Han, X., Bai, Y., 2015. Strategies to alleviate poverty and grassland degradation in Inner Mongolia: intensification vs production efficiency of livestock systems. J. Environ. Manage. 152, 177–182, http://dx.doi. org/10.1016/j.jenvman.2014.07.036. Brown, C.G., Waldron, S.A., Longworth, J.W., 2008. Sustainable Development in Western China: Managing People, Livestock and Grasslands in Pastoral Areas. Edward Elgar Publishing Ltd, Cheltenham, UK. Bureau of Statistics, Inner Mongolia Autonomous Region. Statistical Year Book of Inner Mongolia (2001–2014). China Statistics Press. Beijing (in Chinese). Carlson, T.N., Ripley, D.A., 1997. On the relation between NDVI, fractional vegetation cover, and leaf area index. Remote Sens. Environ. 62, 241–252, http://dx.doi.org/10.1016/S0034-4257(97)00104-1. Courtois, D.R., Perryman, B.L., Hussein, H.S., 2004. Vegetation change after 65 years of grazing and grazing exclusion. J. Range Manage. 57, 574–582 http://dx.doi. org/10.2111/1551-5028(2004)057[0574:VCAYOG]2.0.CO;2. Department of Agriculture and Animal Husbandry, Inner Mongolia Autonomous Region, 2011. Implementation Plan for Grassland Ecological Subsidy and Award System. Department of Agriculture and Animal Husbandry, Inner Mongolia Autonomous Region, Hohhot (in Chinese). Department of Agriculture and Animal Husbandry, Inner Mongolia Autonomous Region, 2003. Inner Mongolia Grazing Withdrawal Project Implementation Plan. Department of Agriculture and Animal Husbandry, Inner Mongolia Autonomous Region, Hohhot (in Chinese). Druska, V., Horrace, W.C., 2004. Generalized moments estimation for spatial panel data: Indonesian rice farming. Am. J. Agric. Econ. 86, 185–198, http://dx.doi. org/10.1111/j.0092-5853.2004.00571.x. Egger, P., Pfaffermayr, M., Winner, H., 2005. An unbalanced spatial panel data approach to US state tax competition. Econ. Lett. 88, 329–335, http://dx.doi. org/10.1016/j.econlet.2005.03.002. Elhorst, J.P., 2014. Spatial Panel Data Models, Spatial Econometrics. Springer, Berlin, Heidelberg, Germany. Ellis, J., 1992. Grasslands and Grassland Sciences in Northern China. National Academies, Washington, DC. Feng, Y., Lu, Q., Tokola, T., Liu, H., Wang, X., 2009. Assessment of grassland degradation in Guinan county, Qinghai Province, China, in the past 30 years. Land Degrad. Dev. 20, 55–68, http://dx.doi.org/10.1002/ldr.877. Frazier, C., Kockelman, K., 2005. Spatial econometric models for panel data: incorporating spatial and temporal data. Transp. Res. Rec.: J. Transp. Res. Board 1902, 80–90, http://dx.doi.org/10.3141/1902-10. Geerken, R., Ilaiwi, M., 2004. Assessment of rangeland degradation and development of a strategy for rehabilitation. Remote Sens. Environ. 90, 490–504, http://dx.doi.org/10.1016/j.rse.2004.01.015. Goodchild, M.F., Anselin, L., Appelbaum, R.P., Harthorn, B.H., 2000. Toward spatially integrated social science. Int. Reg. Sci. Rev. 23, 139–159, http://dx.doi. org/10.1177/016001700761012701.

H. Chen et al. / Land Use Policy 64 (2017) 429–439 Grassland Supervision and Management Bureau, Inner Mongolia Autonomous Region, 2012. Annual Monitoring Report on Ecological Effects of Grazing Withdrawal Project (2011). Grassland Supervision and Management Bureau, Inner Mongolia Autonomous Region, Hohhot (in Chinese). Gu, Y., Li, W., 2013. Research on the grazing ban policy’s effect to grassland quality: analysis on the scale of herdsman households. Acta Sci. Nat. Univ. Pekinensis 49, 288–296, http://dx.doi.org/10.13209/j.0479-8023.2013.042 (in Chinese with English abstract). Han, A., 2004. The selection and composing of the vegetation index in the investigation into the source area of sandstorm around Beijing and Tianjin. Remote Sens. Land Resour. 3, 013 (in Chinese with English abstract). Hua, L., Squires, V.R., 2015. Managing China’s pastoral lands: current problems and future prospects. Land Use Policy 43, 129–137, http://dx.doi.org/10.1016/j. landusepol.2014.11.004. Jiang, G., Han, X., Wu, J., 2006. Restoration and management of the Inner Mongolia grassland require a sustainable strategy. Ambio 35, 269, http://dx.doi.org/10. 1579/06-S-158.1. Lee, L., Yu, J., 2010. Some recent developments in spatial panel data models. Reg. Sci. Urban Econ. 40, 255–271, http://dx.doi.org/10.1016/j.regsciurbeco.2009. 09.002. Li, W., Li, Y., 2012. Managing rangeland as a complex system: how government interventions decouple social systems from ecological systems. Ecol. Soc. 17, 9, http://dx.doi.org/10.5751/ES-04531-170109. Li, Y., Li, W., 2015. China’s rangeland management policy debates: what have we learned? Rangeland Ecol. Manage. 68, 305–314, http://dx.doi.org/10.1016/j. rama.2015.05.007. Li, W., Li, Y., 2016. Rangeland Degradation Control in China: A Policy Review, The End of Desertification? Springer, Berlin, Heidelberg, Germany. Li, S., Verburg, P.H., Lv, S., Wu, J., Li, X., 2012. Spatial analysis of the driving factors of grassland degradation under conditions of climate change and intensive use in Inner Mongolia, China. Reg. Environ. Change 12, 461–474, http://dx.doi.org/ 10.1007/s10113-011-0264-3. Li, X., Gao, J., Brierley, G., Qiao, Y., Zhang, J., Yang, Y., 2013. Rangeland degradation on the Qinghai–Tibet plateau: implications for rehabilitation. Land Degrad. Dev. 24, 72–80, http://dx.doi.org/10.1002/ldr.1108. Li, Z., Han, G., Zhao, M., Wang, J., Wang, Z., Kemp, D.R., Michalk, D.L., Wilkes, A., Behrendt, K., Wang, H., 2015. Identifying management strategies to improve sustainability and household income for herders on the desert steppe in Inner Mongolia, China. Agric. Syst. 132, 62–72, http://dx.doi.org/10.1016/j.agsy.2014. 08.011. Liu, A., Xing, Q., Su, R., Wang, J., Guo, G., 2003. Study on effects of grassland bans and rests in Xilingol League. Grassland Prataculture 3, 1–4 (in Chinese). Liu, S., Yan, D.H., Shi, X.L., Wang, G., Yuan, Z., Yin, J., 2013. Grassland NDVI response to climate factors in different vegetation regionalizations in China. Commun. Comput. Inf. Sci. 399, 370–380, http://dx.doi.org/10.1007/978-3-642-41908-9 38. McNaughton, S., 1979. Grazing as an optimization process: grass-ungulate relationships in the Serengeti. Am. Nat., 691–703, http://dx.doi.org/10.1086/ 283426. Miao, D., Liu, Y., 2013. Annual National Grassland Monitoring Report (2012). Chin. Anim. Husbandry 8, 14–29 (in Chinese). Miao, L., Moore, J.C., Zeng, F., Lei, J., Ding, J., He, B., Cui, X., 2015. Footprint of research in desertification management in China. Land Degrad. Dev. 26, 450–457, http://dx.doi.org/10.1002/ldr.2399. Millennium Ecosystem Assessment, 2005. Ecosystems and Human Well-being: Desertification Synthesis. World Resources Institute, Washington, DC. Ministry of Agriculture, 2005. Implementation and Management Regulations on Grazing Withdrawal Project. Ministry of Agriculture, Beijing (in Chinese). Ministry of Agriculture, 2012. Annual National Grassland Monitoring Report (2011). Ministry of Agriculture, Beijing (in Chinese). Moran, P.A., 1950. A test for the serial independence of residuals. Biometrika 37, 178–181, http://dx.doi.org/10.1093/biomet/37.1-2.178. Pullin, A.S., Bajomi, B., 2008. Are we doing more good than harm? Evaluating effectiveness of nature restoration policy in Europe. Sixth European Conference on Ecological Restoration, Ghent, Belgium.

439

Schönbach, P., Wan, H., Gierus, M., Bai, Y., Müller, K., Lin, L., Susenbeth, A., Taube, F., 2011. Grassland responses to grazing: effects of grazing intensity and management system in an Inner Mongolian steppe ecosystem. Plant Soil 340, 103–115. Sheng, G., Harazono, Y., Oikawa, T., Zhao, H., Zong, Y., Xue, L., 2000. Grassland desertification by grazing and the resulting micrometeorological changes in Inner Mongolia. Agric. For. Meteorol. 102, 125–137, http://dx.doi.org/10.1016/ S0168-1923(00)00101-5. Shi, S., Zou, X., Zhang, C., Su, G., 2009. Investigation of vegetation restoration in the area of Beijing and Tianjin sandstorm source control project. Sci. Soil Water Conserv. 7, 86–92 (in Chinese with English abstract). Tan, M., Li, X., 2015. Does the Green Great Wall effectively decrease dust storm intensity in China? A study based on NOAA NDVI and weather station data. Land Use Policy 43, 42–47, http://dx.doi.org/10.1016/j.landusepol.2014.10.017. The State Council, 2002. Beijing–Tianjin Wind/Sand Source Control Project Plan (2001–2010). The State Council, Beijing (in Chinese). The State Council, 2012. Beijing–Tianjin Wind/Sand Source Control Project Stage II Plan (2013–2022). The State Council, Beijing (in Chinese). Tobler, W., 1970. A computer movie simulating urban growth in the Detroit region. Econ. Geogr. 46, 234–240, http://dx.doi.org/10.2307/143141. Tong, C., Wu, J., Yong, S., Yang, J., Yong, W., 2004. A landscape-scale assessment of steppe degradation in the Xilin River Basin, Inner Mongolia, China. J. Arid Environ. 59, 133–149, http://dx.doi.org/10.1016/j.jaridenv.2004.01.004. Van Andel, J., Aronson, J., 2012. Restoration Ecology: The New Frontier. John Wiley & Sons, Hoboken, New Jersey. Waldron, S., Brown, C., Longworth, J., 2010. Grassland degradation and livelihoods in China’s western pastoral region: a framework for understanding and refining China’s recent policy responses. Chin. Agric. Econ. Rev. 2, 298–320, http://dx.doi.org/10.1108/17561371011078435. Wang, Y., Qiao, Y., 2011. Problems in implementing retire livestock, restore grassland projects and counter measures. Agric. Econ. Prob. 2, 99–103, http:// dx.doi.org/10.13246/j.cnki.iae.2011.02.012 (in Chinese). Wang, X., Han, J., Dong, Y., 2005. Recent grassland policies in China: an overview. Outlook Agric. 34, 105–110, http://dx.doi.org/10.5367/0000000054224319. Wang, J., Brown, D.G., Agrawal, A., 2013. Climate adaptation, local institutions, and rural livelihoods: a comparative study of herder communities in Mongolia and Inner Mongolia, China. Global Environ. Change 23, 1673–1683, http://dx.doi. org/10.1016/j.gloenvcha.2013.08.014. Wang, L., 2009. Some thoughts on the performance and sustainability of retire livestock restore grassland projects. J. Inner Mongolia Agric. Univ. (Soc. Sci. Ed.) 11, 116–118 (in Chinese). Wooldridge, J.M., 2015. Introductory Econometrics: A Modern Approach. Nelson Education, Toronto, Canada. Wu, J., Zhang, Q., Li, A., Liang, C., 2015. Historical landscape dynamics of Inner Mongolia: patterns, drivers, and impacts. Landsc. Ecol. 30, 1579–1598, http:// dx.doi.org/10.1007/s10980-015-0209-1. Xing, Q., Liu, A., Liu, Y., Gaowa, 2005. Study on change of rangeland vegetation using MODIS-NDVI in Xilingol Grassland. Acta Agrestia Sin. 13, 15–19 (in Chinese). Xu, H., Tang, X., Liu, J., Ding, H., Wu, J., Zhang, M., Yang, Q., Cai, L., Zhao, H., Liu, Y., 2009. China’s progress toward the significant reduction of the rate of biodiversity loss. BioScience 59, 843–852, http://dx.doi.org/10.1525/bio.2009. 59.10.6. Xue, R., Zheng, S., Bai, Y., 2010. Impacts of grazing intensity and management regimes on aboveground primary productivity and compensatory growth of grassland ecosystems in Inner Mongolia. Biodivers. Sci. 18, 300–311, http://dx. doi.org/10.3724/SP.J.1003.2010.300. Yang, Y., Fang, J., Ma, W., Smith, P., Mohammat, A., Wang, S., Wei, W., 2010. Soil carbon stock and its changes in northern China’s grasslands from 1980s to 2000s. Global Change Biol. 16, 3036–3047, http://dx.doi.org/10.1111/j.13652486.2009.02123.x. Ye, D., Chou, J., Liu, J., Zhang, Z., Wang, Y., Zhou, Z., Ju, H., Huang, Q., 2000. Causes of sand-stormy weather in northern China and control measures. Acta Geog. Sin. 55, 513–521 (in Chinese).