Spatiotemporal variations of vegetation and its determinants in the National Key Ecological Function Area on Loess Plateau between 2000 and 2015

Abstract China defined 25 National Key Ecological Function Areas in 2010 and adopted various measures to support ecosystem restoration in these areas. During the process of environment policymaking, it is important to observe the variation of vegetation and its driving factors. In this paper, we chose the National Key Ecological Function Area (NKEFA) on Loess Plateau as the study area. Based on MODIS‐NDVI data between 2000 and 2015, the trend analysis was used to depict the change in NDVI and the stepwise regression analysis method was used to quantitatively assess its determinants. The results show that: (a) The vegetation coverage in study area was low in the northwest and high in the southeast, corresponding to the distribution of precipitation and temperature. (b) NDVI in the growing season increased remarkably from 0.2841 in 2000 to 0.4199 in 2015 with a linear tendency of 0.085/10a. About 71.22% of the study area experienced an extremely significant increasing of NDVI, while only 0.03% of the total area suffered from significant decreasing of NDVI. (c) Compared to climatic factors, ecosystem conservation policies, and labor transfer contributed more to the vegetation changes in the study area. In order to ensure ecological security and sustainable development in these areas, it is necessary to maintain the continuity of ecological compensation policy. Moreover, developing targeted eco‐compensation policies and encouraging farmers to participate in nonfarm employment are effective ways to reach a win–win outcome of reducing the ecosystem pressure and improving the welfare of rural households.


| INTRODUC TI ON
The spatiotemporal variation of vegetation is the result of interaction between climate factors and human activities. At large scale, human activities usually refer to human disturbances, such as cultivation activities, road traffic, and urban land use (Gao et al., 2017).

But in National Key Ecological Functional Areas (NKEFA) on Loess
Plateau, human activities like ecological protection policies or labor force transfer may have more influence on vegetation condition.
The Major Function Oriented Zoning issued by the State Council of China in 2010 identified 25 NKEFA. On the one hand, these areas as national ecological security safeguard are targeted to provide ecological products and ecosystem services. On the other hand, NKEFA mostly overlapped with poverty-stricken areas, faced with serious ecological degradation under a continuously extensive economic growth model (Li, Yuan, Gao, & Xu, 2014). By 2015, the government had arranged 251.3 billion yuan for these regions, to strengthen ecological protection and improve the economic level.
Severe ecosystem degradation threatens regional sustainable development and ecological security in NKEFA on Loess Plateau.
Since 1999, the Chinese government has implemented many ecological protection policies in this area, such as the Grain for Green (GFG) project, Natural Enclosing and Prohibiting Grazing project, Natural Forest Conservation project, Ecological Public-welfare Forests Protection policy, Grassland Subsidy policy, and transferring payment for NKEFA. Many studies have shown a significant improvement in local ecosystem after the implementation of ecological protection policies, indicating that these policies had received good results (Su, Peng, Xie, & Huang, 2011), particularly in fragile areas (Ouyang et al., 2016). Since the late 1980s, the net primary productivity (NPP) decreased in South China due to urban expansion, but North China experienced an increase in NPP with the help of vegetation restoration program, especially on Loess Plateau .
Except for ecological protection policies, labor transfer also influences on vegetation change. During the process of urbanization, a large number of rural labors in mountainous areas emigrated to urban areas, searching for more job opportunities and higher salaries, which bring about an impact on vegetation greenness (Li, Li, Tan, & Wang, 2017). More people moved to urban areas, engaging in nonagricultural jobs, which reduced the pressure on local ecosystem, especially in places where people mainly depended on agriculture and animal husbandry for their livelihood. Li et al. (2016) found that the emigration of agricultural labor had significantly improved vegetation cover in Inner Mongolia, which means compared to natural factors, agricultural labor affects more on vegetation cover.
As for the climatic factors impacting the vegetation change, a similar conclusion has been drawn in different researches. That is the correlation between vegetation and climatic factors was higher in monthly scale than yearly scale on Loess Plateau (Bai, Bai, & Wang, 2014;Duo, Zhao, Qu, Jing, & Xiong, 2016;Zhang, Fang, & Shi, 2016).
More researches focused on comparing the impact of human activities and climatic factors on vegetation coverage (Lu, Yu, Liu, & Dhruba, 2015;Zhou et al., 2015). Different contribution assessment methods were used to distinguish the contribution of human activity and climate factors to vegetation growth . Using the residual method, Tong, Zeng, and Wang (2016) found both climate change and human activities were important factors on vegetation variation in Shanxi Province, but the influence of human activities was relatively higher. Using correlation analysis, climate change only had a marginal effect on vegetation greenness in a typical hilly-gully basin on Loess Plateau, with a contribution of 9.3% (Bai, Mo, Liu, & Hu, 2019). At a larger scale, the contribution of human activities may also surpass the impact of climate change. According to Li, Peng, and Li (2017) and Gang et al. (2018) (Du, 1999). Thus, it has been widely applied in many studies (Anyamba & Tucker, 2005;Carlson & Ripley, 1997). Some studies show that NDVI is positively related to vegetation coverage in areas with low vegetation coverage (Watinee & Netnapid, 2013), which means NDVI is suitable for representing the vegetation coverage in North China. Therefore, this paper used NDVI to study the vegetation coverage change on Loess Plateau.
In this research, we selected the NKEFA on Loess Plateau as the study area, based on MODIS-NDVI data between 2000 and 2015, analyzed the spatiotemporal changes in NDVI, and quantitatively explored the impact of climatic factors and human activities on NDVI changes. Besides that, Yanchi County was selected as a case area to carry out a household survey, analyzing the driving mechanism of vegetation change and distinguishing the key factors in ecosystem changes. The study of dynamic vegetation change is helpful to understand the role of ecological protection policy in ecosystem restoration, which can provide some academic support for policy makers in sustainable natural resources utilization and ecosystem conservation.

| NDVI data
The MODIS data (MOD13A3), with a 1-km spatial resolution and 30-day intervals, were quality guaranteed after geometric precision correction, radiometric calibration, and atmospheric correction.
Considering the climate characteristics and vegetation growth status on Loess Plateau, May to September was selected as the growing season. NDVI of growing season was calculated as the mean value of the monthly NDVI from 2000 to 2015, which was used to represent the average NDVI of the year.

| Meteorological data
The mean value of monthly temperature and precipitation data of 77 meteorological stations located in or around the study area were The meteorological data were interpolated according to the spatial resolution of NDVI data.

| Statistical data
Statistical data at county level was used to comprehensively analyze the effects of labor transfer and ecological protection policies on NDVI change. The statistical data was obtained from the statistical yearbook and local survey, including GDP, number of population, farmland area of Yanchi, number of rural workers, number of rural laborers engaged in farming activities, area of newly planted arbor forests and shrub wood with a survival rate exceed 85% and subsidies for afforestation.

| Household survey data
Yanchi County was one of the first round counties implemented GFG project in NKEFA on Loess Plateau (Wang, Hao, Zhai, & Liu, 2017), located in the east of Ningxia Province. The area is a typical zone of fragile agro-pastoral interlaced region, north connecting to Mu Us Sandland and south connecting to Loess Plateau (Zhong et al., 2018).
Due to natural conditions and human activities, Yanchi County was faced with various ecological problems during the past decades, such as land desertification, overgrazing in the grassland, vegetation degradation, water and soil erosion, and scarcity of water resources (Wang, Hao, Zhang, Zhai, & Zhang, 2018). As a result, we selected Yanchi County as the case area for household survey, to observe the driving factors of NDVI in NKEFA. Using stratified random sampling method, a survey of 222 households was conducted in 2015. Survey methods include questionnaire investigation, semistructure interviews, and meetings. The questionnaire covered basic information of rural family, F I G U R E 1 Location of study area and the distribution of meteorological stations labor allocation, livelihood capital, the recognition of ecological compensation, the participation willingness of compensation policy and the expectation to eco-compensation subsidies, etc.

| Trend analysis
The trend of NDVI change between 2000 and 2015 was calculated as the slope of linear regression. The statistical significance of the trend was assessed by two-tailed significance tests. The formula is as follows: where i is the order of year from 1 to n; n is the number of study years; and NDVI i is the value of year i which was calculated by monthly NDVI data in the growing season. If slope > 0, the trend of NDVI increased during the study period; otherwise, it decreased.

| Partial correlation analysis
Partial correlation analysis is used to study the relationship between two specific variables. When two variables associated with the third variable simultaneously, the impact of the third one is excluded through partial correlation analysis. In this way, only the correlation of the other two variables is estimated. The formula is as followed: The partial correlation coefficient is calculated based on the correlation coefficient, the formula is as followed: where r ab⋅c is the partial correlation coefficient between variables a and b after fixing variable c; r ab , r ac , r bc are the correlation coefficients between variables a and b, a and c, b and c, respectively.

| Stepwise regression analysis
In this paper, a stepwise regression statistical model was applied to analyze the determinants of NDVI changes in a case area. The stepwise regression statistical model is a semiautomated process of building a model by successively adding or removing variables. This method introduces the variables that have passed the significance test, which is suitable for the multivariate analysis. The formula is as followed: where β j is the parameter of variable j; x ij is the value of variable j of the year i; α is a constant; ε is the error term.
In this study, the independent variables were selected to reflect the natural factors, labor force factors, and policy factors (Table 1).
Precipitation and temperature were selected to represent the natural factors. GDP per capita, farmland area per labor, and proportion of rural laborers engaged in farming activities were selected as variables which reflect economic pressure on ecosystem. Afforestation area and GFG subsidy funds were selected to reflect the impact of ecological protection policies.

| Spatial patterns of NDVI
The results showed that NDVI in study area is low in the northwest and high in the southeast, corresponding to the distribution of de- the mean values of NDVI are generally lower than 0.3 (Figure 2). belonged to significantly increased area. About 13.96% of the study area showed no significant variation, which were distributed sporadically in north Tongxin, south Dingbian, and east Yanchi (Figure 4).

| Trends of NDVI
Restricted by the hydrothermal condition, the "low in the northwest and high in the southeast" spatial pattern remained unchanged.

| Correlation analysis between NDVI and meteorological factors
The average precipitation of study areas showed a weakly increas-  probably not the determinant of the NDVI change. Similar inferences have emerged in previous study (Li et al., 2016). In order to further analyze the response of vegetation coverage to climate change, the partial correlation coefficients between NDVI and meteorological factors of every spatial raster were calculated ( Figure 6). The results showed the mean partial correlation coefficients between NDVI and precipitation, temperature were 0.237 and −0.226, respectively.
About 95.50% of the study area showed a positive correlation be-

| Relationship between NDVI changes and ecosystem policies
The increase in NDVI was sharp in 2001-2002 and 2011-2012, with the growth rates of 30.23% and 22.03% respectively, which may directly related to the implementation of ecosystem policies.
In August 1999, GFG project started in three provinces including Shaanxi, Gansu, and Sichuan After that, the project extended to 25 provinces, 1897 counties in 2002 (Wang & Chen, 2006), covering the whole study area. From 2000 to 2010, implementation of GFG project in study area had reduced sloping cropland by 1,571 km 2 and increased ecological land by 1,337 km 2 . Among them, the area of grassland and forest increased by 10.89% and 4.24%, respectively . This indicates that GFG project had promoted the ecosystem quality, especially in the first round (2001)(2002) with powerful policy enforcement and foundation support. According to Figure 3, NDVI in the study area increased rapidly after the two policies just implemented. But in the subsequent three years, the NDVI declined with fluctuation, especially in the fourth year (2004)(2005)(2014)(2015). This may be related to farmer willingness to participate in ecological protection policy. Zhao et al. (2017) also found this trend in their study, but they regard this as a result of low survival rate of trees. With a new round of GFG project officially started, NDVI in study area is expected to rise in the next five years. Maintaining the continuity of ecological protection policy is crucial to ecosystem restoration.

| Determinants of NDVI changes in case area
In order to identify the determinants of NDVI changes, this paper selected Yanchi County as a case area to conduct a regression analysis. The average NDVI in Yanchi had increased significantly with a rate of 52.92% from 2000 to 2015. The obvious increase in vegetation coverage in Yanchi has also been confirmed in previous research (Gao et al., 2017). The stepwise regression showed that the proportion of rural laborers engaged in farming activities (p < 0.05) and afforestation area (p < 0.05) were significant driving factors ( Table 2). Stepwise regression analysis and household survey showed that the ecological protection policy played an important role in promoting vegetation restoration in Yanchi. As the basic resource users and policy participants, the willingness of farmers will directly affect the results of ecosystem restoration policies (Li et al., 2010). The diversity of farmers' livelihood is the main factor influencing the farmers' willingness to participate in eco-compensation policies. According to the household survey, 58.1% of rural households had nonagricultural employment and 40.8% of rural labors had a short-term or long-term nonagricultural employment.

Proportion of rural laborers engaged in
Based on indicators such as present way of make a living, major income sources, and the job of major labor force, four household types are identified, including full-time farming households, farming-dependent households, nonfarming-dependent households, and nonfarming households (Zhang, Zhang, Yan, & Wu, 2008).
Among the four types, households who were willing to participate in policies took up 53.7%, 53.4%, 54.7%, and 66.7%, respectively.
Nonfarming households did not depend on farmland, while fulltime farming households had a higher dependence on land, whose income was mainly from agricultural or sheep products. Other relative studies also confirmed that livelihood diversification could effectively reduce their reliance on natural resources, which lead to a higher willingness to participate in eco-compensation policies (Li & Cai, 2014;Wang, 2010 TA B L E 2 The stepwise regression result of NDVI changes in Yanchi 2. As the new round of GFG project has officially started, the vegetation coverage in study area is expected to keep rising generally. The participation willingness of farmers will determine the effect and sustainability of ecological protection policies.
According to the household survey in Yanchi, the livelihood types of farmers, compensation standard, and policy stability will affect farmer willingness. Given the importance of the NKEFA, it is necessary to increase the ecological compensation intensity and maintain the continuity of ecological compensation policy. In order to achieve a win-win situation of farmers' livelihood and ecosystem conservation, a targeted ecological compensation policy should be considered.

3.
In this paper, NDVI was used as the index of ecosystem restoration, which took more consideration of its ecosystem services function in sand fixation and soil conservation. However, as the largest arid and semiarid zone in China, some researchers considered that large scale of vegetation restoration was not suitable for Loess Plateau due to a lack of water resources (Jin et al., 2018).In the future study, we should pay more attention to the comprehensive influence of ecological protection policy on ecosystem services at regional scale. Researches have already found that there is a trade-off between different types of ecosystem services (Dai, Wang, & Zhu, 2015;Rao, Lin, Wang, Zhang, & Lu, 2015). The impact of ecological protection policy on different ecosystem services is also different. According to a study in the Loess Plateau (Feng et al., 2016), the vegetation productivity was enhanced by the ecological policies at the cost of river runoff reduction. Therefore, in the future research, analyzing the trend of different ecosystem services based on regional ecological function positioning is necessary for promoting the scientific implementation and sustainable management of ecological protection policies.