Influences of population pressure change on vegetation greenness in China's mountainous areas

Abstract Mountainous areas in China account for two‐thirds of the total land area. Due to rapid urbanization, rural population emigration in China's mountainous areas is very significant. This raises the question to which degree such population emigration influences the vegetation greenness in these areas. In this study, 9,753 sample areas (each sample measured about 64 square kilometers) were randomly selected, and the influences of population emigration (population pressure change) on vegetation greenness during 2000–2010 were quantitatively expressed by the multivariate linear regression (MLR) model, using census data under the condition of controlling the natural elements such as climatic and landform factors. The results indicate that the vegetation index in the past 10 years has presented an increasing overall trend, albeit with local decrease in some regions. The combined area of the regions with improved vegetation accounted for 81.7% of the total mountainous areas in China. From 2000 to 2010, the rural population significantly decreased, with most significant decreases in the northern and central areas (17.2% and 16.8%, respectively). In China's mountainous areas and in most of the subregions, population emigration has significant impacts on vegetation change. In different subregions, population decrease differently influenced vegetation greenness, and the marginal effect of population decrease on vegetation change presented obvious differences from north to south. In the southwest, on the premise of controlling other factors, a population decrease by one unit could increase the slope of vegetation change by 16.4%; in contrast, in the southeastern, northern, northeastern, and central area, the proportion was about 15.5%, 10.6%, 9.7%, and 7.5%, respectively, for improving the trend of NDVI variation.

It is proved that the impact of climate factors is crucial on vegetation growth in some regions Chuai, Huang, Wang, & Bao, 2013;Piao, Mohammat, Fang, Cai, & Feng, 2006). Especially, temperature is claimed to play a dominated role compared to precipitation (Thavorntam & Tantemsapya, 2013;Tian et al., 2015). Some scholars have explored the relationship between human activities and vegetation change through the correlation analysis method (Cai et al., 2014;Lu et al., 2015). For example, Cai et al. (2014) have explored the relationship between population emigration and vegetation change at the county scale in the karst areas of southwest China, using spearman correlation analysis, and found a positive influence of population emigration on vegetation index. However, they only used one single factor and did not control other variables. Lu et al. (2015), using Pearson's correlation analysis, selected a large number of factors (such as population, labor force, GDP, investment.) at the provincial scale as explanatory variables and analyzed the influences of China's social and economic factors on vegetation index, without controlling the natural factors. Other authors have tried to isolate the influence of human activity from the comprehensive influences using residual analysis (Evans & Geerken, 2004;Ferrara, Salvati, Sateriano, & Nolè, 2012;Sun et al., 2015;Tousignant et al., 2010;Wessels et al., 2007). For example, Wessels et al. (2007) have used residual analysis to isolate the influences of human activities on vegetation productivity in a study on the impacts of land degradation in South Africa. In mountainous areas, Wang et al. (2015) have investigated the influences of climate and human activity factors on the vegetation of southern China using residual analysis, obtaining the regression equations of the vegetation index for temperature and precipitation, whereas the influences of human factors were completely explained by the residual terms of the regression equation. However, the residual analysis method could only reveal the positive or negative effects of human activity instead of identifying the types, intensity, and contribution ratio of human activities.
Such studies have deepened our understanding of the relationship between human activities and vegetation change. However, these approaches could not identify the types, intensity, and contribution ratios of human activities that influence vegetation greenness change and do not exclude the influences of natural factors, such as temperature and precipitation. In addition, these studies always choose a certain administrative unit as the research object. However, due to the large land area and complex topography of mountainous areas, the spatial differences inside the administrative unit are significant and therefore using an administrative unit as the analysis unit increases the uncertainty of the results.
Mountainous areas in China account for two-thirds of the total land area; they are characterized by high intensity of human activity and fragile ecological environments. Furthermore, a number of large rivers rise in China's mountainous areas, such as the Yangtze River, the Yellow River, and the Lancang River, as well as some international rivers; the state of the ecological vegetation of these areas significantly affects the hydrological conditions of these rivers in China as well as in neighboring countries. So, it is very important to evaluate the effects of human activities on mountainous vegetation in China. This study chose the widely used vegetation index NDVI (Normalized Difference Vegetation Index) as an indicator of vegetation conditions (Fu & Burgher, 2015;Guay et al., 2014;He et al., 2012;Liu & Gong, 2012;Starns, Weckerly, Ricca, & Duarte, 2015;Stow et al., 2003;Zhang, Zhang, Dong, & Xiao, 2013). Furthermore, we used two indices, population pressure (population density) and land-use intensity, to express human activity intensity. Then,9,753 samples of 64 square kilometers each were selected, to reduce the uncertainty caused using administrative units with a large area. This allowed us to quantitatively assess the effects of human activities on the vegetation conditions, using the multivariate linear regression (MLR) model under the condition of controlling the influences of natural factors.

| Study area and data source
The study area covers about five million square kilometers ( Figure 1).
According to previous research (Guo & Zhang, 1986), China's mountainous areas can be divided into seven regions: northwest, northeast, north, central, southeast, southwest, and Tibet.
In this study, we mainly used the following data: NDVI data, Network Information Center, Chinese Academy of Sciences (GSCLOUD), with a spatial resolution of 1 × 1 km and a temporal resolution of 1 month. In northern China, the growing season is from April to October (Li, Sun, Tan, & Li, 2016); therefore, the average NDVI value for the growing season was adopted to replace the annual average NDVI value in this study. Temperature and pre- Academy of Sciences. In this data set, there are seven main land use types. These are arable land, woodland, grassland, water area, construction land, unused land, and other areas. Population density data mainly refer to the research report of Tan, Li, Li, and Li (2016), and the spatial distribution diagram of population density is modeled based on nighttime light image data, land-use data and the fifth and sixth nationwide census data. Figure 2 shows the overall design of this study, expressed as a flowchart.

| Interpolation of meteorological data
Original temperature and precipitation data were grid data sets with a precision of 0.5 × 0.5°. These datasets were obtained by interpolation based on 2,472 weather stations across the country. In this study, on the platform of the ANUSPLIN (The Australian National University, Canberra, ACT, Australia) software, we carried out an interpolation on the grid data, using elevation as the covariate via thin-plate spline F I G U R E 1 Regional division and sample spatial distribution of China's mountainous areas. NOTE: the landform map is derived from State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences; measuring scale is 1:4,000,000 interpolation (Liu et al., 2008;Qian, Lu, & Zhang, 2010), thereby obtaining grid data with an accuracy of 1 × 1 km.

| Trend analysis of NDVI and meteorological data
As the interannual variabilities of temperature, precipitation, and NDVI were obvious, trend analysis was used to calculate their variation trends from 2000 to 2010 (Tian et al., 2015;Wang et al., 2015;Zhao et al., 2015). Here, we use temperature as an example to introduce the calculation process for this method: where T_Slope is the slope of the temperature change, m is the fixed number of years of the study, equaling 11 in this study, and t j is the temperature of the j-th year. When T_Slope < 0, the temperature presents a decreasing trend during this study period; otherwise, it presents an increasing trend. In this study, the data were processed at the pixel level, and the average value of each index was determined in the sampling scope using the zonal statistics tool of ArcGIS 10.2.

| Calculation of land-use intensity (LUI)
According to the comprehensive analysis methods to measure LUI proposed in previous studies (Gao, Liu, & Zhuang, 1998;Liu, 1997;Wang, Liu, & Zhang, 2001;Zhuang & Liu, 1997), the land was divided into four land use grades, that is, the unused land grade (with a grading index of 1), which contains saline-alkali land, marsh land, sand land, bare land, and other unused or hardly used land, for instance, alpine desert and tundra; the forest-grass-water land grade (with a grading index of 2), which includes forest land, grass land, and water areas; the agricultural land grade (with a grading index of 3), which includes cultivated land, garden land, and artificial grassland; and the urban settlement land grade (with a grading index of 4), which includes town land, residential land, and industrial and traffic land. The calculation formula for the comprehensive index of LUI is as follows: where I represents the land-use intensity, i is the number of landuse intensity grades, M i refers to the grading index of the i-th landuse intensity grade, and S i represents the area percentage of the i th (1)

| Selection of influencing factors
In terms of studying the influencing factors on vegetation change, this article mainly selects natural factors and human activities (Table 1).
For natural factors, due to the obvious differences in the interannual change of temperature and precipitation, the variation trends of annual total precipitation and annual average temperature were used in the study period. Besides, gradient and aspect determine the vegetation site conditions and are also introduced into the model as the explanatory variables. The differences in land-use intensity can reflect the influence degree of human land-use activities on vegetation change (Zhuang & Liu, 1997), which can quantitatively reveal the comprehensive level of regional land use (Wang , Liu, & Zhang, 2001).
Here, population density change represents the indicator that reflects population pressure.

| Influences of population pressure change on vegetation greenness variation
Before model estimation, we adopted the Variance Inflation Factor (VIF) to carry out a full-collinearity test on the explanatory variables. All operations were implemented by Stata 13.0; The number of samples for the statistics is 9,753; * is 10,000 times that of NDVI_Slope, ** is 1,000 times that of Slope temperature, and *** is the natural logarithmic of original Average Elevation.

Variables
T A B L E 2 Summary statistics of variables

F I G U R E 3 Spatial distribution of average NDVI values in the growing season in 2000 in mountainous areas of China
For all explanatory variables, VIF values were below 10, which means that there was no significant collinearity between variables.  (Table 3), the standard partial regression coefficient shows that the main factors influencing the NDVI change, according to their influencing degrees from high to low, include average elevation, trend of precipitation variation, population pressure change, trend of temperature variation, gradient slope, and aspect.
To reveal the regional differences, we carried out further analysis on the factors that influence vegetation greenness change in Models 5-11. The results show that overall, population pressure change significantly influences the trend of the NDVI variation for the six regional models, except for the northwest. Among these models, the influence of population pressure on vegetation greenness was significant at the 1% significance level for Models 5,8,9,10,and 11, and significant at the 5% significance level for Model 6. As introduced in Tibet was totally different, that is, a population increase by one unit could increase the trend of NDVI variation by 18.5%.
Furthermore, Table 4 shows that there were large regional differences for the influences of land-use intensity on the explained variables. In the northwestern and the Tibet areas, land-use intensity had a significant positive influence on vegetation greenness change at the significance level of 5%. In other regions, land-use intensity change had no significant impact on vegetation greenness change. not the focus of this study, we will further elaborate on this in the Appendix S1.

| DISCUSSION
In this study, on the basis of analyzing the variation trend of the From the beginning of the 21st century, the rural population in China's mountainous areas has been decreasing significantly. In general, population pressure in two-thirds of China's mountainous areas has been decreasing in the past 10 years. For instance, the northern, T A B L E 3 Models of impact of population pressure change on NDVI slope at the national level in mountainous areas in China The figures in [] are marginal effects of population pressure change; the figures in () are t values; *, **, *** are coefficients different from zero at 10%, 5%, and 1% significance levels, respectively; Region dummies = Yes. Standard error adjusted for 9,753 clusters in each sample.
central, and southeastern mountainous areas presented the most obvious decrease. From 2000 to 2010, the rural populations of these three regions fell by 17.2%, 16.8% and 12.6%, respectively. Accordingly, the slope of NDVI change in these regions was relatively large, and vegetation greenness increase was significant ( Figure 6). In contrast, in the northwestern and Tibet areas, the rural population only decreased by 4.1% and 1.8%, respectively, which was significantly lower than the national average of 17% (Li, Sun, Tan & Li, 2016). As a result, the NDVI increased slowly in this area.
With a decrease in rural population, vegetation index generally showed an increasing trend, except for a few regions which showed a decreasing trend and accounted for 18.3% of the total study area during the study period. Some scholars in China have come to similar conclusions (Cai, Yang, Wang & Xiao 2014;Han & Xu, 2008;Li, Sun, Tan & Li, 2016). For instance, Lu et al. (2015) have stated that China experienced both vegetation restoration and degradation with great spatial heterogeneity. In addition, Han and Xu (2008), using the correlation analysis method, found that demographic factors significantly affected vegetation productivity in the undeveloped regions with a great distance to the center of Chongqing, especially in the mountainous areas. In other countries in the developing world, research led to similar conclusions. For example, Olsson, Eklundh, and Ardö (2005) have found that the population emigration in marginal areas of the southern Sahara region increased the cultivated land abandonment rate, thus promoting the spontaneous recovery of vegetation to a certain degree. The results of a similar study have shown that cultivated land abandonment in mountainous areas caused by rural-to-urban labor migrants in St. Lucia, West Indies, had a certain facilitating effect on forest restoration in mountainous areas (Bradley, 2016). Based on these studies, it can be concluded that population pressure decrease positively impacts vegetation greenness change.
However, as mentioned in the introduction, previous researches have rarely analyzed the effects of human activity changes on vegetation greenness while controlling natural factors. In this study, we quantitatively analyzed the influences of human activities on the basis of controlling climatic and landform factors (Table 1)  contributed to the improvement of vegetation conditions. Studies showed that through irrigation, the grassland biomass in Tibet could be significantly increased (Ganjurjav et al., 2014(Ganjurjav et al., , 2015. Above all, the proportion of shrubs and broad-leaved forbs was also increased under irrigation conditions. Generally speaking, with higher land-use intensities and human activities, vegetation greenness decreases. However, in this study, landuse intensity had no significant impact on the dependent variables in most of the models. This may mainly be related to the definition of land-use intensity in this study. We calculated land-use intensity according to equation (2), which divides land-use status into four different grades. Nevertheless, this discontinuity of variables cannot fully reflect the influences of land use and masks a large amount of vegetation responses to land-use change.
Remarkably, in Model 7 and 11 (Table 4), land-use intensity has a significant positive influence on the dependent variables; the increase of land-use intensity can promote vegetation greenness. This situation occurs in northwestern China and is, most likely, mainly related to the development of irrigated agriculture in the area. In the valley and piedmont zones of the northwestern area, the temperature rise in recent years has increased the amount of alpine snow water, which promoted the development of irrigated agriculture in these regions, thus improving regional vegetation (Ta, Dong, & Caidan, 2006). Previous studies have shown that from 1975 to 2005, the large-scale development of cultivated land in Xinjiang has significantly influenced regional vegetation (Wang, Wang, Zhang, & Duan, 2014). The vigorous development of irrigated agriculture in these regions has improved the vegetation conditions to some extent.
In addition, vegetation change is affected not only by natural conditions and human factors, but also by other factors such as land use policy, related projects, and policies of vegetation protection (Li, Wu & Huang, 2012;Lu et al., 2015;Luck, Smallbone, & O'brien, 2009).
Especially since the 1990s, large-scale ecological protection and afforestation projects have been significantly affecting vegetation F I G U R E 5 Standardized coefficients of significant explanatory variables based on multivariate linear regression models shown in Table 4 in subregions F I G U R E 6 The rural population and average values of trend of NDVI variation in different areas restoration (Lu, Fu, Wei, Yu, & Sun, 2011). The variable of land-use intensity in this study can, to a certain extent, reflect the influence of the "grain-to-green" policy. However, land-use changes do not fully reflect the influence of policies on vegetation greenness. Further studies are therefore required to assess the impacts of population change on vegetation greenness change.