You are on page 1of 13

Ecological Indicators 113 (2020) 106278

Contents lists available at ScienceDirect

Ecological Indicators
journal homepage: www.elsevier.com/locate/ecolind

Analysis of spatial variability in factors contributing to vegetation T


restoration in Yan'an, China
Daojun Zhanga,1, , Qiqi Jiaa,b,1, Peng Wangc, Jinting Zhangb, Xianhui Houa, Xiandong Lia,

Wangjun Lid
a
College of Economics and Management, Northwest A&F University, Yangling 712100, China
b
School of Resources and Environmental Sciences, Wuhan University, Wuhan 430079, China
c
Xi'an Geological Survey Center of the China Geological Bureau, Xi'an 710054, China
d
The School of Environmental Science and Engineering, Suzhou University of Science and Technology, Suzhou 215009, China

ARTICLE INFO ABSTRACT

Keywords: As a typical ecologically fragile region, the Loess Plateau of China has much improved in the ecological en-
Vegetation restoration vironment in the past two decades. In order to achieve a more efficient implementation of ecological projects in
Driving force analysis the future, it is of great importance to study the influencing factors and the driving mechanism of vegetation
Ecological policy restoration based on past vegetation restoration practices. However, human activity and natural factors may
Spatial heterogeneity
show different coupling effects in contributing to the vegetation restoration in different locations because of
Yan'an
their spatial heterogeneity. In that case, the traditional global regression based on the Ordinary Least Squares is
at risk of failure. With Yan'an as the study area, vegetation index data in 2000 and 2011 were used to calculate
the vegetation improvement map during the ecological projects; taking this map as the dependent variable, and
the topographical, meteorological, socio-economic, and policy factors as independent variables, Geographically
Weighted Regression model which can well deal with spatial heterogeneity, together with the Ordinary Least
Squares model, was applied in this study. The results show: (1) most of the variables selected in this research
have significant impacts on the vegetation restoration, while meteorological and socio-economic factors make
greater contributions; (2) the effect of factors contributing to vegetation coverage improvement varies sub-
stantially across the study site, with climatic and physical factors dominating, and socio-economic/policy factors
playing either positive or negative roles in different parts of the study area. It could be concluded that spatial
variability in factors contributing to vegetation restoration should be well considered when performing ecolo-
gical policy evaluation, and Geographically Weighted Regression method has unparalleled advantages of
Ordinary Least Squares model in this aspect since it provides scientific reference for policymakers to make
effective decisions according to local conditions and maximize the effectiveness of ecological policies.

1. Introduction certain extent (Ali et al., 2019; Kuriqi et al., 2019a, 2019b). In order to
improve its ecological environment, the state has implemented a series
Vegetation, as an important part of the terrestrial ecosystems, plays of vegetation restoration projects since 1998 (Yin, 2010; Qu et al.,
a key role in climate regulation, carbon fixation, as well as water and 2018; Yin et al., 2018), including the Returning Farmland to Forest and
soil conservation (Huang et al., 2010; Sun et al., 2015; Ellison et al., Grass Program (RFFG), which has attracted worldwide attention. The
2017). As a typical ecologically fragile area, as well as a traditional pilot implementation of the RFFG has been implemented in individual
agricultural and pastoral region in China, the Loess Plateau is facing counties in China (Tang, 2004). Although RFFG literally involves only
many challenges, such as dry climate, severe soil erosion, land de- the conversion of cultivated land to forest and grassland, other related
sertification, and rapid population growth (Shi and Shao, 2000; Guo measures, such as the Forest Land Conservation (FLC) and the Grass
et al., 2010; Jiang et al., 2016). The impact of human activities on the Land Conservation (GLC) are combined in the meantime (Zhang et al.,
ecosystem is even growing in the industrial era, for example, the uti- 2018a, 2019a). Overall, ecological projects have achieved remarkable
lization of water resources has a negative impact on the ecosystem to a results (FAO, 2000) and vegetation has been significantly restored


Corresponding author.
E-mail address: zdj@nwsuaf.edu.cn (D. Zhang).
1
These authors contributed equally to the work and should be considered co-first authors.

https://doi.org/10.1016/j.ecolind.2020.106278
Received 30 November 2018; Received in revised form 31 January 2020; Accepted 2 March 2020
Available online 10 March 2020
1470-160X/ © 2020 Elsevier Ltd. All rights reserved.
D. Zhang, et al. Ecological Indicators 113 (2020) 106278

Nomenclature and abbreviation list k The number of independent variable


xk Independent variables
Abbreviations and acronyms βk Unknown parameters
ε Represents error item
RFFG Returning Farmland to Forest and Grass Program n The sample number, i = 1,2⋯n
FLC Forest Land Conservation X The matrix of the independent variable
GLC Grass Land Conservation Y The column vector of the dependent variable The column
NDVI Normalized Differential Vegetation Index vector of the dependent variable
EVI Enhanced Vegetation Index β The column vector of the unknown parameter
GWR Geographically weighted regression u The geographic coordinate position of each sample i
OLS Ordinary Least Squares regression W(u) The weighted diagonal matrix
MVC Maximum Value Composites βi(u) The local regression coefficient at position u
AIC Akaike information criterion Wij The geographical weight of the sample point j to the
CV Cross-validation current sample point i
dij The Euclidean distance of the sample point j to thecurrent
Variables and parameters sample point i in the local window
b The bandwidth
y Dependent variable

(Wang et al., 2015; Gao et al., 2017). In order to perform more efficient negative impact of human activities, and the tradeoff between eco-
ecological projects, an objective and accurate evaluation of past vege- nomic and environmental costs (Hou et al., 2019; Bejarano et al., 2019;
tation restoration practices should be performed. Kuriqi et al., 2019b). In summary, the previous studies have carried out
Some researchers have conducted in-depth studies on driving forces an evaluation of vegetation restoration projects and made some at-
and the driving mechanism of vegetation restoration from different tempts and exploration of the choice of evaluation indicators and
spatial and temporal scales (e.g., Scholtz et al., 2014; Ndayisaba et al., models. However, there are some weaknesses that need to be
2016; Kang et al., 2017). The factors affecting vegetation restoration strengthened in subsequent research. Firstly, previous studies did not
are generally divided into natural factors and human factors (Hu et al., pay enough attention to human activities, especially the effect of policy
2016; Bao et al., 2017; Qiu et al., 2017). The former includes elevation, factors. Some studies directly referred to the residuals in the regression
slope gradient, temperature, and precipitation; and the latter mainly model as the effect of human activities (Li et al., 2011; Li et al., 2017;
refers to population, economy, policy, etc (Liu et al., 2015; Sun et al., Wang et al., 2017). Furthermore, few studies separated the effect of
2015; Li et al., 2017). It is generally believed that the vegetation re- policy from other social-economic factors, which made it difficult to
storation is the result of combined effects of natural conditions and objectively evaluate the effects of ecological projects. Secondly, it
human activities (Bégué et al., 2011; Justice et al., 2015; Jin et al., should be noted that there is a significant difference in ecological re-
2015; Yan et al., 2017), while different researches have concluded storation effect across the country (Li et al., 2011), while previous
different dominant driving forces. The effect of human activity in the studies seldom pay attention to the spatial heterogeneity of driving
process of vegetation restoration is debatable. For example, Li et al. factors for the vegetation restoration, and the regulation obtained by
(2011) considered that climate change was a decisive factor affecting the traditional Ordinary Least Squares regression (OLS) model can
the growth of vegetation, and in the meantime, human activities could hardly fit each location in the study area (Han et al., 2016; Zhang et al.,
accelerate the rate of vegetation change in the short term; Zhao et al. 2018b).
(2016) found that human activities had double effects on vegetation In order to overcome the shortcomings of classical statistics models
changes, and their constructive role was stronger than their destructive in current vegetation restoration evaluation, this study introduced GWR
effect; Alix-Garcia et al. (2016) provided a case study showing that to provide a decision-making basis for better implementation of vege-
political and economic transition could bring huge changes in vegeta- tation restoration planning in the future. Firstly, unlike prior studies
tion coverage; Qu et al. (2018) concluded that ecological restoration that used residual to represent some hard-to-obtain factors, this study
projects were the main factor for vegetation restoration, while the identified and included as many variables as possible to increase the
benefits were closely related to the combined effect of the topography, explanatory power of the research model. The human activity well
climate, and human management. expressed and measured from different aspects, which included rural
With respect to data processing models and methods, correlation road density, population density, and ecological policies. Secondly, the
analysis, classical linear regression analysis, residual analysis, com- GWR model has been widely used in various fields to overcome the
parative analysis, and sensitivity analysis were commonly used in study negative effects brought by the spatial heterogeneity (Robinson et al.,
of driving force analyses (Zhao et al., 2013; Tian et al., 2014; Liu et al., 2013; Li and Liu, 2016; Zhang et al., 2016; Ahmed et al., 2017), and
2015; Kuriqi et al., 2019b). More recently, the importance of spatial hence we used the GWR model to analyze the differences in vegetation
effects was gradually realized. For example, Du et al. (2014) applied the restoration effects and their causes from the perspective of spatial
Geographically weighted regression (GWR) model to analyze the re- heterogeneity. Besides, in order to compare the results obtained by OLS
lationship between land cover change and driving factors under dif- and GWR respectively, both of these two models were applied in this
ferent sampling scales; Chen et al. (2017) used the space panel model research. The OLS model was used to explore driving forces for the
which can deal with spatial dependence to assess the impact of the vegetation restoration in the whole study area, which could help to
Inner Mongolia grassland protection project on vegetation recovery; select independent variables for the GWR; and the GWR model was
Shen et al. (2017) analyzed the response of the vegetation restoration to established based on local window to detect the spatial variations of
climatic factors from basin-scale and pixel scale respectively. In addi- these driving factors, which was expected to help the decision-makers
tion, Zhang et al. (2019b) and Xu et al. (2020) introduced vegetation put forward effective and targeted measures according to local situa-
restoration potential and potential realization models in vegetation tions.
restoration evaluating, so that the role of provenance policies can be
better separated; other scholars emphasized the sustainability goals, the

2
D. Zhang, et al. Ecological Indicators 113 (2020) 106278

2. Study area observations, the meteorological data used in this work were collected
from the data sets for consecutive years and consecutive days of the
Yan'an City is located in northern Shaanxi Province (Fig. 1), China Meteorological Data Service Center (http://data.cma.cn/site/
35°21′~37°31′N, and 107°41′~110°31′E, has an area of 37,000 km2 index.html) from 1981 to 2010, which were interpolated to get the grid
with a total population of 2.19 million. It administers one district, i.e. data layers of the meteorological factors with resolutions of
Baota, and 12 counties, i.e. Wuqi, Zhidan, Ansai, Zichang, Yanchang, 30 m × 30 m in the study area. The socio-economic data were mainly
Yanchuan, Yichuan, Hunglong, Luochuan, Huangling, Fuxian, and extracted from the land use maps interpreted from the TM images and
GanQuan. With an arid and semi-arid climate, the average annual annual socio-economic statistics from 1999 to 2010, and they were
precipitation is about 500 mm, and the mean annual sunshine ranges taken average values in these years then. The vegetation index data set
from 2,300 to 2,700 h, and the average annual temperature varies is derived from the 16-day MOD13Q1 data product provided by the
7.7 °C–10.6 °C. Yan'an is situated in the middle reach of the Yellow LPDAAC (Land Process Distributed Active Archive Center), with a re-
River and belongs to the hilly and gully region of the Loess Plateau, solution of 250 m × 250 m. Considering the time lag of vegetation
whose terrain is high in the northwest and low in the southeast with an growth relative to land-use type adjustment, the time nodes of the ve-
average elevation of 1,200 m. Yan’an took the lead in carrying out getation index data are identified as 2000 and 2011 respectively, and
large-scale RFFG, FLC, and GLC since 2000, and it had completed a total the final used yearly vegetation data were synthesized with all 16-day
area of about 71,800 ha2 of RFFG by the end of 2016, accounting for MOD13Q1 data in a certain year according to the average vegetation
19.4% of the total land area. index value of the growing season (Zhang et al., 2018a). In ENVI 5.2,
images are subjected to pre-processing such as radiation correction,
atmospheric correction, cropping, and format conversion. All data are
3. Materials and methods masked in the ArcGIS 10.2 to the range of the study area, and the
spatial resolution is unified to 30 m × 30 m. The coordinate reference
3.1. Data sources and processing is defined as the WGS-84 geocentric coordinate system with UTM
projection, and the central meridian is 108° E.
The land use data were obtained by interpretation of the Landsat 5
TM images in 1999 and 2010. The original images were downloaded
from the Geospatial Data Cloud website of the Chinese Academy of 3.1.1. Land use data
Sciences (http://www.gscloud.cn) with a spatial resolution of With the Landsat 5 TM images in 1999 and 2010 as data sources, the
30 m × 30 m. The temporary resolution of Landsat 5 TM is 16-day. To land use database update method is used to manually interpret land use
eliminate the interference of the cloud, and select the highest quality maps on the ArcGIS 10.2 platform. First of all, land use map in 2010
images which can make the best discrimination among different land- (1:10,000) can be directly obtained from Yan'an land use database, and
use types, we choose the target image from May to August within one then to interpret land-use types in 1999 based on the land use map of
year. The topographic data, ASTER GDEM V2 elevation data, which was 2010 by comparing the two remote sensing images for 1999 and 2010
released in 2011, also came from the Geospatial Data Cloud website of respectively. If change is found between the two years’ images in a
the Chinese Academy of Sciences with the same resolution of TM polygon in the land use map of 2010, land use type will be updated for
images. Because meteorology needs to be based on long-term stable the new land use map for 1999; if not, the land use type in 1999 will be

Fig. 1. Location and range of the study area.

3
D. Zhang, et al. Ecological Indicators 113 (2020) 106278

the same as that of 2010. Finally, the land use maps are converted to established. In this study, land-use transfer matrices were obtained
grid layers (Fig. 2). According to the characteristics of land use types in through the overlay analysis of two-year land-use grid maps. As a re-
the study area and research needs, the types of land use are classified sult, the area of the RFFG, the GLC, and the FLC, as well as the popu-
into cultivated land, forest, grassland, water area, construction land, lation density were obtained. The zonal statistics, which also belongs to
and unused land. spatial overlay analysis, was used to calculate the average value of the
growth of EVI, wind speed, humidity, temperature, terrain elevation
3.1.2. Vegetation index data and slope, and the area of land use conversion types for each village.
MOD13Q1 contains two types of vegetation index data, i.e.
Normalized Differential Vegetation Index (NDVI) and Enhanced 3.2.2. Ordinary Least Squares regression (OLS)
Vegetation Index (EVI). It is believed that EVI has higher recognition of OLS is a global linear regression model based on the least square
high-coverage vegetation and greater variance (Huete et al., 2002), method. All samples with equal weights are used at once to establish a
which can better reflect the vegetation spatial difference within the unified model for the analysis of the linear relations between multiple
study area. Thus, EVI is selected to indicate the state of vegetation independent variables and dependent variables in the study area. The
growth and vegetation cover in this study. To eliminate the effect of least square method is used to obtain partial regression coefficients for
different vegetation growing seasons, maximum EVI of every grid is each variable, indicating the degree of contribution of each in-
obtained through the Maximum Value Composites (MVC), and the dependent variable to the dependent variable. Suppose that the de-
difference of maximum EVI values between 2011 and 2000 is used to pendent variable (the improvement of EVI from 2000 to 2011) is y , and
indicate the degree of vegetation restoration. k independent variables of natural factors, social-economic factors, and
ecological policy factors are x1, x2 , , andxk , and then the OLS model
3.1.3. Other data can be expressed as
The terrain elevation is downloaded from ASTER GDEM V2 eleva- yi = 0 + 1 x1i + 2 x2i + k xki + i (1)
tion data (http://www.gscloud.cn) shared by Geospatial Data Cloud
website of the Chinese Academy of Sciences, with a spatial resolution of where 0 , 1, 2 , , and k are k + 1 unknown parameters, ε represents
30 m × 30 m. The terrain slope can be calculated based on the ele- error item, k is the number of independent variable; i = 1, 2 n , and n
vation data with the Slope Tool in ArcToolBox. The data obtained from is the sample number. The least square method can be used to calculate
the meteorological monitoring stations are spatially interpolated (anti- the regression parameter matrix
distance weighted average interpolation method) to form meteor-
= (X TX ) 1X T Y (2)
ological grid layers including the annual mean humidity, the annual
mean temperature, and the annual mean wind speed. Using the rural where X andY represent the matrix of the independent variable and the
road layer (in the form of a polyline) as the original data, the rural road column vector of the dependent variable respectively, and represents
density layer in the study area is obtained through density analysis. All the column vector of the unknown parameter.
these data are unified as 30 m × 30 m grid data at last.
3.2.3. Geographically weighted regression (GWR)
3.2. Methods To overcome the spatial heterogeneity and non-stationarity of
geographic data, British scholars Fotheringham and Brunsdon
3.2.1. Spatial overlay analysis (Fotheringham et al., 1996; Brunsdon et al., 1996) proposed the Geo-
Spatial overlay analysis refers to the statistical analysis of pixel at graphically Weighted Regression (GWR) model, in which local window
the same location in two or more grid layers, by which the relations regression instead of global regression is used. Within each local
between any two different attributes of a spatial object can be window, a local regression model is established with neighboring

Fig. 2. Land use maps of Yan'an in (a) 1999 and (b) 2010 respectively.

4
D. Zhang, et al. Ecological Indicators 113 (2020) 106278

samples of different weights. The weight is usually implemented by the is a column vector representing the value of the dependent variable.
inverse distance function, that is, the closer the sample is to the current The geographical weight is obtained by the distance decay function
point, the greater the weight is in the model. GWR is a regression model (kernel function), and the Gaussian function is a typical kernel function
that allows for spatial variation in parameter estimates in order to (Eq. (6)).
achieve higher model accuracy (Zhang et al., 2018b). When the local
window size is infinite, GWR degenerates into an OLS model. Therefore, wi1 (u) 0 0
the former can be regarded as the promotion of the latter, and the latter 0 wi2 (u) 0
Wi (u) =
is a special case of the former. The mathematical expression of the GWR
model is 0 0 win (u) (5)
yi (u) = 0 (u) + 1i (u) x1i + 2i (u) x2i + ki (u) xki + i (3)
0.5(dij b)2
where yi (u) is the EVI improvement at location u ; u denotes the geo- wij = e (6)
graphic coordinate position of each sample i ; 1i (u), 2i (u), , ki (u)
denote the local regression coefficients of x1i , x2i , , xki , and 0 (u) is the where wij is the geographical weight of the sample point j to the current
constant term. The Eq. (3) can be solved by a weighted least square sample point i, dij is the Euclidean distance of the sample point j to the
method, current sample point i in the local window. b is the bandwidth, meaning
the range of the local window. The optimal bandwidth can be de-
(u) = (X TW (u) X ) 1X T (W (u)) Y (4) termined by the Akaike information criterion (AIC) (Akaike, 1974) or
where W (u) is the weighted diagonal matrix (Eq. (5)), and the elements Cross-validation (CV) (Cleveland, 1979).
on the diagonal are geographical weights, determined by the distance The detailed data processing flow and methodological framework of
from the sample point to the current point, and the elements on the this study are shown in Fig. 3.
non-diagonal line are 0. (X TW (u) X ) 1is a weighted covariance matrix.

Fig. 3. Study design conceptual framework.

5
D. Zhang, et al. Ecological Indicators 113 (2020) 106278

4. Results 4.3.1. Driving force analysis based on OLS


With the vegetation restoration as the dependent variable, and the
4.1. Land use change analysis before and after ecological engineering factors from climate, topography, socio-economy and ecological po-
licies as independent variables, traditional multivariate linear regres-
Land-use changes in Yan'an during the period of 1999–2010 were sion model based on OLS was performed to explore the driving force for
mainly reflected in the sharp decline in cultivated land and a sub- vegetation restoration. We can see from Fig. 5 that the adjustment R2 of
stantial increase in forestland (Table 1). During these years, the largest the regression is 0.34, which indicates that the selected variables can
change was in cultivated land, which decreased by 54.57%. It is fol- account for 34% of the increase in the vegetation index, the F-value is
lowed by forestland with the increase of 29.14%, and the construction 179.61, and that the equation is tested by a significant level of 1%. At a
land area increased by 23.15%. On the one hand, the changes in land significant level of 5%, the humidity, temperature, FLCR, and RFFGR
use types were the result of the implementation of ecological projects, have significant positive effects on the increase of EVI, while the wind
because a large amount of sloping farmland was converted into forest speed, rural road density, population density, and slope gradient are
and grass; on the other hand, construction land expansion has also led significantly negatively correlated with the increase of EVI. From a
to tremendous cultivated land loss. global point of view, the population density has the greatest impact on
The land-use transfer matrix from 1999 to 2010 (Table 2) indicates vegetation recovery, with a coefficient of −0.209, followed by the
that the change in land use structure was mainly among cultivated land, temperature and humidity, with the regression coefficients of 0.205 and
forestland, and grassland, which further proves the above viewpoints. 0.160, respectively. In ecological policies, RFFGR and FLCR contribute
Overall, the coverage rate of forest and grass increased from 74.0% to a lot to the vegetation restoration, with a contribution of 0.105 and
86.8% in Yan'an during 1999–2010. 0.081, respectively, but the contribution of GLCR is relatively small at
0.012.
4.2. Vegetation restoration analysis before and after ecological engineering To quantitatively analyze the effect of ecological policies on vege-
tation restoration, we further performed the sensitivity analysis based
In this study, the increase of EVI from 2000 to 2011 (Fig. 4) is used on the regression model obtained by OLS. Assuming that the density of
to indicate vegetation restoration. It can be seen from Fig. 4 that the RFFGR increases and decreases by 5% respectively, the overall vege-
vegetation in Yan'an showed an overall improvement trend while de- tation coverage in the study area would increase and decrease by 0.5%
terioration could be found in local positions, with a significant spatial accordingly. It was further assumed that if logging and grazing were not
variation. The EVI increased by an average of 0.11, with 0.22 of max- banned on existing forest and grassland, and only afforestation was
imum enhancement. In particular, EVI was improved in 99.7% of total carried out, the vegetation coverage level in the study area would de-
villages, and the decrease was only observed in 10 villages, which were crease by 10%. Different ecological policies have made significant
mainly located in Huangling and Luochuan. contributions to the improvement of vegetation coverage in the study
Vegetation restoration was better in the south than that in the north. area. In particular, the effects of measures against logging and grazing
The EVI in the southern region (i.e. Fuxian, Ganquan, Luochuan, cannot be underestimated, and it is also a strong basis for the RFFG.
Huanglong, Yichuan, and Huangling) increased by an average of 0.13,
which was higher than that in the north. In the northern region (i.e.
Wuqi, Ansai, Zhidan, Zichang, Baota, Yanchuan, and Yanchang), the 4.3.2. Driving force analysis based on GWR
vegetation index increased by an average of 0.09. But in the northwest, The GWR 4.0 software tool (Nakaya, 2016) was applied to perform
especially in parts of Wuqi, Zhidan, and Ansai, due to the constraints of the GWR model where Gaussian function was used as a weight function,
climate, topography and other conditions, the vegetation restoration and the final fixed bandwidth was determined to be 20 km. The esti-
effect was relatively weak. mated results of GWR are shown in Tables 3 and 4. Compared with the
OLS model, GWR can obtain more accurate estimation and provide
4.3. Driving force analysis of vegetation restoration more information. As can be seen from Table 3, the adjusted R2 in-
creases from 0.34 to 0.59, which means that the GWR provides a better
The vegetation restoration is influenced by many factors that are overall fitting than that of the OLS, and the former has stronger ex-
normally divided into natural and human activities (Li et al., 2011, Li planatory power for the growth of vegetation index than the latter.
et al., 2017; Wang et al., 2017; Qu et al., 2018). To quantitatively se- Meanwhile, the GWR reduces the Sigma, the Residual SS, and the AICc
parate the independent contribution of each factor to the vegetation (an improved version of Akaike information criterion; Anderson and
index increase, this paper further selected the factors in varied aspects Burnham, 2004) by 0.38, 7.65, and 2,483 respectively, indicating that
of climate, topography, socio-economy, and ecological policy based on the regression error of the GWR model is less than the OLS, and the
the previous researches and data experiments. Climatic factors included former can provide more accurate prediction results. Another ad-
the annual mean humidity, annual mean temperature, and annual mean vantage of using the GWR model is that it can detect spatial hetero-
wind speed; topography mainly included terrain elevation and slope; geneity. Table 4 provides the regression results of the GWR, which is of
socio-economic factors included the rural road density and population great significance to spatial differences in the driving force of vegeta-
density; according to ecological project implementation modes, the tion index improvement in different areas.
ecological policy factors were divided into the RFFG, FLC and GLC,
which were expressed as the rate of RFFG (RFFGR), the rate of FLC Table 1
(FLCR), and the rate of GLC (GLCR), respectively. Population density, Change of every land use type from 1999 to 2010 (Area: km2; Rate: %).
RFFGR, GLCR, and FLCR were measured as follows: Year 1999 2010 Changes
Population density = population size / total area of administrative
area, where population = construction land size/per capita construc- Land use type Area Rate Area Rate Area Rate

tion land use. Cultivated Land 8868.65 23.96 4028.64 10.89 −4840.02 −54.57
RFFGR = the area implemented for RFFG / the total administrative Forestland 17863.18 48.27 23067.64 62.33 5204.46 29.14
area. Grassland 9526.15 25.74 9044.08 24.44 −482.08 −5.06
FLCR = the area implemented for FLC / total administrative area. Water Area 185.58 0.50 181.19 0.49 −4.39 −2.37
Construction Land 549.00 1.48 678.05 1.83 129.06 23.51
GLCR = the area implemented for GLC / total administrative area.
Unused Land 14.65 0.04 7.63 0.02 −7.02 −47.94
To eliminate dimensional influence, all variables are normalized Total 37007.22 – 37007.22 – 0.00 –
before OLS/GWR modeling, so that data have the same caliber.

6
D. Zhang, et al. Ecological Indicators 113 (2020) 106278

Table 2
Land use transfer matrix from 1999 to 2010 (Area: km2; Rate: %).
19,992,010 Cultivated Land Forestland Grassland Water Area Construction Land Unused Land Total transfer-in

Area Rate Area Rate Area Rate Area Rate Area Rate Area Rate

Cultivated Land 3943.98 44.47 49.63 0.28 24.92 0.26 7.02 3.78 3.01 0.55 0.07 0.48 84.65
Forestland 4667.27 52.63 17720.01 99.20 674.26 7.08 3.64 1.96 2.46 0.45 0.00 0.00 5347.63
Grassland 198.06 2.23 44.23 0.25 8799.28 92.37 0.72 0.39 0.92 0.17 0.87 5.93 244.80
Water Area 4.27 0.05 2.01 0.01 1.59 0.02 173.05 93.25 0.08 0.01 0.19 1.32 8.13
Construction Land 54.70 0.62 47.00 0.26 25.62 0.27 0.33 0.18 542.53 98.82 7.87 53.72 135.52
Unused Land 0.37 0.00 0.30 0.00 0.49 0.01 0.82 0.44 0.00 0.00 5.65 38.55 1.98
Total transfer-out 4924.67 55.53 143.17 0.80 726.87 7.63 12.53 6.75 6.46 1.18 9.00 61.45 –

(1) The spatial driving effects of climatic factors vegetation recovery in the western and eastern regions, while in the
The humidity (Fig. 6a) and temperature (Fig. 6b) were selected to middle of the study area, it shows a relatively negative relationship
represent the climate differences in the study area. While both of these with the vegetation restoration.
two climatic factors shown significant effects on vegetation restoration, (3) The spatial driving effects of social and economic factors
humidity has a wider positive impact throughout the study area. Here we use the rural road density (Fig. 6e) and the population
(2) The spatial driving effects of topographical factors density (Fig. 6f) to reflect the social and economic factors. The rural
The slope gradient has a positive influence on the vegetation re- road density has a higher and more significant positive effect on ve-
covery in 77.96% of the areas, and it increases from the southeast to the getation restoration in the northwest (Wuqi) and in the southeast (Yi-
northwest, with only 28.14% of them significant (Fig. 6c). Concerning chuan and Huangling), while it has a significant negative influence in
elevation in Fig. 6d, it has a significant positive correlation with the northeast and southwest regions. The regression coefficients of

Fig. 4. Spatial distribution map of vegetation index improvement from 2000 to 2011. Notes: this figure shows the spatial distribution of the vegetation index
improvement, taking villages as samples; the brackets indicate the number of villages included in the range.

7
D. Zhang, et al. Ecological Indicators 113 (2020) 106278

Fig. 5. Estimation result of regression coefficients obtained through OLS modeling. Notes: for the entire regression equation, the adjustment R2 is 0.34, F value is
179.61, and P < 0.01; the two dashed lines denote that the t-value is equal to −1.96 and 1.96 respectively; at 0.05 significance level, t < -1.96 means a
significantly negative correlation, while t > 1.96 represents a positive correlation.

Table 3 Regression coefficients obtained by the GWR model for the three
Accuracy Comparisons between OLS and GWR. ecological measures are compared in Fig. 7. Overall, the RFFGR owns
Model R2 (adjust) Sigma AICc Residual SS
the largest coefficient, followed by the FLCR, and that of the GLCR is
the lowest. The regression coefficient of each ecological policy variable
OLS 0.34 0.07 −8478 18.34 changes with spatial position within a certain range, indicating that the
GWR 0.59 0.45 −10146 10.69 influence of policies has certain spatial heterogeneity.
Improvement 0.25 −0.38 −2483 −7.65

5. Discussion
Table 4
Estimation result of regression coefficients obtained through GWR modeling. 5.1. Main findings

Variable Minimum Lower Median Upper Maximum


Quartile Quartile
Yan'an is a national demonstration zone for ecological construction,
and it is also one of the first places in China to carry out the RFFG
Constant Term −0.587 0.302 0.548 0.903 1.741 program, thus this study takes Yan'an as the research area because of its
Humidity −1.56 −0.042 0.042 0.154 0.568 representative. Based on the results of our analysis, the following
Temperature −0.738 −0.015 0.182 0.347 1.398
findings could be obtained.
Wind speed −1.717 −0.217 −0.091 0.133 1.901
Rural road −0.234 −0.064 −0.029 0.007 0.265
density 5.1.1. Cultivated land was converted into ecological land significantly
Population −0.8 −0.316 −0.083 −0.03 0.258 According to Tables 1 and 2, the area of forest land was increased
density
greatly along with the sharp reduction of cultivated land from 2000 to
Slope gradient −0.174 0.016 0.061 0.107 0.568
Elevation −0.237 −0.048 0.034 0.164 0.565 2010, and the main flow direction of reduced cultivated land was forest
GLCR −0.179 −0.083 −0.047 −0.012 0.163 land and grassland, which indicates that the ecological policies had a
FLCR −0.194 0.022 0.058 0.102 0.295 significant impact on the land use structure, and the ecological land
RFFGR −0.089 0.054 0.081 0.128 0.352 increased significantly. This is consistent with the conclusion of Zhang
et al. (2018a), which concerned with Wuqi County in Yanan.
Notes: detailed analyses can be obtained from Fig. 6.

population density are negative in most of the study area, and the ne- 5.1.2. Natural factors played a fundamental role in vegetation restoration
gative driving strength gradually increases from north to south. Vegetation restoration is a complex system due to the combined
(4) The spatial driving effects of ecological measures effects of climate, topography, society, economy, population, and po-
We used the RFFGR (Fig. 6g), FLCR (Fig. 6h), and GLCR (Fig. 6i) to licies. Since the implementation of RFFG, many scholars have done
reflect ecological measures in the study area. The coefficients of the extensive and in-depth research on vegetation restoration. Some re-
FLCR and RFFGR show similar spatial trends. They have a positive searchers have conducted in-depth studies on the driving forces and
impact on the increase of EVI in most areas, and the relatively high- driving mechanisms of vegetation restoration from different spatial and
value areas are generally distributed in the north-western (Wuqi and temporal scales (e.g. Scholtz et al., 2014; Ndayisaba et al., 2016; Kang
Zhidan), central (Baota), south-western (Huangling), and south-eastern et al., 2017). It is generally believed that vegetation restoration is the
(Yichuan, Yanchang, and Huanglong) regions. The GLCR has a positive result of the interaction of natural conditions and human activities (Hu
impact on the increase of EVI in very few locations, i.e., the northwest et al., 2016; Bao et al., 2017; Qiu et al., 2017). Through the GWR
(Wuqi), north-eastern (Zichang) and south-eastern (Huanglong and modeling, it can be seen from this research that although the intensity
Yichuan) regions. Besides, it is relatively small and only significant in a of the RFFG in northern Yan'an is much greater than that of the south,
small part of the study area. the vegetation index in the south is higher, which indicates that natural
factors play a dominating role in vegetation restoration. This can also

8
D. Zhang, et al. Ecological Indicators 113 (2020) 106278

(caption on next page)

9
D. Zhang, et al. Ecological Indicators 113 (2020) 106278

Fig. 6. Spatial distribution maps of regression coefficients and significances for all independent variables. Notes: (a) to (i) are the spatial distribution maps of the local
regression coefficients and the significances for all independent variables, taking village as a basic unit. At 0.05 significance level, t < = −1.96 means significantly
negative correlation, t > 1.96 means significantly positive correlation, and −1.96 < t < = 1.96 means not significant; the numbers in parentheses indicate the
number of villages included.

be verified from the analysis of the OLS. We can see from Fig. 5 that conditions are abnormally dry and hot. In these locations, the RFFG
humidity, temperature, and population density have a much greater should be controlled at a certain scale to avoid further moisture loss and
influence on the vegetation restoration than that of the policy measures. to increase the survival rate of trees.

5.2. Existing inadequacies and outlook


5.1.3. GWR provides more useful information than the traditional OLS
model In this study, both OLS and GWR models were applied to analyze the
While the OLS model provides us an overall comparison among the factors contributing to vegetation restoration in Yan'an, China, and it
regression coefficients, GWR can compare the values of regression was found that the latter could provide more useful information for
coefficients of different variables across the whole study area, which policy-makers. However, Due to the limitation of data source and data
makes it possible to study the spatial heterogeneity of driving forces on accuracy, the following inadequacies may exist. First of all, this study
the vegetation restoration. Results of the GWR model show that the analyzed the vegetation restoration and its driving forces in the study
driving forces on the vegetation restoration do demonstrate the spatial area through the comparison of two time-section data, and the period is
difference, which should be the result of geographical differentiation from 1999 to 2010. The subsequent changes in the effects of vegetation
and various factors combination effect. restoration require further research in the future. Secondly, the basic
(1) In terms of meteorological variables, except for Yanchang and sample unit of this study is the village. Although this guarantees as
Yichuan along the Yellow region and the southeastern part of Wuqi, the much sample capacity as possible and reduces the internal variation of
temperature has a significantly positive effect on the vegetation re- the sample, it brings difficulties to the collection of variables. For ex-
storation with an increasing trend from southeast to northwest. The ample, there may be differences in ecological protection efforts, and
relatively low-temperature environment makes vegetation growth ex- this is not reflected in the model. It can be obtained through field
tremely sensitive to temperature, and thus an appropriate increase in surveys in the future, but the workload would be large. Finally, the
temperature can extend the growth period of vegetation moderately spatial resolution of the vegetation coverage data from MODIS is
(Sun et al., 2015). Water is the main limiting factor in most regions of 250 m × 250 m, which is relatively a low accuracy. Although the
Yan'an. Thus, under the condition of limited water resources, water Landsat TM image owns a spatial resolution of 30 m × 30 m which is
conservancy projects should be constructed to save and intensively use more accurate, its temporal resolution is poor. In the future, it may be
water resources. At the same time, in the process of the RFFG, the considered to use the spatial and temporal adaptive reflectance fusion
government should determine the moderate scale and select suitable model, which can guarantee both temporal and spatial resolution (Zhu
species according to conditions of water and heat to avoid serious soil et al., 2010), to improve the accuracy of the vegetation coverage data.
desiccation.
(2) Terrain slope not only produces local microclimate which plays 6. Conclusions
an important role in vegetation growth but affects vegetation restora-
tion by affecting the intensity of human activities. As is well known to This paper analyzed the effects of land use/cover change and ve-
us, the relatively flat terrain is more conducive to human production getation restoration before and after the ecological engineering in
activities. In the eastern part of the study area with relatively low ter- Yan'an. Based on this research, driving forces and their spatial differ-
rain slope, the lower the slope, the better the vegetation growth. ences for the vegetation restoration are analyzed through the combined
However, in the western region (Wuqi, Zhidan, and Ansai), the higher applications of the OLS and GWR models. The following conclusions
the slope, the better the vegetation recovered. The possible reason is can be obtained from this research.
that in relatively flat areas, vegetation is mainly restored naturally;
while in hilly areas, vegetation restoration is mainly through artificial (1) Due to the implementation of ecological engineering, the land use
afforestation, and the greater the terrain slope, the higher the RFFGR. structure changed dramatically in Yan'an. The forest and grass
(3) The rural road density has a higher and more significant positive coverage increased from 74.0% in 2000 to 86.8% in 2010.
effect on vegetation restoration in the northwest (Wuqi) and southeast (2) Although the vegetation index was generally promoted as a whole
(Yichuan and Huangling) of the study area, and the possible reason is in the study area, significant spatial variation and trends could be
that the larger rural road density provides more convenient traffic observed. The effect of vegetation restoration in the southern region
conditions for the implementation and maintenance of ecological en- was better than that in the northern region, which indicated that
gineering. However, it demonstrates a significant negative influence on natural conditions such as water and temperature play a key role in
the vegetation restoration in the northeast and southwest regions, vegetation restoration.
which is mainly because of the great disturbance caused by human (3) From the perspective of the whole study area, the two meteor-
activities near rural roads. The different roles of rural road density in ological factors, i.e. the temperature and the humidity, and the
different regions suggest that we should adjust different aspects of population density showed much higher contributions to the ve-
human activities according to local conditions, in order to achieve the getation restoration than others, while policy factors also played a
most favorable combination of factors for vegetation restoration. role.
(4) The FLCR and RFFGR in most areas have positive impacts on the (4) Compared to the OLS model, the GWR model could provide more
increase of EVI, and the villages with high-value coefficients are gen- detailed and specific information for the policymakers since it de-
erally distributed in the north-western (south-eastern part of Wuqi and tected not only the driving forces but their spatial variability. The
west part of Zhidan), central (Baota), south-western (Huangling) and GLC showed a significant positive impact on vegetation restoration
south-eastern (Yichuan, Yanchang, and Huanglong), where the scale of in the southeast and northeast, and the RFFR and FLC have a sig-
the ecological projects can be moderately expanded in future. However, nificant contribution to vegetation in most areas, which decreases
the FLCR and RFFGR have negative impacts on the vegetation re- from the center to all around. The governments could maximize the
storation in the northwest of Wuqi, and the junction of Ansai and effectiveness of ecological policies by adjusting their strength and
Zhidan, where the strength of the FLCR is high, but the local climatic

10
D. Zhang, et al. Ecological Indicators 113 (2020) 106278

Fig. 6. (continued)

structure according to the local conditions.

The ecological policies and environmental factors show different


coupling effects in contributing to the vegetation restoration in dif-
ferent locations, which indicates that spatial variations should be fully
considered when formulating ecological policy, as well as evaluating its
implementation effect. There are great limitations to evaluate the ap-
plication effect of ecological policy leaving aside the regional differ-
ences because the OLS model isolates the effects of ecological policies
from other factors and their coupling effects in different regions in
achieving vegetation restoration goals. In comparison, GWR takes these
conditions into account, and thus it is more accurate to use GWR than
OLS for the evaluation of similar policies.
Fig. 7. Regression coefficients obtained by GWR model for ecological measures.

11
D. Zhang, et al. Ecological Indicators 113 (2020) 106278

CRediT authorship contribution statement the radiometric and biophysical performance of the MODIS vegetation indices.
Remote Sens. Environ. 83 (1–2), 195–213.
Jiang, C., Wang, F., Zhang, H., Dong, X., 2016. Quantifying changes in multiple ecosystem
Daojun Zhang: Conceptualization, Methodology, Writing - review services during 2000–2012 on the Loess Plateau, China, as a result of climate
& editing. Qiqi Jia: Data curation, Resources, Writing - original draft. variability and ecological restoration. Ecol. Eng. 97, 258–271.
Peng Wang: Data curation, Writing - review & editing. Jinting Zhang: Jin, G., Li, Z., Wang, Z., Chu, X., Li, Z., 2015. Impact of land-use induced changes on
agricultural productivity in the Huang-Huai-Hai River Basin. Phys. Chem. Earth,
Writing - review & editing. Xianhui Hou: Data curation. Xiandong Li: Parts A/B/C. 79, 86–92.
Writing - review & editing. Wangjun Li: Writing - review & editing. Justice, C., Gutman, G., Vadrevu, K.P., 2015. NASA land cover and land use change
(LCLUC): An interdisciplinary research program. J. Environ. Manage. 148
(148C), 4–9.
Acknowledgments Kang, C., Zhang, Y., Wang, Z., Liu, L., Zhang, H., Jo, Y., 2017. The driving force analysis
of NDVI dynamics in the Trans-Boundary Tumen River Basin between 2000 and
This study benefited from joint financial support by the Programs of 2015. Sustainability 9 (12), 2350.
Kuriqi, A., Pinheiro, A.N., Sordo-Ward, A., Garrote, L., 2019a. Influence of hydrologically
National Natural Science Foundation of China (Nos. 41602336,
based environmental flow methods on flow alteration and energy production in a
71503200, and 71603209), China Postdoctoral Science Foundation run-of-river hydropower plant. J. Cleaner Prod.
(Nos. 2016M592840 and 2017T100773), Shaanxi Provincial Natural Kuriqi, A., Pinheiro, A.N., Sordo-Ward, A., Garrote, L., 2019b. Flow regime aspects in
Science Foundation (No. 2017JQ7010), and Fundamental Research determining environmental flows and maximising energy production at run-of-river
hydropower plants. Appl. Energy 256, 113980.
from Northwest A&F University in 2017 (No. 2017RWYB08). Li, H., Liu, G., Fu, B., 2011. Response of vegetation to climate change and human activity
based on NDVI in the Three-River Headwaters region. Acta Ecologica Sinica. 31 (19),
References 5495-5504 in Chinese with English abstract.
Li, H., Liu, Y., 2016. Neighborhood socioeconomic disadvantage and urban public green
spaces availability: A localized modeling approach to inform land use policy. Land
Ahmed, M.A.A., Abd-Elrahman, A., Escobedo, F.J., Cropper, W.P., Martin, T.A., Timilsina, Use Policy 57, 470–478.
N., 2017. Spatially-explicit modeling of multi-scale drivers of aboveground forest Li, J., Peng, S., Li, Z., 2017. Detecting and attributing vegetation changes on China’s Loess
biomass and water yield in watersheds of the Southeastern United States. J. Environ. Plateau. Agric. For. Meteorol. 247, 260–270.
Manage. 199, 158–171. Liu, Y., Li, Y., Li, S., Motesharrei, S., 2015. Spatial and temporal patterns of global NDVI
Akaike, H., 1974. A new look at the statistical model identification. IEEE Trans. Autom. trends: Correlations with climate and human factors. Remote Sensing. 7 (10),
Control 19 (6), 716–723. 13233–13250.
Alix-Garcia, J., Munteanu, C., Zhao, N., Potapov, P.V., Prishchepov, A.V., Radeloff, V.C., Nakaya, T., 2016. GWR4.09 user manual. WWW Document [online]. Available from:
Krylov, A., Bragina, E., 2016. Drivers of forest cover change in Eastern Europe and https://raw.githubusercontent.com/gwrtools/gwr4/master/GWR4manual_409.pdf
European Russia, 1985–2012. Land Use Policy. 59, 284–297. [accessed on 16 February 2017].
Ali, R., Kuriqi, A., Abubaker, S., Kisi, O., 2019. Hydrologic alteration at the upper and Ndayisaba, F., Guo, H., Bao, A., Guo, H., Karamage, F., Kayiranga, A., 2016.
middle part of the Yangtze river, China: towards sustainable water resource man- Understanding the spatial temporal vegetation dynamics in rwanda. Remote Sens. 8
agement under increasing water exploitation. Sustainability. 11 (19), 5176. (2), 129.
Anderson, D.R., Burnham, K., 2004. Model selection and multi-model inference, Second. Qiu, B., Chen, G., Tang, Z., Lu, D., Wang, Z., Chen, C., 2017. Assessing the Three-North
Springer-Verlag, NY, pp. 63. Shelter Forest Program in China by a novel framework for characterizing vegetation
Bao, A., Huang, Y., Ma, Y., Guo, H., Wang, Y., 2017. Assessing the effect of EWDP on changes. ISPRS J. Photogramm. Remote Sens. 133, 75–88.
vegetation restoration by remote sensing in the lower reaches of Tarim River. Ecol. Qu, S., Wang, L., Lin, A., Zhu, H., Yuan, M., 2018. What drives the vegetation restoration
Ind. 74, 261–275. in Yangtze River Basin, China: Climate change or anthropogenic factors? Ecol. Ind.
Bégué, A., Vintrou, E., Ruelland, D., Claden, M., Dessay, N., 2011. Can a 25-year trend in 90, 438–450.
Soudano-Sahelian vegetation dynamics be interpreted in terms of land use change? A Robinson, D.P., Lloyd, C.D., McKinley, J.M., 2013. Increasing the accuracy of nitrogen
remote sensing approach. Global Environ. Change 21 (2), 413–420. dioxide (NO2) pollution mapping using geographically weighted regression (GWR)
Bejarano, M.D., Sordo-Ward, A., Gabriel-Martin, I., Garrote, L., 2019. Tradeoff between and geostatistics. Int. J. Appl. Earth Obs. Geoinf. 21, 374–383.
economic and environmental costs and benefits of hydropower production at run-of- Scholtz, R., Kiker, G.A., Smit, I.P.J., Venter, F.J., 2014. Identifying drivers that influence
river-diversion schemes under different environmental flows scenarios. J. Hydrol. the spatial distribution of woody vegetation in Kruger National Park. South Africa.
572, 790–804. Ecosphere. 5 (6), 1–12.
Brunsdon, C., Fotheringham, A.S., Charlton, M.E., 1996. Geographically weighted re- Shen, Q., Gao, G., Lü, Y., Wang, S., Jiang, X., Fu, B., 2017. River flow is critical for
gression: a method for exploring spatial nonstationarity. Geogr. Anal. 28 (4), vegetation dynamics: Lessons from multi-scale analysis in a hyper-arid endorheic
281–298. basin. Sci. Total Environ. 603, 290–298.
Chen, H., Shao, L., Zhao, M., Zhang, X., Zhang, D., 2017. GLC programs, vegetation re- Shi, H., Shao, M., 2000. Soil and water loss from the Loess Plateau in China. J. Arid
habilitation and spatial dependency in Inner Mongolia, China. Land Use Policy 64, Environ. 45 (1), 9–20.
429–439. Sun, W., Song, X., Mu, X., Gao, P., Wang, F., Zhao, G., 2015. Spatiotemporal vegetation
Cleveland, W.S., 1979. Robust locally weighted regression and smoothing scatterplots. J. cover variations associated with climate change and ecological restoration in the
Am. Stat. Assoc. 74 (368), 829–836. Loess Plateau. Agric. For. Meteorol. 209, 87–99.
Du, S., Wang, Q., Guo, L., 2014. Spatially varying relationships between land-cover Tang, X., 2004. Five-year history of grain for green project. China's Forestry 8, 13–16.
change and driving factors at multiple sampling scales. J. Environ. Manage. 137 (4), Tian, Y., Yin, K., Lu, D., Hua, L., Zhao, Q., Wen, M., 2014. Examining land use and land
101–110. cover spatiotemporal change and driving forces in beijing from 1978 to 2010. Remote
Ellison, D., Morris, C.E., Locatelli, B., Sheil, D., Cohen, J., Murdiyarso, D., Gaveau, D., Sens. 6 (11), 10593–10611.
2017. Trees, forests and water: cool insights for a hot world. Global Environ. Change Wang F, An P, Cai L, et al., 2015. Monitoring effects of ecosystem restructuring project for
43, 51–61. returning grain plots to forestry in Wuchuan county of Inner Mongolia based on RS
FAO. Global Forest Resources Assessment 2000 Main Report. Rome, 2010: 140. and GIS. Transactions of the Chinese Society of Agricultural Engineering. 31 (11),
Fotheringham, A.S., Brunsdon, C., Charlton, M.E., 1996. The geography of parameter 269-277 in Chinese with English abstract.
space: an investigation of spatial non-stationarity. Int. J. Geogr. Inf. Syst. 10, Wang, H., Liu, G., Li, Z., Ye, X., Fu, B., Lü, Y., 2017. Analysis of the driving forces in
605–627. vegetation variation in the Grain for Green program region, China. Sustainability 9
Gao H, Pang G, Li Z, et al., 2017. Evaluating the potential of vegetation restoration in the (10), 1853.
Loess Plateau. Acta Geographica Sinica. 72 (5), 863-874 in Chinese with English Xu, X., Zhang, D., Zhang, Y., Yao, S., Zhang, J., 2020. Evaluating the vegetation re-
abstract. storation potential achievement of ecological projects: a case study of Yan’an, China.
Guo, R., Li, F., He, W., Yang, S., Sun, G., 2010. In: Spatial and Temporal Variability of Land Use Policy 90, 104293.
Annual Precipitation During 1958–2007 in Loess Plateau, China. Springer, Berlin, Yan, L., He, R., Kašanin-Grubin, M., Luo, G., Peng, H., Qiu, J., 2017. The dynamic change
Heidelberg, pp. 551–560. of vegetation cover and associated driving forces in Nanxiong Basin, China.
Han, Y., Zhu, W., Li, S., 2016. Modelling relationship between NDVI and climatic factors Sustainability 9 (3), 443.
in China using Geographically Weighted Regression. Acta Scientiarum Naturalium Yin, H., Pflugmacher, D., Li, A., Li, Z., Hostert, P., 2018. Land use and land cover change
Universitatis Pekinensis. 52 (6), 1125-1133 in Chinese with English abstract. in Inner Mongolia-understanding the effects of China's re-vegetation programs.
Hou, X., Liu, J., Zhang, D., 2019. Regional sustainable development: the relationship Remote Sens. Environ. 204, 918–930.
between natural capital utilization and economic development. Sustainable Dev. 27 Yin, R., 2010. Assessing China's ecological restoration programs. Environ. Manage. 45 (3),
(1), 183–195. 425.
Hu, Y., Peng, J., Yuan, S., Shu, X., Jiang, S., Pu, Q., Ma, K., Yuan, C., Chen, G., Xiao, H., Zhang, D., Jia, Q., Xu, X., Yao, S., Chen, H., Hou, X., 2018a. Contribution of ecological
2016. Influence of ecological restoration on vegetation and soil microbiological policies to vegetation restoration: a case study from wuqi county in Shaanxi province,
properties in Alpine-cold semi-humid desertified land. Ecol. Eng. 94, 88–94. China. Land Use Policy. 73, 400–411.
Huang, C., Goward, S.N., Masek, J.G., Thomas, N., Zhu, Z., Vogelmann, J.E., 2010. An Zhang, D., Jia, Q., Xu, X., Yao, S., Chen, H., Hou, X., Jin, G., 2019a. Assessing the co-
automated approach for reconstructing recent forest disturbance history using dense ordination of ecological and agricultural goals during ecological restoration efforts: a
Landsat time series stacks. Remote Sens. Environ. 114 (1), 183–198. case study of Wuqi County, Northwest China. Land Use Policy. 82, 550–562.
Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G., 2002. Overview of Zhang, D., Xu, X., Yao, S., Zhang, J., Hou, X., Yin, R., 2019b. A novel similar habitat

12
D. Zhang, et al. Ecological Indicators 113 (2020) 106278

potential model based on sliding-window technique for vegetation restoration po- Zhao, A., Liu, X., Zhu, X., Pan, Y., Chen, S., 2016. Spatiotemporal analyses and associated
tential mapping. Land Degrad. Dev. (published online). driving forces of vegetation coverage change in the Loess Plateau. China
Zhang, D., Ren, N., Hou, X., 2018b. An improved logistic regression model based on a Environmental Science. 36 (5), 1568-1578 in Chinese with English abstract.
spatially weighted technique (ILRBSWT v1. 0) and its application to mineral pro- Zhao, R., Chen, Y., Shi, P., Zhang, L., Pan, J., Zhao, H., 2013. Land use and land cover
spectivity mapping. Geosci. Model Dev. 11 (6), 2525. change and driving mechanism in the arid inland river basin: a case study of Tarim
Zhang, T., Gong, W., Wang, W., Ji, Y., Zhu, Z., Huang, Y., 2016. Ground level PM2. 5 River, Xinjiang, China. Environ. Earth Sci. 68 (2), 591–604.
estimates over China using satellite-based Geographically Weighted Regression Zhu, X., Chen, J., Gao, F., Chen, X., Masek, J.G., 2010. An enhanced spatial and temporal
(GWR) models are improved by including NO2 and enhanced vegetation index (EVI). adaptive reflectance fusion model for complex heterogeneous regions. Remote Sens.
Int. J. Environ. Res. Public Health 13 (12), 1215. Environ. 114 (11), 2610–2623.

13

You might also like