Exploring the Impact of Grain-for-Green Program on Trade-Offs and Synergies among Ecosystem Services in West Liao River Basin, China

Abstract

Natural ecosystems of the West Liao River basin (WLRB) in northeast China have been damaged by both natural and human factors from the 1990s. Since 2000, China’s Grain-for-Green Program (GFGP) has been widely adopted with the aim of improving ecosystem services. An accurate evaluation of the eco-hydrological effects for policy implementation is essential to provide references for further restoration of ecosystem services. This study quantified and characterized the ecosystem services and their trade-offs/synergies using models and statistical methods in the WLRB from 1990 to 2020. Moreover, the impact of key drivers on ecosystem services was evaluated by the difference-in-differences model. Among them, the study mainly investigated how GFGP affects ecosystem services. The results confirmed that the water yield, carbon sequestration, habitat quality, and total ecosystem service of the WLRB decreased in the pre-GFGP period (1990–2000). However, this tendency was reversed in the regions where the GFGP was implemented during the period of 2001–2020. Furthermore, a synergistic relationship was shown among carbon sequestration, soil conservation, and habitat quality. Additionally, there were tradeoffs between water yield and the other three ecosystem services, especially in mountain areas. The GFGP could restore carbon sequestration, habitat quality, and total ecosystem services by 1.3%, 2.1%, and 0.6%, respectively. Nevertheless, GFCP may enlarge the tradeoff and imbalance between water yield and habitat quality. Results highlight the need for the governance of ecosystem protection and suggest natural restoration in the mountain area for maintaining water yield and helping ecosystem restoration. Timely adjustment of the policy implementation areas is the key to improving and balancing multiple ecosystem services in the future.

1. Introduction

Ecosystem services (ESs) are the basis for human production and life to obtain matter and energy from the ecosystem [1,2,3]. However, the rapid socioeconomic development over the past centuries has led to an imbalance among ESs [4]. For example, researchers warn that China’s total ESs value has declined and fluctuated, especially in northern China from 1992 to 2018 [5]. The revegetation of degraded lands is expanding globally with the UN Decade on Ecosystem Restoration to alleviate the conflict between human development and ecosystem conservation [6]. Chinese authorities have taken several measures aimed at mitigating the ecological crisis. These measures cover industrial, urban, forestry, and agricultural activities. Among all these efforts, the Grain-for-Green Program (GFGP) has made remarkable achievements [7,8,9]. This study selects the West Liao River basin (WLRB) as the study area, which is a typical agro-pastoral zone and ecologically fragile area in northern China [10]. Human overgrazing and farmland irrigation have caused ecological damage, such as vegetation degradation, land sanding, and water shortage in the past 30 years [10]. Therefore, understanding the impact of GFGP on ESs in the WLRB is essential for improving management strategies to mitigate human activities impacts and realize sustainable development goals.

Scholars have classified ESs into four categories (provision, regulating, cultural, and support services) [6]. Specifically, due to the water resources being constrained in the WLRB, the increase in agricultural water consumption would largely reduce the water availability of the ecosystem [11]. Therefore, the water yield (WY) is explored in our study. Regulating services [12], including carbon sequestration (CS), soil conservation (SC), and habitat quality (HQ), also play vital roles in WLRB to support ecological and socioeconomic development [10,13]. With the improvement of remote sensing techniques and modeling methods, ESs are evaluated using features such as topography, land use and land cover (LULC), and vegetation greenness inverse from remote sensing. ES evaluation approaches have become more accurate in both temporal and spatial scales [14,15]. A variety of methods have been invented to quantify ESs [15]. Generally, the InVEST (Integrated Valuation of Ecosystem Services and Tradeoffs) model [16], ARIES (Artificial Intelligence for Ecosystem Services) model [17], and SolVES (Social Values for Ecosystem Services) model are the most applied. At present, the ARIES and SolVES models are not fully developed and are mainly applicable in the United States [18]. The InVEST model has successfully been applied to analyze the ESs and their driving factors for carrying out better environmental protection policies worldwide [19]. Therefore, the InVEST model was chosen to evaluate the ESs in our study.

It should be noted that most ESs do not change independently; the synergy and tradeoff relationship promote the interaction among different ESs [20,21,22,23]. The synergy relationship is the joint enhancement of multiple ecosystem services, whereas the tradeoff relationship happens when one ES increases while other ESs cross-balance and decline [24]. Using correlation analysis, tradeoffs among ecosystem services were detected under ecological restoration. Studies have confirmed that improved SC and CS increased water consumption and intensified the shortage of water [25,26]. The correlation method cannot evaluate the spatial tradeoffs. The root-mean-square error (RMSE) is considered a simple but effective method for quantifying tradeoffs between two or more ESs [27,28].

The ESs are significantly linked to environmental conditions, in which the climate and LULC are the principal factors for ES change [29,30,31,32]. For example, the main factor for water yield and soil erosion was identified as precipitation in the Hengduan Mountain region of China [33]. During the last 30 years, the LULC has changed vastly to fulfill the residents’ needs for a better life [34,35]. The considerable change in LULC has also directly impacted carbon emissions, resulting in global climate change [36,37]. Researchers have confirmed that urbanization’s impact on carbon dioxide emission steadily increased during 1990–2016, and energy utilization increased from 571.44 to 4358.19 million tons [38]. The wheat production rate may decrease by 3.8% (supporting services declined) under future global warming in northern China, which may drive a national food crisis [39].

The GFGP is a key ecological restoration measure to address land degradation through changing LULC in China. The GFGP started in 1999 to cease sloping farmland cultivation and convert farmland to grassland and forest, aiming to alleviate the threat of desertification [40]. China’s largest “Three-North Shelterbelt Program” has achieved the expected reduction in local land desertification and soil erosion. The artificial sparsely forested grassland restoration approach has effectively restored Horqin sandy land [41]. However, adverse consequences have also occurred during ecological restoration. For example, GFGP aggravates water conflicts and increases the risk of future large-scale sustainable interventions. However, the collective impact of climate change and human activities on ESs is still inconclusive [4,42], especially in the northern agro-pastoral zone.

Researchers have made efforts to analyze how the GFGP has affected environmental restoration [26,43,44,45,46]. A widely used approach to analyze the effect of GFGP is using scenario simulation. Liu et al. [47] identified the ESs that increased during 2000–2018 by revegetation projects in the agro-pastoral zone of northern China, according to different LULC prediction scenarios. Specifically, the GFGP has been proven to improve ESs by increasing vegetation cover and, thus, reducing soil erosion. The alpine grassland restoration management scenario has been used to simulate the strategies for grassland productivity improvement in the Qilian Mountains [48]. The ES estimation in these studies was mainly carried out according to LULC without considering the influence of climate, which may not present the actual change in ESs by the GFGP [49,50]. Various statistical methods, such as the geographical detector model (GDM) [51], simultaneous equation model (SEM) [52], and machine learning [29], were used to quantify the effects of different environmental factors on ES. Among the various methods, the difference-in-differences (DID) method is considered effective for obtaining the net policy of environment management [53,54,55]. This study applies the DID method to separate the effect of GFGP in ESs.

By combining the InVEST model with statistical modeling methods to quantify the environmental effects of policies, this study explored the impact of GFGP on tradeoffs and synergy among ESs in the WLRB. Results can provide scientific suggestions to policymakers for optimizing future regional ecological restoration policies. This study focused on (1) evaluating the variation of ESs over the last 30 years in the WLRB, (2) estimating the tradeoff among ESs under a changing environment, and (3) demonstrating the feasibility of GFGP for improving ESs in the WLRB.

2. Study Area and Data

2.1. Study Area

The WLRB (115°36′–125°17′E, 40°07′–46°28′N) is located at the eastern edge of northern China’s agricultural and pastoral interlacing zone, with a substantial agricultural and livestock production base. The WLRB is the largest tributary of the Liao River Basin (Figure 1), with elevation ranging from 110 to 2021 m, and a total area of 1.31 × 10⁵ km². Mountains surround the northern, western, and southern parts of the WLRB, and the terrain becomes lower from west to east, which finally transitions to the Liao River Plain.

The climate has strong continental characteristics. For example, the rain and heat appear simultaneously, with uneven seasonal distribution; about 70% of total annual precipitation occurs in summer. The mean annual precipitation, temperature, and actual evapotranspiration from 1990 to 2020 were 486.4 mm, 6.4 °C, and 460.8 mm, respectively. As shown in Figure 2a–c, the meteorological conditions fluctuated during the last 30 years, while precipitation and temperature presented an overall rising trend. The maximum and minimum annual precipitation decreased from 1990 (518.0 mm and 441.6 mm) to 2000 (348.0 mm and 329.0 mm). The maximum and minimum air temperature appeared in 2015 and 2010. The vegetation condition reflected by NDVI (Figure 2d) showed a greening trend in the WLRB. The average population density (Figure 2e) of the study area increased by 28.5%, and the GDP (Figure 2f) expanded 60-fold. Importantly, the study area is also a specific ecologically sensitive and fragile area in Northern China. These unique characteristics make food and ecological security crucial to the region.

2.2. Data

The 90 m resolution digital elevation model (DEM) used in this study was obtained from the Aster Global Digital Elevation. Monthly meteorological data with 1000 m resolution were provided by the National Tibetan Data Center (https://www.tpdc.ac.cn, accessed on 7 March 2023) [56], including annual precipitation, temperature, and solar radiation. The LULC maps were obtained from the Resource and Environment Data Center (RESDC) with a 1000 m resolution. The LULC was reclassified into seven types: cropland, forestland, shrubland, grassland, water, urban land, and unused land. The socioeconomic data, including gridded datasets of China’s GDP and population density, as well as the normalized difference vegetation index (NDVI), at 1000 m resolution, were also obtained from RESDC. Moreover, the root depth was provided by the National Cryosphere Desert Data Center based on the Harmonized World Soil Database (HWSD) (v1.1) [57]. All data were transformed and resampled to the Lambert coordinate system with a spatial resolution of 1000 m.

3. Methods

3.1. Landcover Change Detection

The LULC in the WLRB from 1990 to 2020 is shown in Figure 3a, showing that grassland and farmland were widely distributed. Farmland increased from 26.7% to 30.2%, while grassland decreased from 47.0% to 43.3%, during the study period. The proportion of forest, shrub, and urban land increased, while that of water and unused land decreased. The transition in LULC mainly occurred from 1990 to 2000. After 2000, a restricted policy was published to limit farmland expansion in this region by the Chinese government, which successfully slowed the expansion of unused land and degradation of grassland.

3.2. Quantifying Ecosystem Services

Referring to previous studies, the environmental status of WLRB, local policies, and data availability, we selected four types of ESs to conduct the study: (1) water yield (WY), (2) carbon sequestration (CS), (3) soil conservation (SC), and (4) habitat quality (HQ). These four ESs were estimated by InVEST 3.9.0 in 1990, 1995, 2000, 2005, 2010, 2015, and 2020, by the annual water yield module, carbon storage, and sequestration module, sediment delivery ratio (SDR) module, and habitat quality module of InVEST [58]. The details of quantifying the ESs are shown in the

Table 1. Methods for quantifying ESs.

ESs	Description	Mathematical Expression
Water yield (WY)	Water produced by a watershed and arriving in streams	$Y\left(i\right)=\left(1-\frac{AET\left(i\right)}{P\left(i\right)}\right)\times P\left(i\right),$ where $Y\left(i\right)$, $AET\left(i\right)$, and $P\left(i\right)$ is the annual water yield, actual evapotranspiration, and precipitation of pixel $i$
Carbon sequestration (CS)	The amount of carbon currently stored in a landscape or the amount of carbon sequestered over time	$C_{total,i}=C_{above,i}+C_{below,i}+C_{dead,i}+C_{soil,i},$ where $C_{above,i}$, $C_{below,i}$, $C_{soil,i}$, and $C_{dead,i}$ are the carbon pools of above-ground biomass, below-ground biomass, soil organic matter, and dead organic matter
Soil conservation (SC)	Erosion control ability of the ecosystem to prevent soil loss and the ability to store and maintain sediment	$E_i={RKLS}_i-{USLE}_i=R_i\times K_i\times {LS}_i\times \left(1-C_i\times P_i\right),$ where $E_i$ is the soil conservation capacity, ${RKLS}_i$ is $R_i$ is the rainfall erosivity (MJ mm(ha·h·yr)−1), $K_i$ is the soil erodibility (ton·ha·h(MJ·ha·mm)−1), ${LS}_i$ is a slope length gradient factor (unitless), $C_i$ is a cover-management factor (unitless), and $P_i$ is a support practice factor (unitless) in pixel i [58]
Habitat quality (HQ)	Ability to provide resources and environmental conditions for the survival and development of species or populations	$Q_i=H\left[1-\left(\frac{D_i^z}{D_i^z+k^z}\right)\right],$ where $Q_i$ is the habitat quality in pixel i, H is the habitat suitability of different types of LUCC, $D_i^z$ and $k^z$ are the level of threat and half-saturation constant, and Z is the implicit parameter of the model [60,61]

. The data inputs included topography and climate conditions, such as the LULC map, digital elevation model, meteorological data, soil properties, biophysical table, and watershed boundary. The biophysical table was in line with studies in the adjacent watershed [43,59].

The ESs at a 1 km² scale were upscaled to the subbasin scale to prevent systematic mistakes and achieve zonal management. For each subbasin, a min–max normalization was applied to make different ESs comparable. The sum of normalized ESs was used to represent the subbasins’ total ecosystem service (TES).

3.3. Assessing the Relationship among Ecosystem Services

Pearson’s correlation analysis and partial correlation analysis were applied to evaluate the tradeoffs and potency of synergy amongst ESs. The partial correlation analysis considers the influence of potential factors on the correlation of study subjects [62]. The partial correlation coefficient ($r_{xy·z}$) between variables x and y was calculated as follows to exclude the effect of the potential variable z:

$$r_{xy·z}=\frac{r_{xy}-r_{xz}{r}_{yz}}{\sqrt{\left(1-r_{xz}^2\right)\left(1-r_{yz}^2\right)}},$$

(1)

where $r_{xy}$, $r_{xz}$, and $r_{yz}$ denote the correlation coefficients between x and y, between x and z, and between y and z, respectively.

To further quantify the relationship of tradeoff variables, the RMSE was employed as the tradeoff coefficient [27]. The tradeoff benefit can be determined from the relative location of ESs [25]. The RMSE can then be derived as

$$RMSE=\sqrt{\frac{1}{n-1}\sum _1^n{({ES}_i-\overline{ES})}^2},$$

(2)

where ${ES}_i$ is the standardized value of ES i, and $\overline{ES}$ is the mean ES.

3.4. Identifying the Effect of Driving Factors via PSM-DID

The difference-in-differences (DID) model, which is frequently used to assess the outcomes of policies, has been proven effective in capturing the net effect of policies [53,54]. In this study, ESs in each subbasin were utilized as a quasi-experiment, and the DID model was applied to assess how the GFGP affected ESs. Subbasins with implemented policies were categorized as treatment groups, and those without implemented policies were control groups. This study tried to overcome sample selection bias and endogeneity via propensity score matching (PSM) before setting up the DID model. Considering that the reforestation policy was gradually implemented after 2000, the continuous DID model could be constructed as follows:

$$Y_{it}=\alpha +\sum _{\tau =-M}^N{\beta }_{\tau }{Policy}_{i,t-\tau }+\gamma X_{it}+{\lambda }_i+v_t+{\epsilon }_{it},$$

(3)

where i and t are the different subbasins and years, $Y_{it}$ is the dependent variable, and ${Policy}_{i,t-\tau }$ is the core explanatory variable that indicates the implementation status of the policy (${Policy}_{i,t-\tau }$ = 1 in the year after the policy is implemented; otherwise, ${Policy}_{i,t-\tau }$ = 0). ${\beta }_{\tau }$ statistically represents the net effect of the policy on $Y_{it}$. $X_{it}$ is a control variable. A two-way fixed-effects approach was applied to control the individual effect (${\lambda }_i$) of the geographic element. The year effect ($v_t)$ was considered to control the effects variables that affected all subbasins in a given year. ${\epsilon }_{it}$ is the random error.

The PSM-DID models were built to separate the driving factors of ESs and their tradeoffs. Propensity score matching (PSM) was first applied to reduce sample selection bias. This study removed 3% of the data to ensure no significant difference between the two groups before policy implementation. More details of PSM are provided in Appendix A. Multivariable regression analysis with a two-way fixed-effects method (with individual and period) was used to create time-varying DID models. The GFGP was the core explanation variable in the regression analysis to identify each ES and its tradeoffs. Additionally, other driving factors, including precipitation, temperature, evapotranspiration, NDVI, GDP, population, and urbanization area, were considered as control variables.

4. Results

4.1. Temporal and Spatial Variation of Ecosystem Services

As shown in Figure 4 and Figure 5, the temporal and spatial distribution of these four ESs differed significantly. The annual WY ranged from 0 to 394.9 mm, and the average WY was 45.8 mm (Figure 5a). The distribution of WY decreased from southeast to northwest. Therefore, only 10.6% of precipitation was converted into runoff, and the remaining 89.4% of precipitation returned to the atmosphere through evapotranspiration. The annual CS ranged from 8953 to 46,844 t/km² with an average value of 19,001.8 t/km² (Figure 5b). The lowest CS appeared in the central Horqin sandy area, while the area with higher CS was spatially consistent with the woodland distribution in the southwest of the study area. The SC presented the most significant spatial characteristic among the four ESs (Figure 5c). The annual average SC was 1734.2 t/hm², reaching 316,970 t/hm² near upstream of the WLRB. The SC diminished from southwest to northeast following the river flow. As for HQ, the mean value in the WLRB was 0.38. The highest HQ occurred in the southernmost part of the basin, whereas the lowest value was recorded in the Horqin sandy area.

To better detect the temporal variation of ESs during 1990–2020, the study period was divided into pre-GFGP (1990–2000) and post-GFGP (2000–2020) periods. In the pre-GFGP period, the highest annual WY (82.4 mm) occurred in 1990, followed by a decrease to 17.2 mm in 2000. The southeastern plains presented a considerable decline (Figure 5e). The CS and HQ showed a decreasing trend in 70.8% and 79.7% of the subbasins from 1990 to 2000, with annual average decreasing rates of 1.1% and 2.9%, respectively. The SC overall increased by 1.4% in the pre-GFGP period, and an increasing was shown in 83.5% of the subbasins. In the post-GFGP period, the WY started to increase, concentrated in the southwest mountainous area of the WLRB (Figure 5i). After 2000, the CS and HQ continuously showed a declining trend. However, the decline was halted in the area where GFGP was implemented. Spatially, the declining trend in CS and HQ tended to disappear in the southeast area, which had the highest decline rate before the GFGP was implemented. Furthermore, the SC decreased in the northern part of the WLRB. Overall, 19.6% of the study area represents a key area for future SC restoration.

Considering these four ESs, the average TES was 1.02 in the WLRB. The TES’s spatial distribution is depicted in Figure 6. Due to changes in topography and climate, the TES declined from south to north, reaching the maximum value of 2.65 in the southwest corner of the study area, whereas the lowest value was close to 0 in the Horqin sandy area. Generally, the TES in the WLRB deteriorated between 1990 and 2000, while the TES improved during 2000–2020. The TES value recovered from 0.89 in 2000 to 1.08 in 2020. Compared to the subbasins where the GFGP was not implemented, the recovery rate of TES was noticeably higher in the subbasins with GFGP (Figure 6d).

4.2. Tradeoff and Synergy among Ecosystem Services

The tradeoff and synergy among ESs were investigated by partial correlation analysis. The influencing factors of ESs were excluded in partial correlation analysis, including the meteorological conditions (precipitation, temperature, and evapotranspiration), vegetation status (NDVI), socioeconomic development level (density of GDP and population), and urbanization area. The partial correlation coefficients for the core explanatory variables are shown in the right triangle of Figure 7. The partial correlation coefficients among CS, SC, and HQ were all significantly higher than 0, indicating a potential synergistic interaction of the three. Specifically, the CS and HQ showed the strongest linkage, with a partial correlation coefficient of 0.78. The positive connection between the four ESs and TES revealed that CS and HQ primarily influenced TES, whereas WY had the most negligible impact.

On the contrary, WY was negatively correlated to CS, SC, and HQ, with correlation coefficients of −0.20, −0.23, and −0.25, respectively, indicating that the WY had tradeoff relationships with other ESs. The results may indicate that revegetation programs in these regions might create a potential conflict between water demand and ecosystem restoration. For example, an increase in the artificial forest would typically sequester more carbon and soil, due to the fact that soil water extracted by tree roots may be deeper than crops. Compared with the Pearson’s correlation analysis in the left triangle of Figure 7, the tradeoff relationships among WY, CS, and SC may have been misjudged by the traditional correlation analysis method.

Since the WLRB is an important water-conserving area of China, the tradeoff between WY and ESs is noteworthy for ecological planning. Spatially, the tradeoff between WY and other ESs varied widely, as shown in Figure 8. The zones with higher RMSE were mostly found in the southwestern mountain area of the WLRB, where the average RMSE between WY and CS, SC, and HQ was 0.24, 0.22, and 0.36, respectively. The tradeoffs in these regions were more beneficial for CS, SC, and HQ. As for the plain area, the tradeoff between WY and ESs was enhanced in the eastern study area. The RMSE values were consistently lower in the south of the plain area, indicating a stable synergistic relationship. Tradeoffs benefited WY to the east of WLRB, which have water utilization potential. Notably, subbasins that implemented GFGP had a lower RMSE between WY and CS and HQ. Afforestation or returning farmland to forest/grassland activities may have also intensified the tradeoff between WY and SC.

4.3. Response of Ecosystem Services to Driving Factors

4.3.1. The Effect of the Grain-for-Green Program on Ecosystem Services

The regression results are reported in Figure 9 and

Table 2. Results of benchmark regression.

Model	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	WY	CS	SC	HQ	TES	WY-CS	WY-SC	WY-HQ
GFGP	−0.011 **	0.016 ***	−0.001	0.022 ***	0.006 ***	−0.012	−0.005	0.012 **
	(0.006)	(0.003)	(0.002)	(0.003)	(0.002)	(0.010)	(0.006)	(0.005)
PRE	1.281 ***	0.060 **	−0.009	−0.001	0.333 ***	0.312 *	0.455 ***	0.002
	(0.180)	(0.030)	(0.011)	(0.013)	(0.039)	(0.162)	(0.097)	(0.074)
TEM	0.072	0.119 ***	0.120 ***	0.029	0.085 ***	1.052 ***	0.671 ***	0.641 ***
	(0.083)	(0.039)	(0.024)	(0.035)	(0.027)	(0.177)	(0.099)	(0.081)
AET	−0.982 ***	−0.098 *	0.014	0.034	−0.258 ***	−0.779 ***	−0.489 ***	−0.164
	(0.350)	(0.057)	(0.017)	(0.022)	(0.077)	(0.280)	(0.163)	(0.107)
NDVI	−0.095 **	−0.020	0.031 ***	−0.044	−0.032 **	−0.020	0.019	−0.021
	(0.039)	(0.016)	(0.011)	(0.027)	(0.013)	(0.053)	(0.037)	(0.034)
lnpop	−0.080 *	0.020	0.040 **	−0.032	−0.013	0.349 ***	0.253 ***	0.256 ***
	(0.047)	(0.021)	(0.016)	(0.023)	(0.015)	(0.096)	(0.058)	(0.051)
lngdp	0.140	−0.042	0.017	−0.097 ***	0.004	0.014	−0.122	−0.170 **
	(0.119)	(0.034)	(0.014)	(0.035)	(0.033)	(0.171)	(0.109)	(0.077)
Urban land	0.076	−0.052	−0.102 **	−0.146 ***	−0.056 **	−0.107	−0.093	−0.112 **
	(0.084)	(0.034)	(0.042)	(0.039)	(0.027)	(0.097)	(0.071)	(0.049)
Constant	0.297	0.306 ***	0.042	0.498 ***	0.286 ***	−0.151	−0.082	0.029
	(0.197)	(0.038)	(0.026)	(0.034)	(0.046)	(0.208)	(0.124)	(0.078)
Observations	1465	1465	1465	1465	1465	1465	1465	1465
Adjusted R2	0.940	0.980	0.996	0.989	0.988	0.579	0.729	0.928

Standard errors are in parentheses. ***, **, and * represent significance levels of 1%, 5%, and 10%, respectively. Note, Variable names are abbreviated as follows: GFGP, the effect of the Grain-for-Green Program; PRE, precipitation; TEM, temperature; AET, actual evapotranspiration; NDVI, normalized vegetation index; lnpop, logarithmic population density; lngdp, logarithmic gross domestic product; urban land, the proportion of urban land at the subbasin; WY, water yield; CS, carbon sequestration; SC, soil conservation; HQ, habitat quality; WY-CS, WY-SC, and WY-HQ are the tradeoff coefficients between WY and CS, SC, and HQ.

. The impact of the GFGP on ESs was in the order of HQ > CS > WY > TES > SC. The positive effect of GFGP on the CS and HQ was significant (p < 0.01), as reported in models (1)–(4) of Table 1. The corresponding regression coefficients of 0.013 and 0.021 indicated that the GFGP led to CS and HQ increases by 1.3% and 2.1%, respectively. However, under the same circumstances, GFGP was inversely linked with WY at the 5% confidence level, indicating that GFGP contributed to a 1.1% decline in WY. The impact of GFGP on SC failed the 10% significance test, regardless of whether the control variables were considered. This result suggested that the GFGP had a limited effect on SC. Overall, the GFGP brought a significant improvement in TES. The regression coefficient of TES was 0.006, smaller than that of individual ES, which may have been due to a mutual inhibition effect among ESs. These results fully confirm that the GFGP is an efficient ecological restoration method, which coincides with existing studies in China [40,47].

Models (6)–(8) in Table 1 reveal the impact of GFGP on the tradeoffs between WY and CS, SC, and HQ. The results (Figure 9a) reflect that GFGP significantly enlarged the tradeoff coefficient between WY and HQ, with a 0.012 regression coefficient. This finding suggests that the GFGP may magnify the uneven development of WY and HQ. The water resource reduced by near-term tree planting activities may have been deteriorated by agricultural production promotion. The impact of GFGP on WY-CS and WY-SC was insignificant, indicating a tradeoff relationship between WY and CS (SC) consistent with the previous section. The negative coefficient of GFGP reflected that, due to carbon and soil sequestration processes, vegetation restoration might be beneficial in reducing water consumption.

Since the policy was applied at various timepoints and the model was built using the time-varying DID approach, it was impossible to dynamically examine the policy implementation effects each year. As shown in Figure 10, the dynamic effects of GFGP on ESs were assessed by systematically measuring the policy impacts 5–20 years before policy implementation (periods −1 to −4) and 5–15 years after policy implementation (periods 1 to 3). The dynamic regression coefficients of the core explanatory variables fluctuated around 0 in the periods −4 to −1, confirming that the parallel trend hypothesis was satisfied. In periods 0 to 3, the regression coefficients started to deviate from 0, which was attributed to the effect of the GFGP.

The regression coefficients of the GFGP on WY were less than 0, indicating a reduction in the policy implementation effect. The effect of GFGP decreased from −0.008 to −0.023 in periods 0 to 1. However, the diminishing effect became weakened after period 2. This phenomenon may have been due to the farmland replacement acts that were completed in the first periods of the GFGP. LULC modifications were minor in the latter period of the GFGP when the program’s main goal changed to ensure the survival of planted plants. Similarly, the regression coefficients of GFGP for CS, HQ, and TES also passed the parallel trend test. The regression coefficients of GFGP on CS, HQ, and TES increased from 0 to 0.019, 0.029, and 0.05 in periods 0 to 1, respectively. The regression coefficients of CS and TES increased slowly in periods 2 and 3. The impact of GFGP on HQ lasted for the longest time, increasing to 0.044 in period 3. The GFGP policy may have prevented the expansion of human activities, thus maintaining and improving the HQ of the ecosystem.

4.3.2. The Effect of the Control Variables on Ecosystem Services

The control variables can be divided into climate change and socioeconomic changes due to factors other than the GFGP. As shown in Table 1, precipitation, evapotranspiration, and NDVI were significantly correlated with WY at the 5% confidence level. WY would benefit from sufficient precipitation. However, vegetation greening would increase evapotranspiration and reduce the basin’s runoff [63,64]. Climate warming played the most prominent role in improving CS and SC. The other contributing element to improving SC capacity was the greenness of the vegetation. Despite GFGP’s advancement, the HQ was challenged by urbanization and economic growth.

Overall, the effect of the control variables on TES showed that the high temperature and humid environment were more favorable to the enhancement of ESs. The significant association between TES and anthropogenic factors confirmed that afforestation and prevention of overurbanization remain effective ways to improve ESs in the WLRB. It is worth noting that the warming air temperature played a dominant role in enlarging the tradeoffs between WY and other ESs, especially for the CS. The increasing temperature was conducive to ecological restoration, while the water resources required for vegetation growth in the WLRB were challenging to replenish on time. These findings would intensify the water supply and demand conflict, bringing additional difficulties to land planning. Furthermore, population growth was the most prominent anthropogenic factor influencing the tradeoff between WY and CS, SC, and HQ, with regression coefficients of 0.349, 0.253, and 0.256, respectively. Population change corresponds to changes in supporting facilities such as human gathering places and farmland. When the needs of populations exceed the available resources, the ESs are reduced. The surrounding ecosystems are increasingly modified to provide services such as food and water, often at the expense of other types of ESs [65].

5. Discussion

5.1. Contribution of Ecological Restoration to Ecosystem Services

This study quantized the impact of GFGP on ESs and their tradeoff and synergy in the WLRB. In this study, we used the DID method to exclude the potential effects of meteorological and topographical factors to obtain the net effect of GFGP. In this study, the DID method was applied to exclude the potential effects of meteorological, topographical, and other factors to obtain the net effect of GF policy. Through group regressions, this method has advantage of describing policy implementation impacts when information on policy implementation intensity is missing. The findings showed that GFGP deployment during the previous 20 years successfully slowed the degradation of the natural ecosystem. Nonetheless, the GFGP has not been able to completely prevent the decline of ESs, especially in the plain area of the WLRB. Previous studies extensively investigated the changes in ESs with different ecological restoration policies in China. Results proved that ecological restoration policies are an essential way to improve soil retention, carbon sequestration, water purification, and habitat provisioning [40,66,67]. However, some current policies were confirmed to not be tailored to local environmental conditions, which might have had a negative effect on achieving environmental policy goals [68]. When introduced plants consume more water than native plants, it increases soil water depletion and creates a dry soil layer. This phenomenon can hinder the water exchange between surface water and groundwater and disrupt the hydrological cycle process [69]. Therefore, the adverse effects of policy implementation can be ameliorated by choosing the suitable plant species and structures.

5.2. Policy Implications

Some vegetation restoration strategies still need to be optimized from the following aspects:

(1)

The negative effects of vegetation restoration policies on WY were nonnegligible, despite climatic factors increasing WY in the WLRB. The negative impacts of the revegetation policy were most focused on vegetation restoration, which would lead to soil water deficits for long-term vegetation growth [70]. Additionally, introduced plants would consume more soil water than native plants and affect the local water cycle [68,70]. Studies have confirmed that the runoff decline in the Loess Plateau region was associated with the GFGP by increasing net primary productivity and evapotranspiration [7,25,71]. The GFGP designer should consider the balance to guarantee food security and natural resources, maintain the function of water production in mountainous areas, and scientifically provide water resources for agriculture and cities in plain areas. Water resources in the plain area mainly maintain the balance of the ecological environment and grain supply.

(2)

To alleviate tradeoffs among WY and other ESs, the zonal management of GFGP should be paid more attention. Comprehensively understanding the synergistic or tradeoff relationships of ESs would help make decisions on managing land use and vegetation restoration [33,66]. The tradeoff between soil moisture and erosion control [72], water conservation, and biodiversity conservation [73], etc. have been widely discussed. Maximizing one ES may lead to a decrease in others, while exceeding the threshold may result in irreversible changes [3]. Significant tradeoff relationships among WY and other ESs were explored in our study. The mountain area was considered to be a hotspot of the tradeoff relationship, which was the major water supply and conservation area. In mountain areas, especially with strong tradeoff intensity, excessive forest restoration may exceed the water supply capacity and natural restoration. The Horqin sandy area of the WLRB, which has relatively sufficient water resources and is poor in CS and HQ, needs to improve the regional ESs by optimizing pasture management and planting local vegetation [74]. In the eastern plain of the WLRB, the water resource needs to be restored through water allocation and improving agricultural production capacity.

(3)

Future improvements of ESs required timely adjustments to policy implementation methodologies. Climate change is considered a major threat to ESs, while climate warming would aggravate the tradeoffs among ESs. Policymakers need to take this uncertainty and challenge into consideration. Moreover, in addition to environmental effects, the existing GFGP has economic effects, such as increasing farmers’ income and promoting nonagricultural employment [54,75]. The ecological damage caused by socioeconomic development can be reduced through industrial restructurings, such as developing tourism and planting economic forests.

5.3. Uncertainty and Limitations

Our study provided a straightforward and adaptable method for quantifying the impact of GFGP on ESs. However, there were still uncertainties and limitations. Structural equation modeling can be used to better describe the relationship between different factors than the correlation analysis [76]. We considered the leading influence of changing environment as much as possible, but there are still unappreciated factors. Placebo tests [77] were used to exclude the potential impacts of other random variables on ESs. Details of the placebo test are provided in Appendix B, demonstrating that other unobserved factors did not influence the policy effect of GFGP on TES. Due to data availability limitations, parameters in the model’s biophysical table were referenced from previous research without validation. Therefore, in the future, we will try to collect additional high-scale data and construct high-precision ecohydrological models to discuss the effects of GFGP on ESs in formation mechanisms.

6. Conclusions

This study explored the impact of the GFGP on the ecosystem by simulating and statistically analyzing the evolution and relationship of ecosystem services in the WLRB from 1990 to 2020. The main conclusions are as follows:

(1)

During the pre-GFGP period (1990–2000), annual water yield decreased from 82.4 mm to 17.2 mm, and carbon sequestration and soil conservation were reduced in 70.8% and 79.7% of the subbasins. In the post-GFGP period (2000–2020), the WY started to recover, and the declining tendency of CS and HQ was halted in the group with GFGP. Spatially, the TES declined from south to north of the WLRB, peaking at 2.65 in the southwest corner of the basin and falling to its lowest value (nearly 0) in the central Horqin sandy area.

(2)

A synergistic relationship was shown among carbon sequestration, soil conservation, and habitat quality. The strongest correlation was found between carbon sequestration and habitat quality, with a partial correlation coefficient of 0.78. Water yield had a tradeoff relationship with other ESs, where were primarily present in the mountain area of the WLRB.

(3)

The GFGP was confirmed to have the greatest impact on enhancing habitat quality and carbon sequestration among the four ESs, while having an adverse impact on water yield and soil conservation. Air temperature and population density were major determinants for tradeoffs between water yield and other ESs. The methodology and implications provided in our study can provide guidance for regional ecological planning.