Study on the Impact of Vegetation Restoration on Groundwater Resources in Tianshan Mountain and Yili Valley in Xinjiang, China

Abstract

China has implemented a series of ecological protection and restoration projects in Tianshan Mountain and Yili Valley in Xinjiang, which have significantly improved regional vegetation coverage. Vegetation improves soil structure through roots, especially increasing non-capillary porosity, which enhances the precipitation infiltration performance, thus reducing surface runoff, increasing the interception and infiltration of groundwater resources, and enhancing regional water retention capacity of soil. In order to quantitatively study the impact of ecological conservation and restoration (represented by fraction of natural vegetation coverage, FVC) on groundwater storage (GWS), we investigated GWS changes in this region, identified the main factors, and quantified their relative impacts. Here, we combined data from the Gravity Recovery and Climate Experiment (GRACE) satellite, GRACE Follow-On (GRACE-FO), and Global Land Data Assimilation System (GLDAS) hydrological model from January 2003 to December 2020 and evaluated GWS changes. We used the variable importance in projection and partial least squares regression methods to determine the main influencing factors. We found that (1) before and after 2012, GWS decreased at a rate of 0.80 cm/yr and 0.75 cm/yr (with statistical significance p < 0.01), respectively. (2) Before 2012, the main factors affecting the decrease in GWS were agricultural planting areas, and after 2012, they were temperature, evaporation, and FVC, with relative contributions of 54.72%, 34.59%, and 10.69%, respectively. FVC has a positive regulating effect on the increase in regional GWS.

1. Introduction

Water resources are vital for regional economic and social development. The Yili Valley and Tianshan Mountain in Xinjiang Autonomous Region are in an arid region of northwest China, where water resources are very scarce. Tianshan Mountain contains many glaciers and permanent snow, which, as a ‘natural solid reservoir’, is an important water resource for the surrounding area. Glaciers are sensitive to climate change, and most glaciers on Tianshan Mountain are melting as Earth warms, which increases meltwater and causes continuous losses in ice [1,2,3,4,5]. Rainfall affects GWS dynamics through recharge or extraction of groundwater [6,7,8,9]. The specific performance is as follows: (1) in a warming climate, the increased transpiration decreases the proportion of rainfall converted into surface runoff, and thus reduces the recharge of regional groundwater resources; (2) when rainfall decreases, the frequency and intensity of regional drought events are increased, which indirectly increases the amount of groundwater extracted by human activities such as agricultural irrigation in the region, thus leading to a reduction in groundwater resources. Since 2012, a series of ecological protection and restoration projects have been implemented in the study area, such as natural forest protection, return of farmland to forest, restoration of degraded shelterbelts, etc., and the regional forest coverage rate and comprehensive vegetation coverage of grassland have been effectively improved [10]. Vegetation cover and its change is an important indicator of regional ecosystem environmental change, which is of great significance to hydrology, ecology, and global change. Forest vegetation can retain water and has a certain function of conserving water. The increase in vegetation coverage will increase the interception and infiltration of water resources, and then increase the amount of groundwater. Vegetation improves soil structure through roots, especially the increase in non-capillary porosity, which enhances the precipitation infiltration performance, thus reducing surface runoff, increasing the interception and infiltration of groundwater resources, and enhancing the capacity of regional soil water storage [11,12]. As a large agricultural region, there is also a large regional demand for groundwater extraction for irrigation in Yili Valley [13]. With the increase in population and the continuous expansion of agricultural planting areas (APAs), the intensity of water resources development and utilization is increasing in this region, which increases the severity of water resource shortages.

Traditional studies in arid and semi-arid areas, especially in northwestern Xinjiang, focus on regional glacier dynamic change, the impact of glacier change on water resources, or the response of water resources to climate change [13,14,15]. There are few comprehensive assessment studies on the impact of climate, agriculture, and ecological restoration on regional water resources as proposed in our paper. Therefore, in order to solve the above problems and explore the impact of ecological protection and restoration projects (represented by Fractional Vegetation Cover (FVC)) on regional GWS, this paper chooses 2003–2020 as the research period, with 2012 as the intermediate time node, to carry out relevant research. However, due to the complex terrain, there are few groundwater-monitoring stations in this region and their spatial distribution is extremely uneven. In addition, satellite remote sensing can only monitor shallow surface water resources. Therefore, it is difficult to monitor changes in Groundwater Storage (GWS) using traditional methods. However, the Gravity Recovery and Climate Experiment (GRACE) satellite has a uniform observation scale and spatial distribution and can be used to retrieve regional GWS changes by combining spaceborne observations with a hydrological model [16,17,18]. Thus, we used GRACE and GRACE-FO data to investigate changes in GWS in Yili Valley and Tianshan Mountain region. A diagram about the methodology in this study is shown in Figure 1.

2. Study Area

Our study area was mainly in Xinjiang, China, and included 41 counties (cities and districts) that covered Yili Valley and Tianshan Mountain, with a geographical location of 39° N–46° N and 78° E–92° E, with a total area of 444,200 km² (Figure 2). The region has a temperate continental climate, the average annual rainfall is between 27–680 mm, and the average annual evaporation is between 520–1530 mm, with an extremely uneven spatial and temporal distribution of water resources [15,19,20,21]. The regional arid climate dictates that the forest coverage of the oasis is low, and the vegetation is dominated by irrigated desert grass. Desert vegetation plays a major role in the protection and maintenance of the oasis ecological environment.

The regional geological structure is relatively complex. The Tianshan area in this study is located in the suture zone between the Junggar Block and the Tarim Plate, formed by the collision between the Indian plate and the Eurasian plate. The regional sedimentary layer is deep, and the hydrogeological conditions are significantly different. The aquifers can be divided into single-structure diving in front-inclined plains, confined water zone diving in alluvial plains, and confined water [22,23]. The Ili River Valley is located in the West Tianshan eugeosynclinal fold belt of the southwest Tianshan fold system in the Tianshan-Xing’an geosynclinal fold region. It is a fault valley of the Mesozoic era. The Quaternary strata are developed, the sedimentary thickness is large, the porosity of loose rocks is dominant, and confined water is distributed locally [24,25].

Many glaciers, with an area of 6500 km², cover Tianshan Mountain of Xinjiang. Snowfall covers more than 50% of Tianshan Mountain in winter [26], which is an important source of water in this region. In spring, snowmelt forms runoff and recharges groundwater. The study area is dominated by mountains, with a small area of plains and hills.

3. Data and Methods

3.1. GRACE Terrestrial Water Storage Anomalies (TWSA) Solution

The Gravity Recovery and Climate Experiment (GRACE) satellite was launched jointly by NASA’s Jet Propulsion Laboratory (JPL) and the German Aerospace Center (DLR) in 2002 and retired in 2017, primarily due to the variation in Earth’s gravitational potential, after which, a successor satellite, GRACE-Follow on, was launched in 2018 [27,28]. Using GRACE data, TWSA can be estimated. There are three conventional GRACE TWSA solution methods, namely CSR Mascon solutions, JPL Mascon solutions, and spherical harmonic coefficient solutions. In the previous research, it was found that the TWSA results obtained by the three GARCE gravity satellite solution methods showed a similar trend, but the accuracy of CSR Mascon solutions results was relatively higher [29,30,31,32,33]. Therefore, in this study, we used the CSR GRACE/GRACE-FO RL06 Mascon Solutions (version 02) to obtain TWSA with 0.25° spatial resolution and 1-month intervals. The time span was from January 2003 to December 2020, covering 184 months (the data from June 2003; January 2004; January and June 2011; May and October 2012; March, August, and September 2013; February, July, and December 2014; June, October, and November 2015; April, September, and October 2016; February 2017; from July 2017 to May 2018; and August and September 2018 were missing). For missing data from July 2017 to May 2018, we used a dataset of reconstructed terrestrial water storage in China based on precipitation (2002–2019) at the same stage, which was provided by National Data Center for Tibetan Plateau Science (referred to as TPDC) [34]. For other missing data, we used linear interpolation for gap-filling [35]. In order to verify the reliability of the replacement data from July 2017 to May 2018, we compared the reconstructed TWSA provided by TPDC with CSR Mascon TWSA, as shown in Figure 3. It can be seen that the change trend of the two time series is consistent, with the same rising or falling state at the same stage, and the amplitude difference between the two is not large. The correlation and Root Mean Square Error (RMSE) between the two were 0.93 and 2.19 cm, respectively, showing a high correlation. Therefore, the reconstructed data is relatively reliable. Considering that the two solutions have similar results, the latest time of reconstructing TWSA is December 2019. Therefore, we still adopt the TWSA calculated by CSR Mascon for periods other than July 2017 to May 2018 in this paper.

CSR Mascon solutions are estimated with the same standards as CSR RL06 Spherical harmonics solutions using GRACE Level-1 observations, with C20 and C30 replacement, degree 1 corrections, and glacial isostatic adjustment correction [35,36,37]. Regulation constraint, instead of empirical filtering (such as smoothing or de-striping), was applied to CSR Mascon solutions to remove correlation errors. Anomalies reported in CSR Mascon solutions are relative to the January 2004–December 2009 mean baseline and are expressed in terms of Equivalent Water Height (EWH). CSR Mascon solutions are applicable to all science areas, such as oceanography, land surface hydrology, and the cryosphere.

3.2. Hydrological Model

3.2.1. Global Land Data Assimilation System (GLDAS)

GLDAS can provide hydrological parameters in near-real time by ingesting satellite and ground-based observational data [38]. Here, GLDAS-2.1 Noah monthly 0.25° data was taken to obtain hydrological parameters, and the time span was consistent with CSR RL06 GRACE and GRACE-FO Mascon solutions’. GLDAS-2.1 is forced with a combination of model and observation data from 2000 to present, which includes Soil Moisture (SM) at a depth of 2 m, Snow Water Equivalent (SWE), and plant canopy surface water [39]. However, the model excludes groundwater and Antarctic data [40,41].

The Surface Water Storage (SWS) can be derived by summing SM, SWE, and canopy water. To be consistent with TWSA, the difference between SWS and the mean value of SWS between 2004 and 2009 was calculated, and the Surface Water Storage Anomalies (SWSA) were obtained.

3.2.2. WaterGAP Global Hydrology Model (WGHM)

The WaterGAP Global Hydrology Model (WGHM) was originally developed by Ref. [42], which can describe the status of global water resources, except Antarctica [43]. The most recent model version, WaterGAP 2.2d, was used in this study, which has a 0.5° spatial resolution and a 1-month interval [44,45], and the time span was from 2003 to 2016. The model includes surface reservoir storage (RESS), SM, SWE, GWS, and water storage related to human activities [46,47,48]. Groundwater storage anomalies (GWSA) were calculated by deducting the mean value of WGHM 2.2d GWS during 2004 and 2009. Here, based on WGHM 2.2d GWSA, the evaluation of CSR Mascon GWSA was quantitatively verified by correlation and Root Mean Square Error (RMSE) methods.

3.3. Other Factors

3.3.1. Meteorological Factors

Meteorological factors affecting GWS changes in the study area may include precipitation, actual evaporation, and temperature. Precipitation and actual evaporation of Terra-Climate [49] were used in this study, which has a spatial and temporal resolution of 4 km and 1 month. Temperature was from the 1 km monthly mean temperature dataset for China (1901–2020) [50]. The time span was from January 2003 to December 2020.

To obtain regional annual meteorological data, the spatial resolution of different hydro-meteorological data was first unified using the PRISM (Parameter-Elevation Regressions on Independent Slopes Model) interpolation algorithm proposed by Refs. [51,52], which takes into account the comprehensive influence of geographical factors such as elevation, distance, slope, and height; then, the regional monthly meteorological data was derived by averaging all grid data in the same month. Finally, by calculating the sum of monthly precipitation/actual evaporation or the mean of monthly temperature in 12 months of a year, the regional annual meteorological data were derived. To be consistent with CSR RL06 Mascon GWSA, we calculated the anomalies of meteorological factors relative to the average value from 2004 to 2009.

3.3.2. Hydrological Factors

Glacier Mass Change

The Randolph Glacier Inventory (RGI), which was derived from Global Land Ice Measurements from Space (GLIMS), is a complete catalog of the global contours of glaciers. RGI 6.0 includes the vector boundary, elevation, area, and other information for each glacier [53]. According to the distribution of each glacier recorded in RGI 6.0, the glacier data in our study area were screened out, as shown in Figure 2. Many studies have shown that most glaciers in Xinjiang have retreated in the past 50 years due to climate change [1,2,3,4]. Mass loss in the glacier area of Tianshan Mountains may be caused by glacier ablation [54] and sublimation of snow and ice. Here, the annual rate time series for each glacier in Central Asia from 2003 to 2020, which was provided by Ref. [55], were used to analyze glacier mass change. By calculating the sum of the mass change rates of all glaciers in the same period, the mass change of glaciers in the study area in a certain period can be obtained. The glacier mass change rate from 2003 to 2020 can be obtained by summing the annual mass change rate during this period.

Groundwater Level (GWL)

Due to the remoteness and complex topography of our study area, groundwater monitoring wells are scarce [54,56,57]. Here, the measured groundwater level (GWL) mainly came from two sources, the monthly groundwater depth data from China Geological Environment Monitoring Groundwater Table Yearbook and the daily groundwater depth data from the China Earthquake Networks Center. The spatial distributions of groundwater monitoring wells from the two sources are shown in Figure 2. The time range of data that available was 2006–2017.

Due to partial lack of monthly groundwater level, the average of all monthly groundwater levels that can be obtained within one year were used as the result of the groundwater level of the year to ensure the reliability of the CSR RL06 Mascon GWSA verification data. Next, we converted groundwater depth to GWL with elevation data. Finally, Kriging spatial interpolation and regional averages were used to obtain the regional annual GWL; among them, the Kriging interpolation space autocorrelation model adopts the spherical function model in the semi-variation function, and the maximum and minimum domain window sizes are set to 5 and 2, respectively. In order to ensure the verification method is scientific and reasonable, we selected the monitoring well distribution area near the location and meeting the GRACE resolution requirements as a verification area. According to the above principles and referring to the spatial distribution position of monitoring wells in Figure 2, the currently available monitoring wells are divided into four verification areas (the area delimited by the blue circle in Figure 2). By comparing the annual groundwater level of each region with the GRACE GWSA results of the corresponding region, the accuracy of the GRACE GWSA results of the corresponding region was assessed.

Runoff

The melting glacier replenishes regional water resources in the form of runoff, which is converted into groundwater by infiltration. Therefore, runoff may be one of the factors affecting GWS changes in the study area. We used runoff from Terra-Climate [49] from January 2003 to December 2020 in our study. The processing steps were consistent with precipitation and actual evaporation as described above.

3.3.3. Agricultural Planting Area (APA)

According to the literature, we found that there was a large amount of agricultural cultivation in our study area [13]. During agricultural cultivation, a large amount of groundwater is extracted for irrigation. Hence, APA (units: 10⁴ km²) was taken as one of the possible factors influencing regional groundwater changes, which can be derived from Xinjiang Statistical Yearbook (2003–2020) [58]. To be consistent with CSR RL06 Mascon GWSA, the anomalies of APA were calculated relative to the average value from 2004 to 2009.

3.3.4. Agricultural Water Use (AWU)

It was difficult to obtain AWU in the study area, but AWU in Xinjiang can be obtained through the Xinjiang Statistical Yearbook or China Statistical Yearbook [58]. AWU in the study area from 2003 to 2020 can be roughly estimated using the following formula:

$${AWU}_{sa}={APA}_{sa}\ast {AWU}_{xj}/{APA}_{xj}$$

(1)

where ${AWU}_{sa}$ and ${AWU}_{xj}$ refer to AWU in the study area and Xinjiang, respectively, and ${APA}_{sa}$ and ${APA}_{xj}$ refer to the APA in the study area and Xinjiang, respectively.

3.3.5. Fraction of Vegetation Coverage (FVC)

Here, we used FVC of the five-year survey and evaluation of national ecological status carried out by the Ministry of Ecology. The spatial resolution of FVC was 500 m, and the time span was from January 2003 to December 2020 (Figure 4). We only selected FVC of natural vegetation, such as forest and grassland, for analysis, excluding farmland. We used land use data provided by Resource and Environment Science and Data Center of Institute of Geographic Sciences and Natural Resources Research [59]. The spatial resolution of the land use data was 1 km and the time span was 2005, 2010, 2013, 2015, 2018, and 2020 (Figure 5). First, FVC was resampled to 1 km using bilinear interpolation. Second, we selected the maximum FVC in a year as the annual value. Third, we extracted the FVC of forest and grassland based on land use of the nearest year (

Table 1. Time corresponding table of FVC extraction and land use.

ID	FVC	Land Use
1	2003–2007	2005
2	2008–2011	2010
3	2012–2014	2013
4	2015–2016	2015
5	2017–2018	2018
6	2019–2020	2020

). Finally, we calculated the mean value of forest and grassland FVC of a year as the region FVC. To be consistent with CSR RL06 Mascon GWSA, the anomalies of FVC were calculated relative to the average value from 2004 to 2009.

3.4. Methods

3.4.1. Estimation Model of GWS Changes

TWS changes include changes in soil moisture storage (SMS), snow water equivalent (SWE), surface reservoir storage (RESS), and GWS [41]. Therefore, the terrestrial water storage balance can be described as follows:

$$∆\mathrm{T}\mathrm{W}\mathrm{S}=∆\mathrm{S}\mathrm{M}\mathrm{S}+∆\mathrm{S}\mathrm{W}\mathrm{E}+∆\mathrm{R}\mathrm{E}\mathrm{S}\mathrm{S}+∆\mathrm{G}\mathrm{W}\mathrm{S}$$

(2)

According to Formula (2), regional GWS changes can be obtained as follows:

$$∆\mathrm{G}\mathrm{W}\mathrm{S}=∆\mathrm{T}\mathrm{W}\mathrm{S}-∆\mathrm{S}\mathrm{M}\mathrm{S}-∆\mathrm{S}\mathrm{W}\mathrm{E}-∆\mathrm{R}\mathrm{E}\mathrm{S}\mathrm{S}$$

(3)

The annual glacier mass change rate in the study area is shown in Figure 6. The rate of glacier mass loss gradually increased, and the overall reduction rate was −2.26 Gt/yr (−0.24 cm/yr), which was relatively fast compared to GWS change calculated by CSR Mascon (−0.76 cm/yr) before the loss of glacier mass was deducted. Thus, glacier mass change was an important part of surface mass change in this region.

GRACE gravity satellite monitors the mass change of Earth’s surface, including glacier mass change, which are mainly manifested as TWS change in land area. Thus, TWS changes retrieved by GRACE includes part of the changes in mass due to glacier melting. It was necessary to deduct the influence caused by glacier melting when analyzing TWS changes. Thus, we calculated actual TWS change in the study area as follows:

$$∆\mathrm{T}\mathrm{W}\mathrm{S}=∆{\mathrm{T}\mathrm{W}\mathrm{S}}_{Grace}-∆\mathrm{G}\mathrm{l}\mathrm{a}\mathrm{c}$$

(4)

where $∆{\mathrm{T}\mathrm{W}\mathrm{S}}_{Grace}$ is TWS changes retrieved by GRACE and $∆\mathrm{G}\mathrm{l}\mathrm{a}\mathrm{c}$ is the change in glacier mass. By substituting Equation (4) into Equation (3), the actual GWS change calculation equation in the study area can be obtained:

$$∆\mathrm{G}\mathrm{W}\mathrm{S}=∆{\mathrm{T}\mathrm{W}\mathrm{S}}_{Grace}-∆\mathrm{G}\mathrm{l}\mathrm{a}\mathrm{c}-∆\mathrm{S}\mathrm{M}\mathrm{S}-∆\mathrm{S}\mathrm{W}\mathrm{E}-∆\mathrm{R}\mathrm{E}\mathrm{S}\mathrm{S}$$

(5)

$∆\mathrm{G}\mathrm{l}\mathrm{a}\mathrm{c}$ can be derived from Ref. [55] and $∆\mathrm{S}\mathrm{M}\mathrm{S}$ and $∆\mathrm{S}\mathrm{W}\mathrm{E}$ can be derived from GLDAS. However, $∆\mathrm{R}\mathrm{E}\mathrm{S}\mathrm{S}$ was relatively small and can be ignored. It should be noted that, since only annual scale glacier mass can be determined here, only annual GWS changes reduce glacier mass, and this reduction was used in the subsequent analysis of influencing factors.

3.4.2. Least Square Spherical Harmonic Analysis

The least square spherical harmonic analysis method obtains the seasonal change signal of regional water storage change, and its specific calculation formula is as follows [60,61,62]:

$$y\left(t\right)=a+bt+\sum _i^2{A}_i\cos \left[\frac{2\pi \left(t-{\phi }_i\right)}{T_i}\right]+\epsilon (t)$$

(6)

where y(t) is TWSA/GWSA derived by CSR Mascon/WGHM, and a and b are the constant and slope of the linear trend term, respectively. $t$ is the time series, which in this article mainly refers to January 2003 to December 2020. $A_i$, ${\phi }_i$, and Ti refer to amplitude, phase, and period, respectively; i = 1 or 2, representing the annual and semi-annual terms, respectively; accordingly, $T$ is equal to 1 or 1/2. $\epsilon \left(t\right)$ is the residual term.

Based on CSR Mascon TWSA/WGHM GWSA, the parameters of a, b, A_i, ${\phi }_i$, and $\epsilon \left(t\right)$ in Equation (6) can be determined by the least-squares regression method, and the seasonal term can be calculated from the fitting parameters directly. The seasonal variation signals were subtracted from TWSA/GWSA time series calculated by CSR Mascon/WGHM (Equation (7). Thus, GWSA calculated by GRACE TWSA has deducted the seasonal information.

$$y^{\prime }=y(t)-{\sum }_{i=1}^2{A}_{i}cos[2\pi (t-{\phi }_i)/T_i]$$

(7)

where $y^{\prime }$ is TWSA/GWSA minus the seasonal signal.

The 13-point sliding window method is used to eliminate the seasonal effect of the residual $\epsilon \left(t\right)$, and then the RMSE of the residual after removing the seasonal effect is calculated, which is the uncertainty of CSR Mascon GWS.

3.4.3. Partial Least Squares Regression (PLSR)

PLSR is a statistical analysis method, which combines canonical correlation analysis and multiple regression analysis and has the characteristics of principal component analysis (PCA) [63,64]. It can effectively solve the high linear correlation between independent variables and has the advantage of obtaining high prediction accuracy with less computation. The basic model is as follows:

$$\begin{gathered} Y=\mathrm{A}X+\mathrm{b} \\ X=\left\{\begin{array}{c}{x}_{11},x_{12},\cdots ,x_{1n}\\ x_{21},x_{22},\cdots ,x_{2n}\\ ⋮ ⋮ ⋮\\ x_{m1},x_{m2},\cdots ,x_{mn}\end{array}\right\}, Y=\left\{y_1,y_2,\cdots ,y_n\right\}, \mathrm{A}=\left\{a_1,a_2,\cdots ,a_m\right\} \end{gathered}$$

(8)

where Y is GWSA in the study area from 2003 to 2020; X is influencing factors, such as precipitation, temperature, actual evaporation, runoff, APA, and FVC; n = 18, which refers to the number of observed samples from 2003 to 2020; m is the number of influencing factors; A and b are coefficients and intercepts of the regression model, respectively.

PLSR eliminates linear correlation between independent variables mainly through variable importance in projection (VIP) [65]. The index is based on PCA and represents the explanatory ability of independent variables to dependent variables. The specific calculation formula is as follows:

$${\mathrm{V}\mathrm{I}\mathrm{P}}_i=\sqrt{(\rho {\sum }_{j=1}^{m}Rd\left(Y,t_h\right)W_{ji}^2)/{\sum }_{j=1}^{m}Rd(Y,t_h)}$$

(9)

where $\rho$ is all possible influencing factors of GWS change; m is the number of principal components extracted from all possible influencing factors; t_h is the hth principal component; Y is GWSA; Rd(Y, t_h) is the correlation coefficients of Y and t_h; and W_ji is the weight of influencing factors on corresponding principal component. The higher the VIP, the stronger the explanatory power of the dependent variable. Independent variables with VIP greater than 1 are generally considered to be more important and are often selected to participate in regression analysis [66].

4. Results

4.1. Evaluation of CSR Mascon GWSA

Before analyzing the trend of GWS and its influencing factors in the study area, we used the measured GWL and the WGHM GWSA to verify the accuracy of GWSA.

4.1.1. With Measured GWL

Limited by the time frame of measured GWL data, only GWSA results from 2006 to 2017 were verified, as shown in Figure 7. In the four verification regions, CSR Mascon GWSA and the measured GWL showed a good consistency in the change trend, showing a declining state, and the correlation was high, both above 0.77.

4.1.2. With WGHM GWSA

WaterGAP 2.2d provided GWSA from January 2003 to December 2016, so only CSR Mascon GWSA for this period was validated, as shown in Figure 8. The WGHM GWSA and CSR Mascon GWSA in the same period essentially kept the same change state (

Table 2. Change status of GWS from CSR Mascon at different periods (cm/yr).

Period	January 2003–December 2011	January 2012–December 2016
CSRM	descending (p < 0.01)	descending (p < 0.01)
WGHM	descending (p < 0.01)	descending (p < 0.01)

), except for individual points. The correlation between the two was high (r = 0.67), and the RMSE was relatively small (4.11 cm) compared to the retrieval accuracy of the GRACE gravity satellite on the 400 km scale (1~2 cm) [67,68,69].

In conclusion, both the measured GWL and WGHM GWSA were highly correlated and consistent with CSR Mascon GWSA, and the deviation between WGHM GWSA and CSR Mascon GWSA was small.

4.1.3. Analysis of CSR Mascon GWSA

Here, we calculated the change trend of CSR Mascon GWS in the two stages from 2003 to 2011 and 2012 to 2020, and found that both stages were in a declining state and the decline rates were −0.80 cm/yr (r² = 0.89, p < 0.01) and −0.75 cm/yr (r² = 0.70, p < 0.01), respectively. The uncertainty of CSR Mascon GWS in the whole period was 1.37 cm. Figure 9 shows the spatial variation trend of CSR Mascon GWS from 2003 to 2020 in the study area. GWS had a downward trend from 2003 to 2020, and the strong changes were mainly concentrated in the central and western part of the study area, namely the border area of Yili Kazak Autonomous Prefecture, Aksu Prefecture, and Bayingolang Mongolian Autonomous Prefecture, covering an area of 96,800 km², with a maximum decline rate of −2.94 cm/yr. Combined with the spatial distribution of FVC in Figure 4 and land use in Figure 5, it can be seen that the regions with significant changes in GWS mainly include agricultural land and forest and grass land. According to the analysis results of the change trend of APA and FVC in

Table 3. Trend of APA or FVC during 2003–2011 and 2012–2020 (p < 0.01).

Trend	2003–2011	2012–2020
APA (km2/yr)	1238	−34
FVC (/yr)	0.002	0.007

, regional APA showed a rapid increase from 2003 to 2011. The increase rate of FVC in 2012–2020 is faster than that in 2003–2011, indicating that the significant change of GWS in this region may be caused by land use change and ecological restoration.

4.2. Change Status of Influencing Factors

4.2.1. Hydro-Meteorological

Hydro-meteorological factors, such as precipitation, actual evaporation, temperature, and runoff, may be influencing factors of GWS change in the study area. As can be seen from Figure 10 and

Table 4. Trend of hydro-meteorological factors during 2003–2011 and 2012–2020 (p < 0.1).

Trend	2003–2011	2012–2020
Precipitation (cm/yr)	−0.24	−0.24
Evaporation (cm/yr)	−0.15	−0.17
Temperature (°C/yr)	0.01	0.14
Runoff (cm/yr)	−0.08	−0.01

, during 2003–2011 and 2012–2020, the temperature change rate in the study area changed from 0.01 °C/yr in the earlier period to 0.14 °C/yr in the later period, and the increasing trend accelerated significantly. The change rate of runoff decreased from 0.08 cm/yr in the early period to 0.01 cm/yr in the later period, and the decreasing trend was obviously slower. The evaporation rate decreased from 0.15 cm/yr in the early period to 0.17 cm/yr in the later period, and the decreasing trend accelerated slightly. However, the change rate of precipitation in two periods was same.

4.2.2. Agricultural Planting Area (APA)

A large area of crops were planted in the study area; thus, a large amount of groundwater was extracted for crop irrigation, reducing regional groundwater resources [13]. Thus, agricultural irrigation may be one of the influencing factors of GWS change in the study area. Around 2012, the change rate of APA in the study area changed from an annual increase of 1238 km² in the early period to a yearly decrease of 34 km² in the later year (Table 3), with statistical significance p < 0.01. As shown in Figure 11, APA and GWS showed the opposite trend during the whole study period. APA increased rapidly from 2003 to 2015, and GWS decreased rapidly in the same period. From 2015 to 2018, APA increased slowly, and the decrease in GWS slowed simultaneously. From 2018 to 2020, APA decreased significantly, but GWS continued to decrease, which may be due to other dominant factors. Overall, GWS decreased and APA increased throughout the study period, and the correlation between the two was high (r = −0.77). Therefore, agricultural irrigation may have been one of the factors that reduced GWS in the study area.

4.2.3. FVC of Natural Vegetation

The area of grassland and forest was large in the study area, which account for more than 40% of the total area of the study region, and vegetation protection and restoration projects have significantly improved the quantity and quality of the region’s vegetation. Vegetation absorbs water through roots during growth and recycles water through transpiration, vegetation improves soil structure through roots, especially the increase in non-capillary porosity, which enhances the precipitation infiltration performance, thus reducing surface runoff, increasing the interception and infiltration of groundwater resources, and enhancing the capacity of regional soil water storage. Here, we used the FVC of natural vegetation, such as forest and grass, to analyze its trend and explore its relationship with GWS change. From 2003 to 2011, the increase rate of FVC was 0.002/yr, and from 2012 to 2020, it was 0.007/yr, and the increase trend was significantly accelerated (Table 4), with statistical significance p < 0.01.

4.3. Quantitative Analysis of Influencing Factors

4.3.1. Determination of Main Influencing Factors

Here, we determined the main influencing factors of GWS changes from a quantitative perspective. As discussed, VIP can be used to explain the contribution of each influencing factor to GWS change. According to the size of VIP, the factors that had a relatively large influence on GWS were screened out and the multicollinearity among the independent variables was eliminated. The contribution rates of various factors to GWS changes during 2003–2011 and 2012–2020 are respectively shown in

Table 5. VIP of all influencing factors from 2003 to 2011 and 2012 to 2020.

Period	Precipitation	Evaporation	Temperature	Runoff	Agricultural	FVC
2003–2011	0.62	0.83	0.30	0.56	−1.98	0.78
2012–2020	0.85	+1.08	−1.37	0.63	0.91	+1.01

Note: (1) (+) and (−) are obtained by correlation, where (+) indicates that it plays a positive regulating role, that is, the greater the evaporation and FVC, the more regional GWS; (−) indicates a negative regulatory effect, that is, the higher the temperature or the larger the agricultural planting area, the less regional GWS. (2) VIP is the abbreviation of Variable Importance Projection.

.

As can be seen from Table 5, in the early stage, the contribution of each influencing factor on GWS change from small to large was as follows: temperature, runoff, precipitation, FVC, evaporation, and agricultural. Among all the influencing factors, only the contribution of agriculture was greater than 1. As described by Refs. [65,66], variables with a VIP value greater than 1 should be considered as important and the main influencing factors of changes in the dependent variable, which are usually selected for regression analysis. We found that APA was the main factor that affected GWS from 2003 to 2011, and had a negative regulatory effect on GWS change. Based on this, combined with Table 5, it can be concluded that the main factors affecting GWS change in the study area during 2012–2020 were evaporation, FVC, and temperature, among which the first two play a positive regulating role and the third plays a negative role. The main influencing factors from 2003 to 2011 and from 2012 to 2020 were used in the subsequent quantitative analysis to determine the degree to which each factor influences GWS.

4.3.2. Contribution of the Main Factors to Changes in GWS

The main factors that influenced GWS were determined in Section 4.3.1. From 2003 to 2011, there was only one major influencing factor of GWS change, namely APA. From 2012 to 2020, there were three main factors affecting the change of GWS, namely evaporation, temperature, and FVC. Here, we used the PLSR method to establish the relationship between three main influencing factors and GWS change during 2012–2020. First, we normalized the three factors and GWS changes, and then we used the PLSR model of the normalized data to obtain a standardized regression equation (with statistical significance p < 0.05, correlation r = 0.62 and RMSE = 1.69 cm), as shown below:

$$y=0.55x_{eva}-0.87x_{tem}+0.17x_{fvc}-2.93$$

(10)

Based on the standardized regression equation, the ratio of the absolute value of the standardized regression coefficient of each independent variable to the sum of the absolute value of the standardized regression coefficient of all independent variables was calculated as the relative influence weight of each variable. From this, we concluded that the relative influences of temperature, evaporation, and FVC on GWS were 54.72%, 34.59%, and 10.69%, respectively. After 2012, temperature had the greatest influence on GWS change, followed by evaporation, and the positive adjustment effect of FVC on GWS change was at least 10%.

5. Discussion

5.1. FVC of Ecological Protection and Restoration Area

The increase in vegetation coverage will increase the interception and infiltration of water resources, and then increase the amount of groundwater [11,12,70]. FVC is a comprehensive quantitative index of plant community covering the land surface and an important parameter to describe vegetation community and ecosystem. Vegetation cover and its change is an important indicator of regional ecosystem environmental change, which is of great significance to hydrology, ecology, and global change. Therefore, this paper selected FVC as an indicator of ecological protection and restoration effectiveness to carry out relevant analysis. However, the vegetation coverage of the whole study area was used to participate in the analysis, resulting in a certain one-sidedness of the analysis results. Strictly speaking, vegetation area or FVC within the scope of ecological protection and restoration projects should be selected for analysis, but it is difficult to obtain these data.

5.2. CSR Mascon GWSA versus WGHM GWSA

As can be seen from Table 2 and Figure 8, CSR Mascon GWSA and WGHM GWSA showed the same increase or decrease at the same stage. However, there were differences in the degree and magnitude of change between the two, in that the amplitude of WGHM GWSA in the same period was significantly larger than that of the CSR Mascon GWSA. Refs. [46,71] pointed out that the WGHM hydrological model has some limitations in simulating GWS in local areas. Moreover, differences in data acquisition means and processing methods between WGHM and CSR Mascon can also lead to these differences in amplitude.

Besides, due to the difficulty of data acquisition, the influence of regional geological structure was not considered in the process of retrieving regional water reserves using GRACE gravity satellites, which will be the focus of our next study. However, the variation trend of GWS calculated by CSR Mascon was consistent with previous studies’ [15,56,57,72], which verified the accuracy of CSR Mascon GWSA from the side.

5.3. APA versus AWC

In our analysis of the main influencing factors of GWS changes, we used APA instead of AWU, mainly because AWU in the study area was estimated (Equation (1)), and the results were uncertain to some extent. We discussed the changes in APA and AWU and the correlation between them, as shown in Figure 12. The trends of the two were relatively consistent, showing a rapid increase from 2003 to 2014 and a decrease from 2017 to 2020, but changes in APA and AWU were slightly different from 2014 to 2017. The difference may be caused by statistical errors in the data, but the overall trends of the two were positive and the correlation between them was high (r = 0.82). Our analysis showed that APA can represent AWU and can be used to analyze the factors that influence regional GWS.

5.4. Analysis of the Main Influencing Factors of GWSA

Before the implementation of the ecological conservation and restoration project in 2012, APA was the main influencing factor leading to the reduction of GWS in the study area, which was consistent with the traditional cognition. However, after 2012, the main influencing factors of GWS reduction were temperature, evaporation, and FVC, and evaporation had a positive regulating effect on the decrease in GWS, which was inconsistent with traditional cognition. This is because the complex karst landform in the study area leads to a relatively complex groundwater environment, most of which are confined aquifers. However, the analysis results of the main influencing factors of GWS change in this paper were consistent with previous studies’ [15,72].

Furthermore, we only calculated the relative contribution rates of the main influencing factors. In future research, our results could be supplemented with additional quantitative analyses of the absolute contribution rate of all influencing factors. Previous studies lacked any systematic quantitative analysis on the variation in GWS and its influencing factors in this region. In addition, restricted by difficulties in deriving industrial water use, urban domestic water use, and other data, we used only hydro-meteorological, FVC, and agricultural cultivation in our study.

6. Conclusions

In order to quantitatively study the impact of ecological protection and restoration (characterized by FVC) on groundwater resources, firstly, CSR GRACE/GRACE-FO RL06 Mascon solutions (version 02) and the GLDAS hydrological model were used to obtain GWS changes in Yili Valley and Tianshan Mountain of Xinjiang from January 2003 to December 2020, and the results were verified by WGHM GWS and measured groundwater level. Secondly, the VIP method was used to screen the main driving factors of GWS changes in the study area before and after 2012. Finally, the PLSR method was used to establish the quantitative relationship between GWS changes and the main driving factors in the corresponding period around 2012, and the relative influence weights of each main driving factor were determined, and then the quantitative influence degree of FVC on GWS changes was obtained. The results showed the following:

(1)

The GWS inversion results showed a certain accuracy and decreased at a rate of −0.80 cm/yr and −0.75 cm/yr before and after 2012, respectively.

(2)

Before 2012, APA was the main driving factor affecting regional GWS changes. After 2012, the main driving factors were temperature, evaporation, and FVC.

(3)

The relative contribution rates of temperature, evaporation, and FVC after 2012 were 54.72%, 34.59%, and 10.69%, respectively

(4)

FVC (that is, ecological protection and restoration project) played a positive regulating role in the reduction of regional GWS, and the relative influence weight was at least 10%.