Vegetation Restoration Outpaces Climate Change in Driving Evapotranspiration in the Wuding River Basin

Abstract

For the management of the water cycle, it is essential to comprehend evapotranspiration (ET) and how it changes over time and space, especially in relation to vegetation. Here, using the Priestley–Taylor Jet Propulsion Laboratory (PT-JPL) model, we explored the spatiotemporal variations in ET across different time scales during 1982–2018 in the Wuding River Basin. We also quantitatively evaluated the driving mechanisms of climate and vegetation changes on ET changes. Results showed that the ET estimate by the PT-JPL model showed good agreement (R² = 0.71–0.84) with four ET products (PML, MOD16A2, GLASS, FLDAS). Overall, the ET increased significantly at a rate of 3.11 mm/year (p < 0.01). Spatially, ET in the WRB is higher in the southeast and lower in the northwest. Attribution analysis indicated that vegetation restoration (leaf area index) was the dominant driver of ET changes (99.93% basin area, p < 0.05), exhibiting both direct effects and indirect mediation through the Vapor Pressure Deficit. Temperature influences emerged predominantly through vegetation feedbacks rather than direct climatic forcing. These findings establish vegetation restoration as a key driver of regional ET, providing empirical support for optimizing revegetation strategies in semi-arid environments.

1. Introduction

Evapotranspiration (ET) is an important component of the local water cycle and climate, and its measurement is critical to water management, agricultural irrigation, ecosystem health, and economic sustainability. ET consists of the transpiration of vegetation and evaporation from the soil, water bodies, and vegetation canopy [1]. It is an efficient method of water budgeting and energy transport between the ground and the atmosphere [2]. The eco-geographical and water conditions of the region are affected by changes in the feedback of soil heat flux, the feedback of sensible heat flux in the atmosphere, and the distribution ratio of latent heat flux during evaporation [3]. As water becomes more scarce, accurate measurement and modeling of ET are essential for optimizing agricultural production and addressing climate change challenges [4]. As a result, it is necessary to quantify the primary causes of regional ET change in order to reveal the multifactorial compound driving mechanisms of ET evolution.

ET is influenced by numerous factors, including climate, vegetation, topography, and others [5]. Changes in the underlying surface may still exert substantial influence on ET even under relatively stable climatic conditions [6]. According to an analysis of the key drivers of ET in the Yellow River Basin, slope had little impact on ET while precipitation and NDVI dominated ET change [7]. Subsequent interaction analysis showed that the change in ET was the result of multifactor interaction, with the interaction between precipitation and elevation having the most significant impact on ET [8]. According to research findings on how ET reacts to climate change in northern China’s transitional climate zone, there is a relationship between ET and temperature and rainfall, with ET decreasing in arid climates and increasing in humid climates [9]. The analysis of the interannual variation in ET in Chinese terrestrial ecosystems reveals that radiation is the primary driver of ET in the Loess Plateau and Inner Mongolia [10]. In addition to climatic and topographic factors, the impacts of vegetation on ET variations should not be neglected [11]. For instance, studies have revealed a strong positive correlation between NDVI and ET under specific conditions [12]. Since the effects of various factors on ET in a given region are often not easily inferred from studies at other scales or regions, it is evident that the conclusions of studies at different scales frequently differ significantly. For example, in alpine meadows, plant transpiration is a major contributor to ET. Plant transpiration in arid alpine steppe is severely limited due to the scarcity of vegetation cover [13]. Since China is a large country, the main factors influencing ET in various regions vary greatly depending on the climate, topography, and distribution of vegetation in each region [14]. Climatic factors have a greater impact on ET in some places than in others while vegetation change has a greater effect in other places [15]. Therefore, decisions on regional water management and vegetation restoration depend on a better understanding of these changes.

The Wuding River Basin (WRB) is located in the Chinese Loess Plateau, a highly fragile ecosystem with significant soil erosion and a conflict between humans and nature. Since 1999, when the “Grain for Green” project was initiated, the region’s land use pattern and ecological environment have changed dramatically [16]. As a key area for the implementation of soil and water conservation projects on the Loess Plateau, it is a typical wind–water composite erosion area in the midstream section of the Yellow River, and it is emblematic of the sandy and coarse sand regions within the middle reaches of the Yellow River [17]. However, there have been few studies on ET change in this region, with the majority focusing on climatic factors. The effects of vegetation factors on ET have not been well studied, with most studies focusing solely on the direct effects of climate factors and vegetation on ET. Yet the effects of meteorological factors and vegetation on ET are complex and they can be indirectly influenced by other factors in addition to the direct impacts. Therefore, more in-depth research is required in specific areas to better understand the effects of vegetation and climatic factors on ET and to identify the most influential factors.

Therefore, we selected the WRB in the northern Shaanxi region as a case area to simulate ET and analyze the spatiotemporal variation in ET in the WRB from 1982 to 1999, 2000 to 2018, and 1982 to 2018. We used the Priestley–Taylor Jet Propulsion Laboratory (PT-JPL) model to simulate ET spatiotemporal variation in ET in the WRB, and used multiple regression and structural equation models to explore the response mechanism of ET to climate factors and vegetation changes. In order to provide a theoretical basis for water resource management and ecological planning strategies in the WRB, this study aims to (1) investigate the spatiotemporal variations in ET across different scales in the WRB; (2) quantify the relative contributions and elucidate the driving mechanisms of vegetation dynamics and distinct climatic factors on ET changes; and (3) establish a scientific basis for optimizing sustainable water allocation while informing policy formulation for coupled socio-ecological system management.

2. Study Region and Data

2.1. Study Region

The WRB (37°02′31″–38°55′52″N, 108°02′39″–110°34′22″E) is a first-level tributary of the Yellow River with a total length of 491 km, and exhibits northwest-high and southeast-low topography (566–1824 m) (Figure 1). Characterized by a typical arid and semi-arid monsoon climate, the region has an average annual temperature of 9.5 °C and receives 300–550 mm of precipitation [18]. The WRB encompasses diverse geomorphological types: loess hilly and gully areas, the Heyuan Liangjian area, and aeolian sand zones [19]. Dominant land uses include grassland, cropland, and woodland [20]. As a core implementation area of China’s Grain-for-Green Program initiated in 2000, the Loess Plateau had suffered severe vegetation degradation and ecosystem deterioration due to excessive reclamation prior to 2000, triggering intense soil erosion. Post-2000 vegetation restoration efforts dramatically increased forest-grass coverage from 31.6% in 1999 to 67% by 2020 (https://www.forestry.gov.cn/c/www/hm/541377.jhtml) (accessed on 15 February 2025). Situated at the transitional zone between the Mu Us Desert’s southern margin and the Loess Plateau’s northern extremity, the WRB demonstrates representative characteristics. Its selection as the study area stems not only from its typical soil-water conservation demands but particularly from its pronounced vegetation dynamics, enabling clear analysis of interactive effects between vegetation restoration and natural climatic factors on hydrological processes, thereby providing theoretical support for optimizing regional ecological management strategies.

2.2. Datasets

2.2.1. Data Source

The AVHRR_0.05 LAI dataset (Advanced Very High Resolution Radiometer at 0.05° resolution) is central to this study, providing vegetation structural parameters with high spatiotemporal consistency from 1982 to 2018. This dataset drives the PT-JPL model simulations by integrating critical remote sensing metrics, specifically leaf area index (LAI) measurements, to quantify vegetation dynamics and energy exchange processes. The AVHRR_0.05 data are publicly accessible via the Department of Geography, The University of Hong Kong (https://www.glass.hku.hk/download.html) (accessed on 24 April 2024).

Additionally, climate variables (including temperature, precipitation, specific humidity, and downward shortwave and longwave radiation) were sourced from the China Regional Land Surface Meteorological Elements Driving Dataset v2.0 (1951–2020) provided by the National Tibetan Plateau Data Center (https://data.tpdc.ac.cn/zh-hans/data/e60dfd96-5fd8-493f-beae-e8e5d24dece4) (accessed on 19 February 2025). These data were further processed to generate detailed datasets for the study area from 1982 to 2018.

Land cover information was obtained from the China Multi-period Land Use Remote Sensing Monitoring Dataset (CNLUCC), originally published by Xu, Liu, et al., in Resource and Environmental Science Data Platform, and is available via the repository https://www.resdc.cn/DOI/DOI.aspx?DOIID=54 (accessed on 7 October 2024).

We used the monthly net radiation dataset from the Research Center for Eco-Environmental Sciences at the Chinese Academy of Sciences via the National Tibetan Plateau Data Center to assess the impact of climate change and human activities on vegetation carbon sequestration within China. The VPD (Vapor Pressure Deficit) was calculated using raw data in this study:

$$RH=0.263pq{\left[e^{\left({\displaystyle \frac{\mathrm{1.7.67}\left(T-T_0\right)}{T-29.68}}\right)}\right]}^{-1}$$

(1)

$$VPD=0.61078\times e^{{\displaystyle \frac{17.27\times \left(T-273.16\right)}{T-273.16+237.3}}}\times \left(1-RH\right)$$

(2)

where $p$ and $q$ are the atmospheric pressure (Pa) and specific humidity (dimensionless); $T$ and $T_0$ are the temperature (K) and reference temperature (273.16 K). $RH$ is relative humidity.

The Rad (Radiation Absorbed Dose) was calculated by the following formulas:

$$Rad=LRad+SRad$$

(3)

where $LRad$ is downward longwave radiation, $Rad$ is the total radiation, and $SRad$ is downward shortwave radiation.

2.2.2. Data Processing

All of the above climate and vegetation data are normalized to max-min:

$$X_i={\displaystyle \frac{x_i-x_{min}}{x_{max}-x_{min}}}$$

(4)

where $X_i$ denotes normalized data; $x_{max}$ and $x_{min}$ symbolize the maximum and minimum values within the sample data; and $x_i$is the annual data of the sample.

3. Methods

3.1. The PT-JPL Model

In the PT-JPL ET model, the latent evapotranspiration formulation of the Priestley–Taylor equation is the core algorithm. In order to gain a deeper understanding of fluid dynamics and energy equilibrium within ecosystems, the PT-JPL model offers a more accurate description of the surface hydrological cycle [21]. Canopy net radiation is calculated as the difference between surface net radiation and the net radiation of the underlying soil:

$$R_{nc}=R_n-R_{ns}$$

(5)

$$R_{ns}{=R}_{nc}^{ -\mathit{kRnLAI}}$$

(6)

$$R_n=R_{short}-R_{long}$$

(7)

where $R_{nc}$ and $R_{ns}$ are net radiation of canopy and surface soil (W·m⁻²), respectively; $\mathit{kRn}$refers to the extinction coefficient, set at 0.6 (dimensionless) [22];$R_n$represents the surface net radiation (W·m⁻²); $R_{short}$ is the incoming net shortwave radiation (W·m⁻²), which can be calculated by $R_{short}=\left(1-a\right)I_t$;$R_{long}$ is the outgoing net longwave radiation (W·m⁻²), and is calculated as $R_{long}=R_{ld}-R_{lu}$; $I_t$ denotes the entirety of incoming shortwave solar radiation, quantified in watts per square meter (W·m⁻²). $a$ signifies the surface albedo, representing the fraction of shortwave radiation that is reflected back from the surface. Meanwhile, $R_{ld}$ and $R_{lu}$ refer to the downward and upward flux densities of longwave radiation (W·m⁻²), respectively.

According to the PT-JPL model, ET is the total of three main components of canopy transpiration ($E_c$), soil evaporation ($E_{s }$) and interception evaporation ($E_i$).

$$ET=E_c+E_{s }+E_i$$

(8)

$$E_c=\left(1-f_{wet}\right)f_g{f}_t{f}_m\alpha {\displaystyle \frac{\Delta }{\Delta +\gamma }}{R}_{nc}$$

(9)

$$ES=\left(f_{wet}+f_{sm}\left(1-f_{wet}\right)\right)\alpha {\displaystyle \frac{\Delta }{\Delta +\gamma }}\left(R_{ns}-G\right)$$

(10)

$$E_i=f_{wet}\alpha {\displaystyle \frac{\Delta }{\Delta +\gamma }}{R}_{nc}$$

(11)

where $\alpha$ and $\gamma$ are the PT coefficient of 1.26 (unitless) and the psychrometric constant of 0.066 (kPa/°C), respectively. $\Delta$ represents the gradient of the saturated vapor pressure curve (kPa/°C).

$$f_{wet}=R{H}^4$$

(12)

$$f_g={\displaystyle \frac{f_{APAR}}{f_{IPAR}}}$$

(13)

$$f_t=exp\left(1-{\left({\displaystyle \frac{T_{max}-T_{opt}}{T_{opt}}}\right)}^2\right)$$

(14)

$$f_m={\displaystyle \frac{f_{APAR}}{f_{APA{R}_{max}}}}$$

(15)

$$f_{sm}=R{H}^{VPD/\beta }$$

(16)

$$f_{APAR}=b_1\times \left(1-e^{-k_1\times LAI}\right)$$

(17)

$$f_{IPAR}=b_2\times \left(1-e^{-k_2\times LAI}\right)$$

(18)

$$G=R_n\left[{\Gamma }_c+\left(1-P\right)\left({\Gamma }_s-{\Gamma }_c\right)\right]$$

(19)

$$P={\displaystyle \frac{NDVI-NDV{I}_{min}}{NDV{I}_{max}-NDV{I}_{min}}}$$

(20)

In the model formulation, $f_{IPAR}$ quantifies the fraction of photosynthetically active radiation (PAR) intercepted by vegetation canopies, whereas $f_{APAR}$ characterizes the absorbed PAR fraction utilized for photosynthetic processes, and the empirical coefficients b₁ and b₂ are calibrated to 0.95 and 0.9355, respectively. $RH$ and $VPD$ are the relative humidity (%) and the Vapor Pressure Deficit (kPa), respectively; $T_{max}$ and $T_{opt}$ denote the maximum air temperature of plant growth and optimum temperature of plant growth (°C); the parameter $\beta$ represents the response of soil moisture restriction to VPD (range 0–1); $R_n$ is the net radiation (W·m⁻²); ${\Gamma }_s$, set at a value of 0.325, represents the parameter of better bare soil area, whereas ${\Gamma }_c$ (equal to 0.05) describes scenarios ideal for extensive vegetation coverage. The specific parameters follow the previous literature [23]. The flowchart of the PT-JPL model is shown in Figure 2.

3.2. Indicators Used in the Verification of PT-JPL Model Results

For data validation, we selected the statistics of deviation, RMSE (Root Mean Square Error) and MAE (Mean Absolute Error), to validate the results of our simulations [10]. They are calculated by the following equation:

$$RMSE=\sqrt{{\displaystyle \frac{1}{n}}{\displaystyle \sum }_{i=1}^n{\left(S_i-O_i\right)}^2}$$

(21)

$$MAE={\displaystyle \frac{1}{n}}{\displaystyle \sum }_{i=1}^n\left|S_i-O_i\right|$$

(22)

where $S_i$ is the model-simulated value, $O_i$ is the observed value, and n is the sample size.

We use the $R^2$ (coefficient of determination) to indicate the degree of correlation between the PT-JPL simulation results and four different ET products, with values between 0 and 1. The closer to 1, the better the fitting of the model. The calculation equation is as follows:

$$R^2=1-{\displaystyle \frac{{{\displaystyle \sum }}_{i=1}^n{\left(S_i-O_i\right)}^2}{{{\displaystyle \sum }}_{i=1}^n{\left(O_i-\overline{O}\right)}^2}}$$

(23)

where $\overline{O}$ is the mean of the observed values.

3.3. Trend Analysis

Using the statistical methodologies of Mann–Kendall test and Sen’s slope estimator, the spatiotemporal trend analysis across three time periods (1982–1999, 2000–2018, and 1982–2018) was carried out. The Theil–Sen estimator was calculated using Equation (24):

$$Slope=Median\left({\displaystyle \frac{X_m-X_n}{m-n}}\right),\forall m<n$$

(24)

in the formula, $Median()$ calculates the median value; the slope denotes the median estimated slope when m ≠ n; $X_n$ and $X_m$ signify the values of variable X in the n-th and m-th years, with n and m serving as year indicators.

The Mann–Kendall method is used to assess the overall trend of time series data, which can manifest no clear trend, an upward trend, or a downward trend. In the absence of obvious trends, it is assumed that the time-varying data points are independent and homogeneously distributed, with no continuous correlation. The Mann–Kendall test is useful because it does not require a normal distribution of the dataset or assume a linear pattern in trend variations. The test’s validity is based on the assumption that the sampling intervals are long enough to guarantee the independence of measurements taken at different time points, preventing any correlation between them.

$$Z=\left\{\begin{array}{c}{\displaystyle \frac{S-1}{\sqrt{Var\left(S\right)}}} (S>0)\\ 0 \left(S=0\right)\\ {\displaystyle \frac{S+1}{\sqrt{Var\left(S\right)}}} (S<0)\end{array}\right.$$

(25)

$Slope$ and $Z$ values serve distinct purposes in assessing trend magnitude and performing significance tests within the frameworks of Sen’s slope and the MK trend test. A positive $Z$ value indicates an upward trend, whereas a negative Z value indicates a downward trend. |Z| ≥ 1.96 demonstrates that the results have surpassed the 95% significance threshold.

3.4. Partial Correlation

Partial correlation (PC) is a method for determining the correlation between two variables when specific factors are excluded. It is used to determine the level of proximity between elements. When correlation analysis involving three variables reveals interrelationships among all variables, it is imperative to eliminate the impact of the third variable before investigating the association between the initial two variables. When establishing relationships between variables, this approach considers the intricate interplay among them. The partial correlation coefficient between a and y is calculated using Equation (26):

$$r_{ya,b}={\displaystyle \frac{r_{ya}-r_{yb}{r}_{ab}}{\sqrt{\left(1-r_{yb}^2\right)\left(1-r_{ab}^2\right)}}}$$

(26)

in which$r_{ya,b}$ is the partial correlation coefficient that quantifies the relationship between the dependent variable y and the independent variable a while keeping another independent variable b constant. The magnitude of the partial correlation coefficient (PCC) indicates the direct linear relationship between two variables; a higher absolute PCC value indicates a stronger linear association, whereas a lower value indicates a weaker correlation. Within the scope of this research, the partial correlation coefficients for the five drivers of ET were calculated.

3.5. Attribution Analysis

Multiple Regression Analysis (MRA) is used to determine the relative contributions of diverse driving factors of ET in the Water Resource Basin (WRB). This calculation is performed using Equations (27) and (28):

$$Y_{ET}=b_0+b_1{X}_1+b_2{X}_2+\dots +b_i{X}_i+\mu$$

(27)

$$W_{b{a}_i}={\displaystyle \frac{b_i{X}_{i_{t}rend}}{Y_{E{T}_{a_{t}rend}}}}{Y}_{E{T}_{n_{t}rend}}$$

(28)

where $Y_{ET}$ denotes normalized ET; $b_0$ and $b_i$ are constant error and standard regression coefficient; $\mu$ refers to system error. The influence of each driving factor on the trajectory of ET is quantified by$W_{b{a}_i}$; $Y_{E{T}_{a_{t}rend}}$ denotes the observed trend in ET, while the normalized trend of ET is illustrated by $Y_{E{T}_{n_{t}rend}}$.

3.6. The Driving Factors of ET

In this study, the main drivers of ET change during the growing season were identified spatially. When ET began to rise, the positive composite contribution of meteorological factors and vegetation was the most significant driving factor. When ET was declining, the negative factors with the least overall impact were the primary drivers of decreased ET change. Meanwhile, the individual roles of Tem (Temperature), Pre (Precipitation), RN, VPD (Radiation), and LAI in responding to ET changes were determined.

3.7. Path Analysis

Partial least squares structural equation modeling (PLS-SEM) is a method for exploring and testing hypotheses about system relationships. It facilitates a thorough and accurate mediating effect analysis of very complex models. PLS-SEM was selected for its capacity to analyze complex models with multiple latent variables and mediating relationships, particularly suited for exploratory research, and its robustness with small-to-medium sample sizes. The method is able to rigorously test mediation effects on our dataset. The PLS-SEM analysis in this study was performed using Smart PLS 3.3.1. The model had satisfactory goodness-of-fit indicators, with an SRMR of 0.025 [24].

4. Results

4.1. Validation of ET Simulation

To validate the precision of ET simulations generated by the PT-JPL model, we compared them to four different ET products (PML, MOD16A2, GLASS, and FLDAS). PT-JPL shows the strongest agreement with GLASS product (R² = 0.84, RMSE = 38.07 mm/year, MAE = 28.64 mm/year). The lowest agreement is observed with MOD16A2 (R² = 0.71, RMSE = 264.67 mm/year, MAE = 255.65 mm/year) (Figure 3). Agreement with the PML product is notably higher than with the FLDAS product. The PT-JPL model’s ET simulations show a strong positive correlation with the four referenced ET products, with R² values above 0.7. These findings indicate that the PT-JPL model’s ET simulations are highly accurate and capable of precisely delineating the spatiotemporal dynamics of ET variations in the WRB region spanning from 1982 to 2018.

To validate the accuracy of ET simulation using the PT-JPL model, we calculated ET for the watershed based on the water balance method. Specifically, we obtained annual runoff data for the watershed from 1982 to 2018 at Baijiachuan Hydrological Station accessed from the Yellow River Basin hydrological yearbook and precipitation data were sourced from daily measured meteorological data from 1982 to 2018 at four meteorological stations in Yulin, Hengshan, Suide, and Jingbian within the WRB, which were accessed from the China meteorological data network (http://data.cma.cn) (accessed on 17 October 2024). We used the water balance formula ET = P − Q − ΔS to calculate the water balance ET, where P, Q, and ΔS represent precipitation, runoff, and changes in soil moisture within the watershed, respectively. In the Loess Plateau region, ΔS can be negligible. As shown in Figure 4, there is a significant positive correlation between the ET simulated by PT-JPL and the water balance ET, with a determination coefficient (R²) of 0.67 and a root mean square error (RMSE) of 60.633 mm/year, indicating the high precision of the ET simulation using the PT-JPL model and its suitability for subsequent analysis. (Figure 4).

4.2. Spatiotemporal Patterns of ET

4.2.1. Interannual Trends of ET

The ET of the WRB mean time series (1982–2018) increased, with annual average ET ranging from 283~594 mm to an average of 445.37 mm (Figure 5). From 1982 to 2018, ET exhibited a significant upward trend (p < 0.05) with an interannual change rate of 8.48 mm/a. ET fluctuated relatively little between 1993–1997 and 2002–2005, with a notable increase in 1997 and a gradual upward trend since. ET in WRB had minimum and maximum values of 283.56 mm and 593.25 mm in 1989 and 2015, respectively.

4.2.2. Multi-Year Mean Spatial Distribution Patterns of ET

Before 1999, the ET in the WRB ranged from 185.58 to 576.64 mm, with higher values in the southeast and lower values in the northwest (Figure 6a). After 1999, the ET range in WRB increased from 286.52 to 672.76 mm. Therefore, the range of ET was increased throughout the study period, reaching a maximum of 621.74 mm, in the southeast direction of the WRB.

4.2.3. Spatial Distribution, Temporal Trends, and Significance of ET Changes

Figure 7 illustrates the spatial variations in ET over the growing seasons of 1982–1999, 2000–2018, and the entire period of 1982–2018. Between 1982 and 1999, ET decreased across the WRB, with the northwest and southeast regions experiencing the greatest reductions. Notably, the northwest region of Wushen Banner witnessed a sharp decline, with an annual decrease of −2.61 mm. In contrast, elevated ET levels were predominantly found in the southwest and central sectors of the basin, encompassing the western territories of Jingbian County and Hengshan Counties (Figure 7a). Between 2000 and 2018, the study area’s ET changes tended to be upward, with annual growth rates ranging from −1.39 to 14.87 mm. Notably, substantial increases in ET were primarily observed in the eastern and central regions of the WRB, which included counties such as Hengshan, Zizhou, Mizhi, and Suide (Figure 7b). In contrast, a significant decline in ET was primarily observed in the southwestern sector of the WRB, where the lowest rate of change reached −1.39 mm per year. Overall, there was a gradient of decreasing growth rates extending from the northeast to the southwest. Throughout the comprehensive study period from 1982 to 2018, the WRB demonstrated an overall positive trend in ET with annual increments ranging from 3.39 and 16.44 mm, with accelerated growth in the northeastern, central, and southwestern zones.

4.3. Temporal–Spatial Characteristics of LAI and Climatic Factors

4.3.1. The Temporal–Spatial Characteristics of LAI and Climatic Factors

Figure 8 depicts the temporal variations in key meteorological variables and the LAI over the WRB timescale, spanning from 1982 to 2018. Temperature data show a consistent warming trend over this period, with an average annual increase of 0.05 °C. Notably, the coldest year was recorded in 1984, whereas 1998 was the warmest. Precipitation levels oscillated between a minimum of 255 mm and a maximum of 599 mm, with an average depth of 407.36 mm. This series exhibited a significant upward trend (p < 0.05), with an interannual variation rate of 3.55 mm/year. During the observation period, saturated vapor pressure fluctuated significantly (p < 0.05). The yearly change rate is 0.01 kPa, indicating a gradual yet discernible shift. Meanwhile, annual radiation values fluctuated within a narrow band, ranging from 456 to 473 W/m², with an averaging of 464.89 W/m². Despite these fluctuations, a general upward trajectory is discerned, accompanied by an annual growth rate of 0.17 W/m². From 1982 to 2018, LAI values varied between 0.2 mm²/mm² and 1.19 mm²/mm², with an average of 0.59 mm²/mm². This variability highlights the dynamic changes in vegetation cover within the basin. LAI volatility has increased (p < 0.05) over the years, with an interannual rate of change of 0.02 mm²/mm² annually. The year 2018 saw the highest LAI since 1982, measuring 0.6 mm²/mm² above the average. On the other hand, 1989 saw the lowest LAI, measuring 0.39 mm²/mm² below the average.

4.3.2. Multi-Year Mean Spatial Distribution Patterns of LAI and Climatic Factors

The spatial distribution and dynamics of the LAI and various climatic factors (where RN stands for the annual cumulative value) within the WRB across three distinct time periods are depicted in Figure 9, Figure 10 and Figure 11. Prior to 1999, Tem and net RN had higher values in the southeastern part of the WRB and lower values in the southwest. Specifically, the southeastern areas, which include Suide County, Zizhou County, Mizhi County, and parts of Hengshan County, saw the highest recorded temperatures of 11.38 °C, and the highest net radiation of 447.41 W/m². This pattern of higher values in the southeast and lower values in the southwest persisted throughout the study period. The minimum precipitation of 318.57 mm, the minimum LAI of 0.07, and the maximum VPD of 0.46 kPa occurred in the northwest of the WRB. The Pre and LAI were higher in the southeastern margin of the WRB, smaller in the central region, and gradually transitioned to the northwest margin. The areas with higher median values were those east and south of Suide County, Mizhi County, and Zizhou County, while the LAI was lower in Wushenqi and other areas, and the VPD showed a smaller distribution in the southwest and a larger distribution in the northwest and southeast (Figure 11e).

4.3.3. Spatial Trends and Significance of Changes in LAI and Climatic Factors

The changes in meteorological conditions and LAI in the WRB over three different time periods are shown in Figure 12, Figure 13 and Figure 14. Before 1999, Tem and RN increased across the entire WRB. Additionally, Pre had a negligible decrease in 96.32% of the basin, with southeast and central areas showing the largest decline. On the other hand, rainfall increased in the southwest and northwest, particularly in the western part of Jingbian County and the northern region of Wushenqi. The precipitation change rate ranged from −7.56 mm/a to 2.41 mm/a. After 1999, the whole WRB experienced an increase in precipitation, while more than half of the region saw a decrease in VPD (65.19%) and RN (85.63%), 31.5% of the region saw a decrease in temperature, 98.98% saw a significant increase in LAI, and only a small part of the southern region experienced a significant degradation in vegetation. During the entire study period, the WRB experienced an increase in Tem, Pre, and RN, while the southwest saw a decrease in VPD and the northwest saw a decrease in LAI. Moreover, 99.93% of the WRB experienced significant greening, with an increase in LAI in the southeast and a decrease in LAI in the northwest, especially in the north of Wushenqi, with the LAI change rate ranging from −0.01 to 0.07 mm/a. Throughout these three time periods, the VPD of the entire WRB remained stable with a negligible rate of change.

4.4. Changes in ET Caused by Vegetation Restoration and Climate Change

We used a partial correlation analysis to examine the impacts of various meteorological factors and LAI on ET (Figure 15, Figure 16 and Figure 17). Before 1999, approximately half of the regions (47.73%) showed a positive correlation between Tem and ET, with the majority located in the WRB’s northern and northwestern parts. In contrast, Pre and VPD did not have a significant effect in most areas within the WRB. However, ET had a significant positive correlation with LAI across the entire WRB. After 1999, ET and Tem were negatively correlated in the southeast and positively correlated in the northwest, while VPD was the opposite. Pre and RN had no significant effect in most areas of the WRB, and ET and LAI were significantly positively correlated in most areas. Compared to before 1999, the area with a significant positive correlation with LAI increased to 97.57%. A partial correlation analysis of ET and meteorological factors, as well as LAI, was conducted over the 37-year period from 1982 to 2018. ET was significantly correlated with Tem, RN, and LAI in certain areas of the WRB (p < 0.05). The correlation between ET and LAI was significantly positive in 99.93% of the WRB, with 71.42% of the areas showing a significant positive correlation between ET and Tem, primarily concentrated in the northwest of the WRB. Furthermore, 79.43% of the areas exhibited a positive correlation between ET and RN. There was no significant correlation observed between ET and Pre or VPD.

Figure 18, Figure 19 and Figure 20 illustrate the spatial distribution of the contributions of meteorological factors and LAI to changes in ET within the WRB over three different time periods. Among these factors, LAI has the greatest impact on ET. The LAI has a direct influence on ET that exceeds 8 in the central and southeastern regions of the WRB. From 1982 to 1999, there were no significant spatial variations in the contribution rates of Tem, Pre, and VPD to ET within the WRB, and their combined contribution to ET was minimal. RN and LAI contribute slightly more to ET, while Tem mainly concentrates at −2 to 2 °C/a. LAI contributes the most to ET, exceeding 4/a in most regions and up to 6–10/a in the Midwest and Southeast. After 1999, LAI’s contribution to ET decreased slightly, but it remained the dominant factor in ET change, with a contribution rate of 0–4/a to the central part of the WRB and more than 4/a to the southeast. Throughout the entire study period from 1982 to 2018, LAI contributed the most to ET, with a contribution rate of more than 8/a in the southeast, including Suide County, Zizhou County, and Mizhi County. Tem was the second-largest contributor, with a significant regional contribution rate ranging from 4 to 10 °C/a, and accounting for 15.3% of the area, mainly in the northern part of the WRB.

Figure 21 depicts the spatial distribution of the dominant factors influencing ET change in the WRB. Before 1999, vegetation factors were the main driving factors affecting ET change in the WRB, accounting for 92.53% of the total. Tem and RN also had an impact on the ET change in the WRB, and Pre had the minimum impact (only accounting for 0.29%). After 1999, the control effect of vegetation on ET in the WRB decreased by 2.18%, but it remained the dominant factor of ET change, followed by temperature change, which was 5.87% higher than before 1999. In conclusion, the ET changes were observed from 1982 to 2018, and vegetation was the main controlling factor of ET in the WRB, determining the ET changes in the WRB. Only a few areas in the northwest had ET changes due to temperature (Figure 21c).

We used path analysis to better understand the driving mechanisms of ET in the WRB, as influenced by various environmental factors and LAI, and the final fitted SEM is presented in Figure 22. The fitting model performed satisfactorily, with an SRMR of 0.025. The SEM with selected variables (Tem, Pre, RN, VPD, and LAI) explained 95.6% of the variability in ET (R² = 0.956). Since 2000, LAI’s direct impact on ET has increased significantly, from 0.745 to 1.059. From 1982 to 1999, Pre had a positive effect on ET (0.018), but after 2000, Pre exhibited a negative effect on ET (−0.144). Overall, LAI was the best natural variable at reflecting ET changes, with a total effect of 0.944. Additionally, as a mediating variable, LAI had an indirect effect on ET due to the influence of Pre on LAI (0.559). Temperature (Tem) also demonstrated an indirect effect (0.471) on ET changes through LAI (Tem → LAI → ET, Tem → LAI → VPD → ET). In addition to the direct effect of Pre on ET, we also found that Pre indirectly affected ET changes through RN, LAI, and VPD (Pre → RN → ET, Pre → LAI → ET, Pre → VPD → ET), with impact coefficients of −0.323, 0.559, and −0.252, respectively. There was little correlation between TEM, Pre, and VPD and ET, and the total impact coefficients were −0.015, −0.068, and 0.068, respectively. The increase in RN exerted a significant positive effect on ET (p < 0.05). Concurrently, LAI indirectly affected ET changes via VPD (LAI → VPD → ET), with an impact coefficient of 0.003. Tem positively affected LAI, VPD, and RN, with total impact coefficients of 0.471, 0.759, and 0.79, respectively, indicating that the increase in Tem positively influenced the changes in LAI, VPD, and RN.

5. Discussion

5.1. Response Mechanism of ET to Vegetation Change

The average annual ET in the WRB on the Loess Plateau showed a significant upward trend from 1982 to 2018, ranging from 283.56 to 593.25 mm. This trend is consistent with the findings of studies in the Loess Plateau, where vegetation restoration is generally considered to be the dominant driver of ET increase [7]. For example, Guo et al. found that the evapotranspiration of vegetation on the Loess Plateau increased at a rate of 4.929 mm/year from 1981 to 2020 (p < 0.05), which further verified the long-term upward trend of ET [25]. In addition, Wang et al. showed that from 2000 to 2015, the ET of cultivated land on the Loess Plateau showed a significant upward trend (p < 0.05), which was mainly driven by agricultural intensification (green farmland) [26]. In terms of spatial distribution, this study showed that the annual average ET in the WRB of the Loess Plateau showed a gradient of high in the southeast and low in the northwest (185.58–576.64 mm), which was consistent with the spatial pattern of decreasing cultivated land transpiration from south to north (200–300 mm) in Wang et al. [26].

A variety of analytical techniques, including trend analysis, partial correlation assessment, multiple regression examination, contribution dissection, and the application of structural equation modeling, were used to compare the impacts of vegetation dynamics on ET within the WRB. This study’s findings indicate that the lowest LAI between 1982 and 2018 occurred in 1989, which may have been caused by extensive deforestation, which resulted in substantial vegetation loss and subsequent soil erosion. Additionally, the same year had the lowest ET level, which may have been directly related to the exceptionally low LAI within the WRB at that time. While climatic factors set the broad spatial template for ET, LAI acts as the primary modulator of ET variability at the basin scale. From the standpoint of land–atmosphere mutual feeding, LAI significantly improves the ET change in the WRB [27]. Additionally, vegetation restoration increases ET, and makes the surface atmosphere wetter, which leads to the strengthening of water vapor convergence and the increase in cloud cover, both of which speed up the recirculation of precipitation. Eventually, some of the increased ET will eventually fall to the surface as precipitation [28]. Areas with high ET values are concentrated in the southeastern part of the study area, specifically in areas with lower latitudes. The government’s “Grain for Green” policy since 1999 is responsible for this phenomenon, as it has significantly increased vegetation cover in many areas and contributed to ecological restoration of the region [29]. Precipitation occasionally follows an increase in vegetation, because vegetation can enhance land surface’s ability to retain and protect water, which can facilitate water cycle processes and raise ET [27,30]. Additionally, afforestation can also affect water retention by improving soil quality through carbon sequestration [31,32]. From 1982 to 2018, the correlation between LAI and ET was the strongest in the WRB, with a significant positive correlation found in 99.93% of the areas (p < 0.05). While LAI and ET trends are positively correlated in most regions, localized discrepancies likely reflect interactions with climatic factors (e.g., precipitation, radiation) or vegetation functional traits. For instance, in arid southwestern areas, LAI increases may not fully translate to ET enhancement due to soil moisture constraints, highlighting the complex trade-offs between vegetation restoration and hydrological processes.

LAI is a direct reflection of the extent and density of vegetation coverage. As the LAI rises, it exacerbates the process of water vapor being released into the atmosphere via leaf surfaces. As a result, plants must increase water uptake from the soil to accommodate the heightened demands stemming from the amplified ET process [2]. Meanwhile, increasing LAI allows for more efficient light capture for photosynthesis, which increases transpiration and, as a result, releases more water into the atmosphere [2]. But prior to the implementation of “Grain for Green” project, the WRB suffered persistent soil erosion, which made it challenging for plants to obtain water. Plants use a variety of physiological and morphological adaptation mechanisms to defend themselves against such arid circumstances [33]. For example, plant roots may become more developed to enhance their ability to absorb water from deeper layers of soil; leaves may become smaller or have a waxy layer to reduce water loss due to transpiration. Some plants may even go dormant during drought, resuming growth when conditions improve. These adaptive changes reflect the survival strategies adopted by plants in response to adverse environmental conditions. However, if soil erosion exceeds plant life’s adaptive thresholds, even the hardiest species may struggle to maintain normal growth and reproduction. This could lead to a decline in vegetation health and decrease overall ecosystem functionality [34]. Therefore, reestablishing the ecology and managing the water resources of the Loess Plateau and WRB require reducing soil erosion and implementing science-based vegetation restoration techniques.

5.2. Response Mechanism of ET to Climate Change

The effects of climate factors on ET in the WRB area were investigated using trend analysis, partial correlation analysis, multiple regression analysis, contribution analysis, and structural equation modeling. According to this study, the ET in the WRB area averaged 445.37 mm and peaked at 593.25 mm in 2015. According to the China Climate Bulletin and the climate data from the same period, the ET in the WRB was significantly impacted by climate change in 2015 [35]. In 2015, climatic indicators such as Tem, RN, and VPD all exceeded historical averages, and these changes probably raised ET and affected the local water cycle and soil moisture regime [36]. A rise in VPD signifies increased evaporation of water vapor from the soil and transcription to water vapor from plants. This escalation significantly amplifies the extent to which vegetation is subjected to drought stress conditions, and at the same time, plants close their stomata in order to reduce water loss, thereby reducing plant photosynthesis and limiting vegetation growth, resulting in a decrease in LAI [37]. Furthermore, the ecological restoration initiatives in the WRB may have also contributed to the observed phenomena. These initiatives typically include vegetation restoration projects intended to improve the ecological environment and reduce soil erosion. Nonetheless, revegetation initiatives may unintentionally worsen soil moisture depletion and influence ET dynamics if they introduce water-intensive plant species or ignore the region’s natural water-carrying capacity [29]. Harsh climatic, low vegetation cover, poor land quality, and the lack of precipitation in the interior could all be contributing factors to the lower ET in the northwest of the WRB. Poor land and unfavorable climatic conditions (like high temperatures and drought) can restrict plant growth, slow down the flow of water from the soil to the atmosphere, and lower overall ET activity. The interannual temperature rate of change was 0.05 °C/a, with the lowest temperature in 1984 and the highest in 1998. This trend may be due to the increase in greenhouse gases (CO₂, CH₄, and N₂O) and global warming preceding the policy of “Grain for Green”, which led to an increase in WRB temperatures. There was a positive correlation between Tem and ET in 71.42% of the WRB, and high temperatures promoted evaporation and plant transpiration [37]. One possible explanation for the negative correlation between ET and Tem in some areas is the lack of soil moisture supply in the region. Under conditions where soil moisture is restricted, an increase in temperature may lead to a further decrease in soil moisture, limiting evaporation. In addition to this, when soil moisture is insufficient, plants reduce water loss by closing the stomata, thereby reducing transpiration [38].

The annual rainfall distribution of the WRB showed a gradual upward trend, with an average precipitation of about 407.36 mm. It peaked at around 598.99 mm in 2017. The observed increase in precipitation levels may be due to the increase in temperature promoting transpiration of plants, which subsequently leads to an increase in rainfall. The Loess Plateau, characterized by its distinctive soil composition and historical overexploitation, has long faced serious soil erosion problems [39]. Soil erosion not only alters the landscape but also causes many ecological and environmental problems. The correlation between RN and ET was positively correlated in most of the WRB, suggesting that increased solar radiation generally energizes water molecules, facilitating their transition from a liquid to a gaseous state and thereby boosting evaporation. However, under specific circumstances, the RN can suppress ET in areas surrounding the WRB. Intense solar radiation, driven by high temperatures and low relative humidity, accelerates evaporation rates. If evaporation surpasses the available water supply, it leads to surface water depletion, subsequently inhibiting further evaporation [13,38]. As the landscape changes, the expansion of exposed loess surfaces enhances direct solar radiation absorption, leading to elevated surface temperatures and intensified radiant energy. Consequently, soil moisture evaporates more readily, hastening the depletion of soil moisture content [40]. Increased radiation accelerates soil moisture depletion and limits the ability of plant roots to absorb water, which in turn influences transpiration—a clear indicator of plants’ self-protective mechanisms. There was no significant correlation between ET and VPD, and VPD had a negative contribution to ET in half of the regions, partly because the lower VPD affected the transition of water from the liquid to gaseous state. Water molecules are less prone to transitioning into vapor at low humidity, reducing water evaporation [41].

In the PT-JPL model, vegetation changes (e.g., LAI) directly affect ET mainly through canopy net radiation (Rnc) (Equation (9)). However, climatic factors (e.g., temperature, precipitation) can indirectly affect ET by regulating vegetation growth. For example, increased temperature may promote photosynthesis in vegetation, which in turn may increase LAI. Increased precipitation is likely to improve soil moisture conditions and support vegetation expansion. These processes are partially captured in the model through the dynamic changes in the LAI. In this study, the structural equation model (SEM) was used to further verify this mechanism, and the results showed that temperature had a significant indirect effect on ET through LAI (Tem → LAI → ET, total effect coefficient 0.471), indicating that the influence of climate factors on ET was indeed partially realized through vegetation mediation.

5.3. Implications for Water Resource Management and Vegetation Restoration

Habitat quality is the foundation of biodiversity and the key to ensuring regional ecological security. In the context of promoting the construction of ecological civilization, achieving a balance between ecological environmental protection and social and economic development has become the core goal of land use transformation. As a pilot area for comprehensive management and sustainable development of the water environment in the national basin, the WRB is facing new requirements for high-quality development, which puts forward higher standards for the ecological environment in the region, so it is important to deeply explore the causes of habitat quality changes and the driving factors behind them to promote local biodiversity and improve the overall ecological environment.

As one of the driving factors that may affect habitat quality change, previous studies have shown that the increase in vegetation does exacerbate water uptake and transpiration, which may further lead to water shortage in the Loess Plateau, but the response to vegetation change in different regions is different. Quantitative findings on this change are essential to accurately guide vegetation layout adjustments and avoid the risk of water scarcity. Therefore, the implementation of water conservation measures and careful planning of water resources allocation is an extremely important task to alleviate regional water stress. To achieve this, ecological conservation and water management can be reconciled through the implementation of a variety of management strategies. For example, afforestation not only enhances soil water retention, but also promotes groundwater recharge, benefiting downstream users and improving the water use efficiency of existing vegetation [30,42].

In addition, with the continuous promotion of China’s key ecological projects, such as the control of sand and dust sources in the Beijing–Tianjin area, the construction of the vast three-north shelterbelt system, and the pilot project of the satisfactory land conservation area, China’s ecological restoration and governance work has become increasingly stable. The historical trajectory of land desertification has shifted from an era of expansion to one marked by decline. In recent years, the cultivated land area in the WRB has gradually decreased while the forest and grassland area has increased steadily, which directly reflects the positive role of the current policy in improving the ecologically fragile areas of the basin (such as land desertification and soil erosion), but the long-term effects of different forest and grass construction on the land are unknown, so managers need to have a deep understanding of the local ecological environment, select suitable plant species to optimize the current vegetation structure, and find the best balance between ecological and hydrological benefits. In order to achieve the purpose of protecting the ecological environment and making efficient use of water resources [43], in the future practice of vegetation restoration, we ought to opt for water-efficient and ecologically vital plant species, taking into account their ET traits and the local environmental context. These species should serve as the foundational elements in restructuring regional vegetation patterns and bolstering their ecological functionalities. The availability of water largely determines the variation in ET rates between different land types [39,44]. Therefore, in arid regions characterized by scant rainfall, it is imperative to meticulously select plant species with high water consumption and refrain from extensive planting of such species. Particularly in locales boasting dense vegetation and expansive green spaces, those plants exhibiting robust water absorption capabilities should be systematically phased out to safeguard limited water resources.

5.4. The Uncertainty and Limitations

This study carries a degree of uncertainty. First, in the PT-JPL model, the key inputs used to estimate ET and the accuracy of each component cover multi-source data such as land use types, meteorological observations, and remote sensing imagery. Bilinear interpolation is used to align the resolutions of different datasets, enhancing the spatial resolution of previously low-resolution remote sensing and meteorological datasets. Nonetheless, this procedure could potentially introduce supplementary uncertainties stemming from the inherent simulations [1]. Secondly, the PT-JPL model mainly considers canopy interception, vegetation transpiration, bare soil evaporation, deep infiltration, and soil water storage changes, but does not consider lakes, reservoirs, and water resource utilization behaviors (such as agricultural water supply, water conservancy projects, and the construction of dams and reservoirs) [45,46]. In addition, some scholars have found that large changes in China’s lakes, especially in the northern region, will result in substantial alterations to regional hydrological cycle elements, encompassing both groundwater levels and runoff patterns. This is because lakes and reservoirs can influence the size and timing of runoff by regulating the amount of water they store [47,48]. Anthropogenic processes, such as agricultural, industrial, and domestic water withdrawals, may have considerable impacts on water resources, directly or indirectly altering the hydrological cycle [48], and these processes are ignored in this study. However, this study will provide fresh insights into how climate change and vegetation restoration influence ET in arid and semi-arid regions, which has far-reaching implications for the sustainable management of regional water resources.

6. Conclusions

Based on the PT-JPL model, this study conducted a comprehensive spatiotemporal analysis of ET trends, partial correlation, multiple regression, and contribution analysis in the WRB in three time periods from 1982 to 1999, 2000 to 2018, and 1982 to 2018. It quantified the contribution of climatic factors and vegetation factors (LAI) to ET and showed the correlation between these drivers and ET using partial correlation analysis. Key findings include the following:

(1) Over time, ET (Slope = 8.48 mm/a, p < 0.05), Tem (Slope = 0.05 °C/a, p < 0.05), Pre (Slope = 3.55 mm/a, p < 0.05), VPD (Slope = 0.01 kpa/a, p < 0.05), RN (Slope = 0.17 W/m²/a, p < 0.05), LAI (Slope = 0.02 m²/m²/a, p < 0.05). Spatially, the ET increased from northwest (258.01 mm) to southeast (621.14 mm), Tem decreased from southeast (11.78 °C) to southwest (5.69 °C), Pre decreased from southeast (500.743 mm) to northwest (334.887 mm), RN decreased from southeast (480.16 W/m²) to southwest (450.3 W/m²), VPD decreased from northwest (0.47 kpa) to southwest (0.29 kpa), and LAI increased from northwest (0.16 m²/m²) to southeast (1.17 m²/m²).

(2) Multi-variate analysis revealed that over the 37-year study period, 99.93% of the areas in the WRB were significantly positively correlated with LAI, 71.42% of the areas exhibited a significant positive correlation with TEM, mainly concentrated in the northwest of the WRB, and 79.43% of the areas were positively correlated with RN. ET was the main driving factor of ET change, TEM and RN also had a certain impact on ET change in the WRB, and Pre had the least impact.

(3) In terms of spatial distribution, the LAI mainly controlled the ET fluctuation in most areas of the WRB, accounting for about 97.14% of the survey area. The change in ET affected by temperature accounted for 2.82% of the study area, predominantly situated in the central and northwestern parts of the WRB. At the same time, RN and VPD factors jointly affected the ET dynamics in the western and northern WRB.

The finding of this study identifies vegetation restoration as the dominant factor driving the increase in ET, providing a scientific basis for optimizing vegetation layouts to balance ecological benefits with water sustainability. The findings underscore the necessity of incorporating vegetation dynamics into water conservation policies, particularly in arid and semi-arid regions such as the Loess Plateau, to mitigate the risks of water scarcity and enhance ecosystem resilience under changing climatic conditions. These insights are crucial for guiding regional water resource management and ecological restoration strategies, highlighting the pivotal role of sustainable vegetation restoration in harmonizing hydrological cycles and environmental conservation within fragile ecosystems.