Vapor Pressure Deficit Thresholds and Their Impacts on Gross Primary Productivity in Xinjiang Arid Grassland Ecosystems

Abstract

Understanding vegetation responses to atmospheric drought is critical for arid ecosystem management under climate change. However, the threshold of the response mechanism of grassland in arid regions to atmospheric drought remains unclear. This study investigates how vapor pressure deficit (VPD) regulates grassland gross primary productivity (GPP) in Xinjiang, China, using MODIS and other multi-source remote sensing data (2000–2020). The results show intensified atmospheric drought in central Tianshan Mountains and southern Junggar Basin, with VPD exhibiting a widespread increasing trend (significant increase: 15.75%, extremely significant increase: 4.68%). Intensified atmospheric drought occurred in the central Tianshan Mountains and southern Junggar Basin. Integrated analyses demonstrate that VPD has a dominant negative impact on GPP (path coefficient = −0.58, p < 0.05), primarily driven by atmospheric drought stress. A ridge regression-derived threshold was identified at 0.61 kPa, marking the point where VPD transitions from stimulating to suppressing productivity. Spatially, 58.75% of the total area showed a significant increase in GPP. These findings advance the mechanistic understanding of atmospheric drought impacts on arid ecosystems and inform adaptive grassland management strategies.

1. Introduction

Climate change is profoundly reshaping ecosystem structure and function globally. Vapor pressure deficit (VPD), defined as the difference between the saturation vapor pressure and the actual vapor pressure of the air, serves as a key measure of atmospheric aridity. VPD directly impacts vegetation water use efficiency by regulating stomatal behavior and carbon assimilation. However, threshold-dependent responses of grassland productivity to VPD remain poorly quantified, particularly in arid environments, significantly limiting the development of effective adaptive management strategies.

Prior studies have indicated that elevated VPD is generally associated with heightened atmospheric aridity, while lower VPD values are more common in humid environments [1,2]. Global observations have shown that when VPD exceeds a certain threshold, it typically leads to a decrease in plant stomatal conductance, thereby inhibiting photosynthetic efficiency [3]. This physiological response may partially offset the carbon uptake benefits resulting from rising atmospheric CO₂ concentrations [4,5], leading to complex effects on surface energy balance. Notably, the coupling between VPD and soil moisture through land–atmosphere feedback mechanisms further amplifies vegetation stress during compound drought–heatwave events [6]. Conversely, hydraulic diversity in plant communities can buffer ecosystem-scale fluctuations in water fluxes under drought conditions [7].

While current academic consensus establishes VPD as a primary regulator of photosynthetic processes through stomatal modulation [8], its ecological impacts exhibit critical biome-specific and context-dependent variations. Seasonal VPD fluctuations in mid-to-high latitudes enhance autumn photosynthetic capacity under low atmospheric aridity [9], yet these benefits are counterbalanced by spatially heterogeneous interactions with temperature (TEM) and soil moisture [10]. Comparative studies in Central Asia suggest that forests exhibit heightened VPD sensitivity compared to grasslands and shrublands, which are predominantly constrained by soil moisture [11]. This biome-specific response is further complicated by the regulatory effects of VPD on grassland ecosystems: direct stomatal control of transpiration and indirect modulation via soil moisture–radiation feedback [12]. Recent evidence highlights a climatic dichotomy: soil moisture dominates vegetation water constraints in arid/semi-arid zones, whereas VPD-driven limitations prevail in humid regions [13]. Crucially, threshold-dependent responses emerge when VPD exceeds regionally adaptive limits through synergistic interactions with TEM and soil moisture [10], but such thresholds remain undefined for Xinjiang’s hyper-arid grasslands. Although grassland ecosystems account for 32% of global natural vegetation and play a crucial role in water vapor exchange and energy balance, systematic studies investigating the relationship between grassland productivity and VPD in arid regions (e.g., Xinjiang) remain limited [14]. This gap is critical given Xinjiang’s unique hyper-arid conditions, where vegetation thresholds to drought-causing factors vary markedly by type [11].

Given that the continuous rise in VPD may exacerbate ecological degradation risks in arid areas [15], particularly under high-emission scenarios where VPD dominates drying trends over 20–27% of global land [16], and considering that elevated atmospheric demand for water (reflected by high VPD) can limit forest carbon uptake as severely as soil drought [17], understanding the environmental effects of VPD in complex topographical regions like Xinjiang is imperative, yet remains poorly understood. Our primary objectives are to analyze the spatial and temporal evolution characteristics of VPD, elucidate its regulatory mechanisms on grassland gross primary productivity (GPP), and critically, to identify and quantify the specific VPD thresholds that impact GPP and assess the changes in their influence on grassland GPP. Through a comprehensive analytical framework combining linear trend analysis, partial correlation assessment, ridge regression modeling, and structural equation modeling (SEM), we systematically evaluated meteorological drivers of GPP variations. By revealing the spatiotemporal variation patterns of VPD, clarifying that VPD is a key meteorological factor regulating vegetation productivity, its ecological effects, and defining the critical VPD thresholds for Xinjiang’s grasslands, this study fills a key regional research gap. It provides a vital theoretical and scientific basis for developing differentiated ecological management strategies to enhance resilience in this vulnerable and ecologically significant arid region.

2. Materials and Methods

2.1. Study Area

Xinjiang, located in the heart of the Eurasian continent (73°40′ E–96°18′ E, 34°25′ N–48°10′ N), encompasses a total area of 1.6649 million square kilometers. It forms a closed inland basin system that exemplifies a typical arid ecological unit globally (Figure 1a). The unique topographical pattern, characterized by the ‘Three Mountains Surrounding Two Basins’—the Altai Mountains, Tianshan Mountains, and Kunlun Mountains encircling the Junggar Basin and the Tarim Basin—creates a significant topographic barrier that obstructs the transport of maritime moisture [18,19,20]. This results in an extremely arid climate with average annual precipitation (PRE) below 200 mm across most areas. This PRE exhibits a notable gradient, ranging from 150–250 mm in northern Xinjiang to 25–100 mm in the south [21,22].

The pronounced spatial differentiation in water and heat drives the vertical zonation of the mountain–oasis–desert system, where natural grasslands, covering 31.5% of the total area (Figure 1b), function as ecological corridors, playing an irreplaceable role in maintaining the regional carbon–water cycle balance [23]. Notably, influenced by the coupling effects of westerly wind circulation and local topography, PRE is distributed unevenly throughout the year (with over 60% occurring in summer), which, combined with intense land surface evapotranspiration processes (with average annual potential evapotranspiration reaching 1800–2200 mm), leads to an extreme sensitivity of grassland productivity to climatic fluctuations. In the context of global warming, the rate of TEM increase in this region significantly exceeds the global average [22], further exacerbating the competition for water resources in vegetation. Consequently, VPD, a key driver of plant water stress and evapotranspiration, exhibits a distinct spatial pattern characterized by lower values in mountainous areas and higher values in plains/basins, with a weaker increasing trend observed in the mountains compared to the plains/basins. This pattern is attributed to the combined effects of topography, climate, and vegetation type. Furthermore, influenced by the zonal differentiation of regional hydrothermal conditions and vegetation types, GPP of vegetation across Xinjiang displays significant spatial heterogeneity, generally manifesting as higher values in Northern Xinjiang and lower values in Southern Xinjiang. These contrasting spatial patterns of VPD and GPP highlight the complex interplay between atmospheric demand, water availability, and ecosystem productivity in this arid region, making the grassland ecosystems a critical indicator of climate–ecological coupling responses in arid areas.

2.2. Data Sources

The data sources for this study encompass multidimensional datasets, including remote sensing observations and meteorological reanalysis products. All data processing follows a unified spatiotemporal framework to ensure the comparability of results. Prior to analysis, initial quality control checks were performed on all input datasets to identify and exclude obvious outliers and significant missing values. The GPP dataset has a temporal resolution of 8 days. To construct continuous time series, gaps in the 8-day GPP data were filled using linear temporal interpolation. These data were then aggregated into monthly means using arithmetic averaging and further used to calculate annual cumulative values using R 4.41. Grassland type data were extracted from the MCD12Q1 global land cover product (NASA EOSDIS Land Processes DAAC, USGS Earth Resources Observation and Science Center, Sioux Falls, SD, USA), which uses the International Geosphere-Biosphere Program (IGBP) classification system. We selected five categories: “Closed shrublands”, “Open shrublands”, “Woody savannas”, “Savannas”, and “Grasslands”. These categories were spatially overlaid using ArcGIS 10.8 (Esri, Redlands, CA, USA) to generate a composite grassland distribution map. Climate variables were obtained, as detailed in

Table 1. Description of datasets used in this study.

Data	Year	Original Resolution	Data Sources
GPP	2000–2020	500 m	https://lpdaac.usgs.gov/products/mod17a2hv006 (accessed on 14 December 2024)
Grassland range	2021	500 m	https://lpdaac.usgs.gov/products/mcd12q1v006 (accessed on 14 December 2024)
LAI	2000–2020	0.05°	http://globalchange.bnu.edu.cn/research/laiv6 (accessed on 21 June 2025)
DEM	2020	90 m	https://zenodo.org/records/4700264 (accessed on 14 December 2024)
TEM	2000–2020	0.50°	https://crudata.uea.ac.uk/cru/data/hrg/cru_ts_4.06 (accessed on 14 December 2024)
PRE	2000–2020	4000 m	https://www.climatologylab.org/terraclimate.html (accessed on 14 December 2024)
SSR	2000–2020	4000 m	https://www.climatologylab.org/terraclimate.html (accessed on 14 December 2024)
VPD	2000–2020	4000 m	https://www.climatologylab.org/terraclimate.html (accessed on 14 December 2024)

Note: Variable names and corresponding abbreviations. Gross primary productivity (GPP), leaf area index (LAI), digital elevation model (DEM), temperature (TEM), precipitation (PRE), downward surface shortwave radiation (SSR), and vapor pressure deficit (VPD).

. VPD, downward surface shortwave radiation (SSR), and air temperature (TEM) data were averaged monthly using arithmetic averaging to mean annual mean values, while precipitation (PRE) data were summed to calculate annual cumulative totals using R 4.41. All datasets underwent spatial resampling using a bilinear interpolation method in ArcGIS 10.8 to unify the spatial resolution to 500 m. Rigorous pixel-scale matching was then conducted in ArcGIS 10.8 to eliminate spatial heterogeneity errors caused by differences in data sources and projection systems. The temporal coverage of all datasets spans from 2000 to 2020.

2.3. Data Analysis Methods

2.3.1. Rate of Change Analysis

Linear regression is a statistical method used to establish a linear relationship between a dependent variable and one or more independent variables for prediction and analysis. This study employs linear regression to explore the spatiotemporal trends in grassland productivity in Xinjiang and its relationship with meteorological factors (VPD, PRE, TEM, and SSR) from 2000 to 2020. By analyzing the data over this period, we aimed to reveal the spatiotemporal dynamics of these factors in Xinjiang [24].

The linear regression trend for a single meteorological factor is calculated using the following equation:

$$slope={\displaystyle \frac{n\times {\displaystyle {\sum }_{i=1}^{\mathrm{n}}i\times x_i-\left({\displaystyle {\sum }_{i=1}^n{x}_i}\right)\left({\displaystyle {\sum }_{i=1}^{n}i}\right)}}{n\times {\displaystyle {\sum }_{i=1}^n{i}^2-{\left({\displaystyle {\sum }_{i=1}^{n}i}\right)}^2}}}$$

(1)

In the equation, “ Slope” represents the trend of the variable; “ n” denotes the number of years (2000–2020, 21 years); $x_i$ refers to the value of the meteorological factor for the i-th year; and i takes values from 1 to 21, corresponding to the years of study in the study area. Model parameter estimation and significance assessment were performed using R 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria). By estimating the model parameters, we quantified the trend in changes in the meteorological factors and assessed their significance.

To more clearly identify temporal trends, we performed an F-test on different pixel values in R 4.4.1. The formula is as follows:

$$U={{\displaystyle {\sum }_{i=1}^n\left(\widehat{y}-\overline{y}\right)}}^2$$

(2)

$$Q={{\displaystyle {\sum }_{i=1}^n\left(y_i-\overline{y}\right)}}^2$$

(3)

$$F=U\times {\displaystyle \frac{n-2}{Q}}$$

(4)

In the equations, U represents the sum of squared errors; Q denotes the sum of squares due to regression; n equals 21; $\widehat{y}$ is the predicted value for the i-th year; $\overline{y}$ is the mean value of the dependent variable (GPP) over the 21 years; and $y_i$ is the actual value of GPP for the i-th year.

To more clearly illustrate the variation trends in each meteorological factor, we computed p-values using the F-test with different pixel values, thereby enabling a further classification of significance levels (as defined in

Table 2. Classification of significance levels in linear regression.

Category	Slope (R)	p-Value Range
Extremely Significant Decrease	R < 0	p < 0.01
Significant Negative	R < 0	0.01 ≤ p < 0.05
No Significant Decrease	R < 0	p ≥ 0.05
No Significant Increase	R > 0	p ≥ 0.05
Significant Increase	R > 0	0.01 ≤ p < 0.05
Extremely Significant increase	R > 0	p < 0.01

) [25]:

2.3.2. Partial Correlation Analysis

The partial correlation coefficient measures the degree of linear relationship between two variables while controlling for the effects of other variables. To minimize the influence of other climatic factors on grassland GPP in the arid regions of Xinjiang, we employed partial correlation analysis using MATLAB 2024b (MathWorks, Natick, MA, USA). This method has been widely applied in previous studies to explore the interactions between vegetation and individual meteorological factors [26]. When using three additional meteorological factors as control variables, the partial correlation coefficient between GPP and a single meteorological factor can be calculated using the following formula:

$$R_{\left.xy\right|z_1,z_2,z_3}={\displaystyle \frac{R_{xy}-R_{{xz}_1}\times R_{{yz}_1}-R_{{xz}_2}\times R_{{yz}_2}-R_{{xz}_3}\times R_{{yz}_3}}{\sqrt{(1-{R_{{xz}_1}}^2)(1-{R_{{xz}_2}}^2)(1-{R_{{xz}_3}}^2)(1-{R_{{yz}_1}}^2)(1-{R_{{yz}_2}}^2)(1-{R_{{yz}_3}}^2)}}}$$

(5)

GPP in the arid regions of Xinjiang is denoted as x, and the meteorological factor of interest is represented by y. Three controlling climatic factors are defined as, z₁, z₂, and z₃. Correlation coefficients were computed in MATLAB 2024b as follows: $R_{xy}$ between x and y; $R_{x{z}_1}$,$R_{x{z}_2}$, and $R_{x{z}_3}$ between x and each controlling factor z₁, z₂, z₃; and $R_{y{z}_1}$, $R_{y{z}_2}$, and $R_{y{z}_3}$ between y and each z₁, z₂, z₃. To quantify the direct association between GPP (x) and the target factor (y) independent of the confounding effects of the controlling factors (z₁, z₂, z₃), the partial correlation coefficient $R_{\left.xy\right|z_1,z_2,z_3}$ was calculated. Statistical significance of the impacts of individual meteorological factors on GPP across spatial pixels was assessed by deriving p-values from t-tests (also performed in MATLAB 2024b), enabling subsequent categorization of significance levels (as defined in

Table 3. Classification of partial correlation significance.

Category	Correlation (R)	p-Value Range
Extremely Significant Negative Correlation	R < 0	p < 0.01
Significant Negative Correlation	R < 0	0.01 ≤ p < 0.05
Not Significantly Negatively Correlated	R < 0	p ≥ 0.05
Not Significantly Positively Correlated	R > 0	p ≥ 0.05
Significant Positive Correlation	R > 0	0.01 ≤ p < 0.05
Extremely Significant Positive Correlation	R > 0	p < 0.01

) [27]:

2.3.3. Ridge Regression Analysis

Ridge regression, incorporating L2 penalty to manage predictor collinearity, was applied to isolate the effects of target climatic drivers on grassland GPP in arid Xinjiang by minimizing interference from correlated covariates. We employed ridge regression analysis in MATLAB 2024b to model the relationship between GPP and the four meteorological factors (VPD, PRE, TEM, SSR) simultaneously, thereby mitigating multicollinearity issues [28].

This method has been widely applied in previous studies to explore the relationships between GPP and key meteorological factors within a multivariate framework, providing more robust coefficient estimates. Ridge regression effectively alleviates multicollinearity issues by introducing a regularization parameter λ when estimating the coefficients β. All ridge regression computations were performed using MATLAB 2024b. The formula for ridge regression is as follows [29]:

$${\hat{\beta }}_{ridge}={(X^{T}X+\lambda I)}^{-1}{X}^{T}y$$

(6)

In the ridge regression formula, X is the matrix of independent variables, y is the dependent variable matrix, λ is the regularization parameter, and I is the identity matrix.

Prior to analysis, all variables were standardized in MATLAB 2024b so that variables with different units could be analyzed together:

$$X_m={\displaystyle \frac{x-\min (x)}{\max (x)-\min (x)}}$$

(7)

where X_m is the normalized data of the climatic factors including VPD, PRE, TEM, and SSR.

Normalization of each independent variable was essential to ensure that variables with different units could be appropriately compared and analyzed. The formula for normalization is as follows:

$$Y_m={\displaystyle \sum _{i=1}^n{a}_i}{X}_{im}+b$$

(8)

In the equation, $Y_m$ represents the normalized GPP, $X_{im}$ denotes the normalized meteorological factors, and a_i are the regression coefficients.

To better illustrate the impact of each climatic factor on GPP across different regions, this study simultaneously calculated the relative contribution rate using MATLAB 2024b, defined as follows:

$${\eta }_{c1}=a_1{X}_{trend}$$

(9)

$${\eta }_{rc1}={\displaystyle \frac{\left|{\eta }_{c1}\right|}{\left|{\eta }_{c1}\right|+\left|{\eta }_{c2}\right|+\left|{\eta }_{c3}\right|+\left|{\eta }_{c4}\right|}}$$

(10)

In these equations, ${\eta }_{c1}$ represents the absolute contribution of a specific meteorological factor to GPP, while a₁ is the regression coefficient of that climate factor obtained through ridge regression analysis. $X_{trend}$ refers to the standardized trend of the climate factor, derived through normalization. ${\eta }_{rc1}$ denotes the relative contribution rate of the meteorological factor to GPP, indicating the proportionate contribution of the meteorological factor to changes in GPP. This comprehensive contribution analysis was implemented in MATLAB 2024b. This metric was used for comparative evaluation of the relative importance of different factors.

2.3.4. Structural Equation Modeling

Structural Equation Modeling (SEM), as a statistical approach that integrates the strengths of confirmatory factor analysis and path analysis, demonstrates unique value in investigating multivariate interactions within complex systems [30]. In this study, SEM was employed to construct a model of the multifactorial interactions and coupling relationships between GPP and meteorological variables. By formulating path hypotheses, the analysis focused on elucidating the direct effect of VPD on GPP, while also uncovering indirect pathways among various meteorological factors. All variables were standardized (mean = 0, SD = 1) before analysis. Raster data were resampled to match GPP resolution (500 m), and pixels with missing values were excluded. Model fitting was performed annually (2000–2020) using the lavaan package in R. Standardized path coefficients (β) were calculated to quantify driver contributions, with significance determined at p < 0.05 (*). The semPlot package was used for visualization. Analyses were conducted in R 4.4.1. During model validation, modification indices were utilized to optimize parameter estimation, and standardized path coefficients were applied to quantify the relative contributions of different driving factors.

3. Results

3.1. The Temporal and Spatial Variation Trend of GPP

Linear regression analysis of GPP trends in Xinjiang during 2000–2020 revealed distinct spatiotemporal variations in GPP across the region (Figure 2a). Spatially, a predominant portion of the area (comprising 58.75%) experienced a significant GPP increase. Areas exhibiting an extremely significant increase (making up 41.21%) were primarily clustered in the mid-section of the northern Tianshan Mountains slopes and the western foothills of the Kunlun Mountains. Conversely, regions showing a decline in GPP (constituting 6.38%) were sporadically distributed within the Ili River Valley and parts of the Junggar Basin.

On an interannual scale, the mean GPP for Xinjiang demonstrated a statistically highly significant increasing trend over the 2000–2020 period (p < 0.001, Figure 2b). Specifically, the period from 2000 to 2010 showed relatively moderate fluctuations, whereas after 2010, the rate of GPP increase accelerated and exhibited greater interannual variability, with GPP peaking in 2016 (325.70 gC/m²/year).

3.2. Trends in Meteorological Factors in Grasslands

Systematic analysis of the spatiotemporal evolution of the four meteorological factors revealed significant spatial heterogeneity and topographic dependence in their changes. VPD showed a widespread increasing trend across the region (significant increase: 15.75%, extremely significant increase: 4.68%), with the central Tianshan Mountains and the southern Junggar Basin forming the core zones of change, followed by the eastern edge of the Altai Mountains and the northern slope of the Kunlun Mountains (Figure 3a,e). TEM changes displayed spatial heterogeneity (no significant change: 95.08%), with slight cooling observed in some areas, while localized warming signals appeared in topographic transition zones (Figure 3b,f).

SSR and PRE changes demonstrated a typical latitudinal differentiation pattern. Influenced by the barrier effect of the Tianshan Mountains, northern Xinjiang exhibited an increasing trend in SSR (no significant Increase: 47.43%, significant increase: 7.93%), accompanied by a decrease in PRE (no Significant decrease: 43.04%, significant decrease: 1.29%); conversely, southern Xinjiang showed the opposite pattern (Figure 3c,d,g,h). Notably, the Ili River Valley experienced the highest increase in SSR across the entire region, which may be related to the significantly greater increase in PRE in the Kunlun Mountains (significant increase: 3.02%, extremely significant increase: 0.64%) compared to surrounding areas. Compared with the widespread significant trend observed in VPD, the significant change zones for SSR and PRE were concentrated primarily in mountain–basin transition areas.

3.3. Impact of Meteorological Factors on GPP

3.3.1. Heterogeneous Adaptation Mechanisms of GPP to Multi-Dimensional Meteorological Stressors

Partial correlation analysis revealed complex response relationships between meteorological factors and grassland productivity (Figure 4). The VPD-GPP correlation exhibited distinct topographic differentiation: 1.42% of areas showed significant positive correlations in the central Altai Mountains and parts of Tianshan Mountains, while 2.78% demonstrated significant negative correlations in the northwestern Junggar Basin and oasis margins of Tarim Basin (Figure 4a,e).

TEM and PRE predominantly exerted positive influences on GPP, with no significant negative correlation areas detected (negative correlations ≤ 1.09%). Notably, 78.58% (TEM) and 87.61% (PRE) of regions did not show significantly positive correlations (Figure 4b,d,f,h). For SSR, positive effects slightly dominated (85.59% were not significantly positive correlated), though 10.39% exhibited non-significant negative correlations. Significant negative correlation zones (0.13%) were concentrated in eastern Tianshan Mountains and southern Tarim Basin (Figure 4c,g).

3.3.2. Major Controlling Factors of GPP

To facilitate the interpretation of ridge regression coefficients ($a_i$), we categorized them into four levels: high-intensity negative effects (−1.6 ≤ $a_i$ < −0.5), indicating a statistically significant inhibitory effect of the independent variable on GPP; low-intensity negative effects (−0.5 ≤ $a_i$ < 0); low-intensity positive effects (0 < $a_i$ ≤ 0.5), reflecting limited positive predictive power of the independent variable on GPP; and high-intensity positive effects (0.5 < $a_i$ ≤ 1.6), identifying the independent variable as a strong driving factor with substantial explanatory power for GPP changes.

Overall, the impacts of TEM and PRE on grassland productivity were overwhelmingly positive, with only minimal areas exhibiting negative effects (TEM: low-intensity negative 0.01%, high-intensity negative 0.69%; PRE: low-intensity negative 0.16%, high-intensity negative 13.18%). In contrast, the positive and negative effects of VPD and downward SSR were more balanced (VPD: low-intensity negative 10.99%, high-intensity negative 7.47%, low-intensity positive 35.89%, high-intensity positive 45.65%; SSR: low-intensity negative 3.64%, high-intensity negative 3.38%, low-intensity positive 42.58%, high-intensity positive 50.40%), suggesting that changes in these meteorological factors play a significant role in the decline of grassland productivity. Notably, the effects of VPD showed a distinct regional distribution, with positive effects primarily concentrated in the northern Altai Mountains and central Tianshan Mountains, while negative impacts were noticeably concentrated in the northwestern Junggar Basin and some northern areas. The spatial distribution characteristics of VPD effects were highly consistent with the regions of positive and negative correlation observed in the partial correlation analysis. Additionally, there was overlap between regions where SSR had negative impacts and areas where VPD exerted positive effects.

Analysis of TEM and PRE impacts demonstrated that both factors primarily exhibited a strong positive correlation with GPP, evidenced by the high prevalence of high-intensity positive effects (TEM: 82.00%, PRE: 71.19%). Unlike VPD, TEM and PRE did not display widespread significant negative impacts (Figure 5b,d,f,h). This finding, quantified by the effect size distribution, was consistent with the results from the partial correlation analysis, further validating that the dominant effects of TEM and PRE on GPP were strongly positive.

3.3.3. VPD-Driven Suppression of GPP

To comprehensively evaluate the mechanisms through which the four meteorological factors influence GPP, we employed SEM for multi-factor path analysis. SEM, as a multivariate statistical method that integrates factor analysis and causal pathways, effectively elucidates the multi-level interactions within complex systems.

The results of the model (Figure 6) indicate that TEM, PRE, SSR, and VPD significantly regulated GPP (p < 0.05). Among these factors, VPD demonstrated a significant negative correlation with GPP, with the direction and magnitude of its standardized path coefficient indicating a systematic suppressive effect in arid regions. The results indicate that although partial correlation analysis demonstrated regional positive effects of VPD on GPP, the integrated impact of VPD on GPP across the arid ecosystems of Xinjiang revealed a significant negative response (path coefficient = −0.58, p < 0.05). This discrepancy in responses between local and regional scales suggests that the ecological effects of VPD on GPP exhibited notable spatial heterogeneity and environmental dependency. It is particularly noteworthy that under extreme arid conditions, a unique threshold response relationship between VPD and GPP may be established.

3.4. Independent Effects of Atmospheric Drought on GPP

3.4.1. Contributions of Multi-Factors Dominated by VPD and Spatial Heterogeneity Regulation of GPP

Integrating the results of partial correlation, ridge regression analysis, and SEM, VPD emerged as the dominant meteorological factor exerting a significant inhibitory effect on GPP (path coefficient= −0.58, p < 0.05). To further elucidate the control mechanisms of atmospheric drought on grassland productivity in Xinjiang, we conducted a coupled analysis of VPD threshold effects and spatial heterogeneity, revealing the underlying regulatory principles.

Using a ridge regression model, the spatial heterogeneity of meteorological drivers affecting grassland productivity was analyzed (Figure 7). The spatial pattern indicated that the relative contribution of VPD peaked in the central Tianshan Mountains and the eastern Altai Mountains, highlighting the sensitivity of mountainous ecosystems to atmospheric drought stress (Figure 7a). High contribution zones for TEM were distributed in a belt along the western edge of the Junggar Basin, while the maximum contribution area for SSR was concentrated in the Ili River Valley.

In terms of hierarchical factor contributions, VPD dominated with a relative contribution significantly higher than that of other meteorological variables, confirming atmospheric drought as the key limiting factor for grassland growth in arid regions (Figure 7b). SSR and PRE followed, reflecting their fundamental regulatory roles in controlling light use efficiency and water availability, respectively. TEM’s contribution is comparatively weaker.

3.4.2. Regulation of GPP by VPD Threshold Effects and Cumulative Drought Stress

Building on the spatial heterogeneity analysis from ridge regression, we further revealed the systemic stress mechanisms of atmospheric drought on grassland productivity through dynamic fitting between VPD and ridge regression coefficients ($a_i$). Linear regression between the ridge regression coefficients ($a_i$ for VPD) and VPD values revealed a significant negative correlation (y = −0.179x + 0.109, R² = 0.9575, Figure 8b), confirming that increasing atmospheric drought imposes systemic stress on grassland growth. Threshold analysis based on the linear regression model further indicated a critical VPD value of 0.61 kPa, defined as the x-intercept (i.e., where y = 0) of the fitted line beyond which ecosystem responses fundamentally shifted: within the low VPD range (≤0.61 kPa), GPP was promoted, whereas exceeding this threshold resulted in GPP suppression. The linear regression approach was empirically justified by its strong explanatory power (R² = 0.9575) between ridge coefficients and VPD, and its direct quantification of the systemic stress response aligned with our objective to identify the tipping point (y = 0) where VPD transitions from promoting to suppressing GPP.

The spatial distribution of the multi-year (2000–2020) average VPD revealed pronounced spatial heterogeneity across the study area (Figure 8a). The hatched regions in Figure 8a, representing major VPD contribution zones (from Figure 7a), are characterized by persistent atmospheric drought conditions. Notably, VPD values in these key areas exhibited a continuous upward trend, with increases in some locations already reaching the 0.61 kPa critical threshold. The sustained rise in VPD within high-contribution zones may progressively constrain the ecological resilience of grasslands. Analysis of the interannual variations in the main VPD contribution zones (Figure 8c) showed that the regional average VPD rose significantly from 0.66 kPa in 2000 to 0.74 kPa in 2008, representing an increase of 12.1%. Although the rate of increase in VPD slowed in the past decade (2010–2020), the ongoing upward trend remained significant, with short-term peaks frequently occurring within the fluctuation range (e.g., reaching 0.71 kPa in 2017). This phenomenon suggests that even with a deceleration in growth rate, the long-term cumulative effects of VPD may weaken the physiological resilience of grasslands through progressive stress. The regional average VPD (0.74 kPa) already exceeded the critical threshold of 0.61 kPa. Furthermore, the expanded fluctuation amplitude of VPD over the past decade (±0.05 kPa) led to an increased frequency of extreme drought events, necessitating caution regarding its potential disruption of ecosystem functions.

4. Discussion

4.1. Overall Increasing Trend in GPP and Regional Response Differences

Our results indicate a significant overall increasing trend in GPP across Xinjiang (Figure 2a), with most regions exhibiting sustained spatial growth (Figure 2b). This pattern aligns with the global vegetation productivity response to climate warming and CO₂ fertilization effects [8]. Notably, Zhu et al. (2023) [31] highlighted significant regional differentiation in ecosystem response pathways: in northern areas, changes in aboveground gross primary productivity were primarily driven by leaf area index, whereas southern regions relied more on functional trait regulation. This mechanism may explain the significant increase in VPD observed in the central Tianshan arid zone (p < 0.05, Figure 3a) and its inhibitory effect on GPP found in our study. In soil moisture-limited areas like central Tianshan, vegetation photosynthetic capacity is more sensitive to atmospheric drought, with functional trait responses potentially playing a dominant role.

The inhibitory effect of VPD on GPP involves a dual mechanism. Firstly, elevated VPD directly restricts CO₂ uptake by increasing stomatal resistance, thereby reducing leaf-level light use efficiency [32]. Kowalski et al. (2024), studying European grasslands, further confirmed a significant positive correlation between increased VPD and vegetation vitality loss, with VPD explaining more variance than PRE alone [33]. This finding is consistent with the observed slowing of GPP growth in high-VPD areas in our study (Figure 8b,c). Secondly, elevated VPD may exacerbate the spatiotemporal coupling of vegetation water stress, especially in arid regions where soil moisture and atmospheric demand imbalance can trigger “hot drought” compound events, leading to nonlinear declines in carbon sink function [10].

Regional heterogeneity analysis revealed that the slowing of GPP growth in relatively humid areas like the Ili River Valley (Figure 2a) contrasts with the findings of Zhu et al. (2023) [31], who suggested that vegetation in humid regions typically regulates productivity through functional traits. The significant VPD inhibition on GPP in these relatively humid zones observed in our study (Figure 8c) may stem from unique hydrothermal configurations: rapid increases in growing-season VPD coupled with swift soil moisture depletion. This pattern echoes the “atmospheric drought-dominated mechanism” proposed by Kowalski et al. (2024) for grassland vitality loss [33].

4.2. Threshold Response of Grassland Productivity to Atmospheric Drought

Our study revealed a distinctive response mechanism of grassland ecosystems in Xinjiang’s arid region to atmospheric drought, identifying a critical VPD threshold of 0.61 kPa. This provides a key theoretical basis for regional ecohydrological research. Unlike He et al. (2023) [34], who proposed VPD percentile thresholds (e.g., the 40.07th percentile triggering mild GPP loss) for Chinese grasslands, our study—based on the extreme arid climate and desert vegetation water regulation strategies in Xinjiang—used ridge regression to determine a specific kPa threshold (0.61 kPa). This discrepancy likely arises from the unique TEM–moisture coupling characteristics of Xinjiang—where annual PRE is less than 150 mm and vegetation rapidly responds via stomatal regulation (Figure 5a)—resulting in a lower water stress tipping point than in more humid regions. This reflects the high sensitivity of arid ecosystems to atmospheric drought.

It is noteworthy that although current VPD levels below the threshold may optimize water use efficiency (WUE) by improving stomatal conductance, according to the stomatal hydraulic feedback hypothesis proposed by Buckley et al. (2005), plants rapidly close their stomata to reduce xylem tension and thus prevent embolism formation [35]. However, complex interactions among multiple factors may mask deeper underlying risks. Our SEM revealed a standardized path coefficient of −0.58 (p < 0.05) for the effect of VPD on GPP. This finding contrasts with Li et al. (2024), who identified soil moisture as the dominant driver in 44.6% of the Northern Hemisphere study area [5]. Similarly, research conducted in Southwest China in 2023 highlighted that those grasslands, due to their shallow root systems, are less capable of accessing deep soil moisture and are therefore more sensitive to atmospheric drought compared to deep-rooted forests [36]. Moreover, the global model projections by Yuan et al. (2019) suggest that the continued rise in VPD may further offset the CO₂ fertilization effect [8]. The observed exceedance of the VPD threshold in parts of our study region (Figure 8b,c) may serve as an early regional warning signal of this emerging trend.

4.3. Research Limitations and Future Directions

Despite the systematic analysis of VPD effects on grassland productivity in Xinjiang’s arid region presented in this study, several key scientific issues require further exploration. Firstly, productivity estimates based on MODIS GPP data (at a 500 m resolution NASA EOSDIS Land Processes DAAC, USGS Earth Resources Observation and Science Center, Sioux Falls, SD, USA) may not adequately capture the small-scale patchy distribution characteristics of certain grasslands (less than 100 m). Additionally, due to spatial limitations and the relatively short temporal scale of the study period, the conclusions drawn may not be universally applicable to all specific sub-regions or capture longer-term dynamics. Furthermore, this study has not explored the potential of using machine learning approaches, such as explainable analysis based on Shapley values, for identifying key factors and their thresholds.

This study focused on atmospheric drought (characterized via VPD), which helps reveal its independent effects. Nevertheless, in arid regions, the dominant role of soil moisture stress in regulating productivity is well-established. Consequently, concurrent soil moisture data were not incorporated. This omission precludes disentangling the relative and interactive contributions of atmospheric drying (VPD) and soil drought to GPP suppression [37]. Therefore, future research must prioritize investigating the compound effects of these co-occurring stressors, as their interaction likely defines critical thresholds for ecosystem function.

5. Conclusions

This study systematically elucidates the response mechanisms of grassland ecosystems to atmospheric drought stress in the arid regions of Xinjiang. Firstly, based on linear regression analysis, GPP exhibited a significant upward trend across the region from 2000 to 2020, with 58.75% of the study area showing significant increases. However, under high VPD conditions, particularly in key atmospheric drought-impacted zones identified by ridge regression contribution analysis, the rate of GPP increase notably slowed down, demonstrating the strong inhibitory effect of atmospheric drought on ecosystems in arid areas. Secondly, through integrated analysis using partial correlation, ridge regression, and SEM, VPD was unambiguously identified as the dominant meteorological factor influencing grassland GPP, exhibiting a significant negative path coefficient (path coefficient = −0.58). Finally, ridge regression analysis identified a precise VPD threshold of 0.61 kPa for Xinjiang’s arid grasslands. Below this threshold, VPD generally promotes GPP; exceeding it suppresses productivity. This threshold is lower than those reported for more humid regions, revealing the heightened sensitivity of plant water regulation under extreme drought conditions. Crucially, the regional average VPD in the main contribution zones already exceeded this critical threshold, having increased significantly from 0.66 kPa in 2000 to 0.74 kPa in 2008 (a 12.1% increase). In summary, this study advances the quantitative understanding of the relationship between atmospheric drought and grassland productivity in arid regions, identifying key thresholds and drivers, and provides a scientific basis for regional ecological conservation and climate adaptation strategies.