Differences in the Sensitivity of Gross Primary Productivity and Ecosystem Respiration to Precipitation

Abstract

The spatiotemporal variability of precipitation profoundly influences terrestrial carbon fluxes, driving shifts between carbon source and sink dynamics through gross primary productivity (GPP) and ecosystem respiration (ER). As a result, the sensitivities of GPP and ER to precipitation (S_GPP and S_ER), along with their differential responses, are pivotal for understanding ecosystem reactions to precipitation changes and predicting future ecosystem functions. However, comprehensive evaluations of the spatiotemporal variability and differences in S_GPP and S_ER remain notably scarce. In this study, we utilized eddy covariance flux data to investigate the spatial patterns, temporal dynamics, and differences in S_GPP and S_ER. Spatially, S_GPP and S_ER were generally strongly correlated. Among different ecosystems, the correlation between S_GPP and S_ER was lowest in mixed forest and highest in broadleaf and needleleaf forest. Within the same ecosystem, S_GPP and S_ER exhibited considerable variation but showed no significant differences. In contrast, they differed significantly across ecosystems, with pronounced variability in their magnitudes. For example, shrubland exhibited the highest values for S_GPP, whereas needleleaf forest showed the highest values for S_ER. Temporally, S_ER demonstrated more pronounced changes than S_GPP. Different ecosystems displayed distinct trends: shrubland exhibited an upward trend for both metrics, while grassland showed a downward trend in both S_GPP and S_ER. Forest, on the other hand, maintained stable S_GPP but displayed a downward trend in S_ER. Additionally, S_GPP and S_ER exhibited a notable non-linear response to changes in the aridity index (AI), with both showing a rapid decline followed by stabilization. However, S_ER demonstrated a wider adaptive range to precipitation changes. Generally, this research enhances our understanding of the spatiotemporal variations in ecosystem carbon fluxes under changing precipitation patterns.

1. Introduction

The carbon dioxide fixed through photosynthesis at the ecosystem level, known as gross primary productivity (GPP), along with the carbon dioxide released via ecosystem respiration (ER), directly governs the global carbon budge balance [1,2,3,4]. GPP and ER demonstrate an intertwined and complementary relationship in the ecosystem, showcasing unique dynamic characteristics despite their close interconnection [5,6,7,8]. Numerous studies have demonstrated that GPP is predominantly driven by climatic factors, resulting in significant interannual variability [9,10,11,12]. Regarding ER, studies have primarily focused on the effects of temperature [13,14,15], which may be attributed to the inclusion of soil respiration [16,17]. Conversely, ER exhibited a positive relationship with vegetation photosynthetic capacity [13,18,19], suggesting the influence of available carbon substrates on both plant and soil respiration [20]. The sensitivity (S) of ecosystem carbon flux, specifically GPP and ER, to climate change can be understood as the magnitude of its response to these changes [21]. For vegetation productivity, S is generally considered to be an intrinsic characteristic [22,23], with spatial variability primarily driven by differences in ecosystems and local environmental conditions [23,24]. Existing studies have demonstrated that S can also vary in response to various environmental factors, such as drought, heatwaves, and changes in precipitation patterns [11,25,26]. In the context of ER, research has predominantly focused on the temperature sensitivity of soil respiration (Q₁₀), exploring its relationships with climate drivers [27,28] and soil microbial communities [29,30]. Soil respiration exhibits a non-linear relationship with soil temperature, with CO₂ efflux increasing up to a peak (28–30 °C) and decreasing at higher temperatures (>30 °C). Therefore, Q₁₀ can accurately represent the temperature dependence of soil respiration for temperatures below 26–28 °C [31].

Precipitation serves as a pivotal environmental driver influencing both GPP and ER. However, their sensitivities to variations in precipitation (S_GPP and S_ER) can differ markedly [32,33], highlighting the complex interactions within the ecosystems. Numerous studies have investigated the sensitivity of vegetation productivity to precipitation [23,24,25], yielding a consensus conclusion: S_GPP is greater in arid regions and smaller in mesic regions [11,25]. However, there remains a notable gap in the research regarding the spatiotemporal differences between S_GPP and S_ER, which may be attributed to the intricate nature of ER, involving various underlying processes and factors. For instance, under optimal soil moisture conditions, the respiration of soil microbes and roots is stimulated, whereas excessive water leading to anaerobic conditions can suppress root respiration [31,34,35].

This study directly explores the similarities and differences in the sensitivity of S_GPP and S_ER, along with their main environmental drivers, using the FLUXNET2015 dataset. The dataset follows a standardized methodology for data processing, which includes rigorous quality control, filtering, the interpolation of meteorological data, and accurate partitioning of flux data [36]. Specifically, the objectives of this study were to (i) examine the spatial distribution and differences of S_GPP and S_ER; and (ii) elucidate the temporal trends and variations in S_GPP and S_ER.

2. Materials and Methods

2.1. Data Collection

To make use of as much observed data as possible, we utilized the FLUXNET2015 dataset to obtain daily mean eddy-covariance carbon dioxide fluxes and meteorological data (https://fluxnet.org/, accessed on 1 May 2021) [36]. Initially, we excluded site-years where missing values in the relevant columns exceeded 30% of the total data, obtaining the dataset to 188 sites (Figure 1) and their corresponding Whittaker biomes (Figure 2). We first reclassified global terrestrial ecosystems into five major categories based on the MODIS IGBP land cover product (0.05°, https://lpdaac.usgs.gov/products/mcd12q1v006/, accessed on 11 March 2022): forest, grassland, shrubland, wetland, and cropland [37]. Given that forests encompass various types and are distributed across multiple climate zones globally, we further subdivided the forest category into needleleaf forest, broadleaf forest, and mixed forest. To investigate the relationship between precipitation sensitivity and drought severity, we used the global average aridity index (AI) at a 0.05° resolution from 1970 to 2000. The data were sourced from the Global Aridity Index and Potential Evapotranspiration (ET0) Database v3 [38], where AI = MAP_rec/MA_ET₀. Here, MAP_rec refers to mean annual precipitation (mm year⁻¹), and MA_ET₀ represents mean annual reference evapotranspiration (mm year⁻¹), which is calculated using the fully parameterized FAO-56 Penman–Monteith formula.

2.2. Definition and Calculation of S_GPP and S_ER

Researchers typically represent the sensitivity of carbon dioxide flux to precipitation using the slope of an ordinary least squares (OLS) linear regression [11,23]. Nonetheless, considering that the slope of OLS linear regression is fixed and symmetric, while the actual response of vegetation productivity to changes in environmental factors is often asymmetric [25,39,40], we calculated the sensitivity based on previous studies as the ratio of carbon dioxide flux (GPP and ER) to the net change in precipitation, i.e.,

$$S_{\mathrm{GPP}}=\left[\left({\displaystyle \frac{\mathrm{G}\mathrm{P}\mathrm{P}-{\mathrm{G}\mathrm{P}\mathrm{P}}_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}}{\mathrm{G}\mathrm{P}\mathrm{P}}}\right)/(\mathrm{P}-{\mathrm{P}}_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}})\right]\times 100$$

(1)

$$S_{\mathrm{ER}}=\left[\left({\displaystyle \frac{\mathrm{E}\mathrm{R}-{\mathrm{E}\mathrm{R}}_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}}{\mathrm{E}\mathrm{R}}}\right)/\left(\mathrm{P}-{\mathrm{P}}_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}\right)\right]\times 100$$

(2)

where S_GPP and S_ER represent the sensitivities of GPP and ER to precipitation, respectively. GPP denotes the gross primary productivity at the site-year, while GPP_mean refers to the multi-year average annual total gross primary productivity at the site. P indicates the total annual precipitation at the site-year, and P_mean represents the multi-year average annual precipitation at the site. Similarly, ER stands for the annual total ecosystem respiration at the site-year, and ER_mean denotes the multi-year average annual total ecosystem respiration at the site. Positive values of S_GPP and S_ER indicate a direct relationship, where precipitation and GPP or ER increase or decrease together. Negative values reflect an inverse relationship, where increases in precipitation lead to decreases in GPP or ER, and vice versa.

2.3. Data Analysis

We first conducted a paired t-test to assess overall differences between S_GPP and S_ER, as well as across ecosystems. To explore differences within the same vegetation type, we used one-way ANOVA followed by the least significant difference (LSD) test [41]. ANOVA was performed initially to determine if there were significant differences among group means (p = 0.05) The null hypothesis for ANOVA posits that all group means are equal:

H0:μ1 = μ2 = μ3 = ⋯ = μk

(3)

If the ANOVA results were significant, indicating that at least one group mean differs, we proceeded with the LSD test. The LSD value was calculated using the following formula:

$$LSD={t_{\alpha /2,df}}_e\sqrt{2{\displaystyle \frac{MSE}{n}}}$$

(4)

where ${t_{\alpha /2,df}}_e$ is the critical value from the t-distribution for a given significance level α and degrees of freedom ${ df}_e$ from the error term of ANOVA; MSE is the mean square error from the ANOVA; n is the number of observations per group. Given the data limitations, in this study, the significance level for the LSD test was set at p = 0.05. We then utilized linear regression to assess the relationship between S_GPP and S_ER, both overall and within different vegetation functional types.

Additionally, to investigate the range and differences in the degree of variation between S_GPP and S_ER, we calculated and compared their coefficients of variation (CV).

$$\mathrm{CV}={\displaystyle \frac{SD}{MEAN}}$$

(5)

Linear regression was used to explore the relationships between S_GPP, S_ER, and their variations with AI and mean annual precipitation (MAP), as well as to examine the influence of environmental factors (mean annual temperature, MAT; MAP; and vapor pressure deficit, VPD) on S_GPP and S_ER across each ecosystem. For the temporal variation analysis, we first selected sites within each ecosystem that had data spanning more than 10 years. Then, for each site, we focused on the specific time period (2000–2014) and used the Mann–Kendall test and linear regression to detect trends in S_GPP and S_ER [42,43].

3. Results

3.1. The Overall Differences and Variability of S_GPP and S_ER Across Ecosystems

The overall differences between S_GPP and S_ER were not statistically significant (Figure 3A); however, S_GPP exhibited markedly greater variation compared to S_ER (Figure 3C). Likewise, no significant differences were observed in S_GPP and S_ER across ecosystems (Figure 3B), yet their variations differed markedly (Figure 3D). For example, the variation in S_ER was greater than that of S_GPP in wetland and broadleaf forest, whereas S_GPP exhibited considerably greater variation than S_ER in grassland and needleleaf forest (Figure 3D). These disparities in S_GPP and S_ER across ecosystems underscore their distinct responses to changes in precipitation.

In addition, our results revealed significant differences in S_GPP and S_ER across all ecosystems. Specifically, shrublands exhibited the highest S_GPP, and wetlands the lowest, while croplands and broadleaf forests showed moderate levels (Figure 4A). For S_ER, needleleaf forest exhibited the highest values, broadleaf forest and cropland showed intermediate levels, and wetland had the lowest (Figure 4B). Analysis of the variation in S_GPP and S_ER across all ecosystems revealed that broadleaf forest exhibited the greatest variation for both metrics, as mixed forest showed the least variation for S_GPP and needleleaf forest showed the least variation for S_ER (Figure 4C,D).

3.2. Relationship Between S_GPP and S_ER

S_GPP and S_ER exhibited a highly significant positive correlation, p < 0.001, R² = 0.27 (Figure 5A). The slope of the linear regression (slope = 0.31) indicates that S_ER was more stable compared to S_GPP. This suggests that for every one-unit change in S_GPP, S_ER changes by 0.31 units.

Furthermore, we examined the relationship between S_GPP and S_ER across ecosystems. With the exception of mixed forests, all other ecosystems exhibited a significant positive regression relationship between S_GPP and S_ER. For example, in cropland, forest, and grassland, S_GPP and S_ER showed significant positive correlations, with p and R² of 0.011, 0.32; <0.001, 0.36; and 0.006, 0.18, respectively (Figure 5B). Broadleaf forest displayed the strongest linear relationship between S_GPP and S_ER, with p < 0.001, R² = 0.61, while no linear relationship was observed in mixed forest. Furthermore, the regression slopes between S_GPP and S_ER were less than 1 for all ecosystems, indicating that S_ER is more stable than S_GPP.

3.3. Changes in S_GPP and S_ER with AI and MAP

Considering the distinct ecological implications of positive and negative values for S_GPP and S_ER, we delineated these values to examine their variations in response to AI and MAP. When S_GPP and S_ER were positive, they initially experienced a dramatic decline with increasing AI, followed by stabilization, with critical breakpoints at 0.65 and 0.85 (Figure 6A,B). This suggests that as water availability increases, water no longer acts as a limiting factor for GPP and ER, and their sensitivity to water variations diminishes. Conversely, S_GPP and S_ER revealed distinct behaviors in response to variations in MAP: S_GPP underwent a sharp decrease followed by a continued decline with rising MAP, whereas S_ER gradually diminished as MAP increased (Figure 7A,B). When S_GPP and S_ER were negative, they did not exhibit significant changes in response to variations in AI and MAP (Figure 6C,D and Figure 7C,D).

3.4. Temporal Trends of S_GPP and S_ER

The MK trend analysis revealed a pronounced difference in the temporal trends in S_GPP and S_ER across ecosystems. Specifically, the S_GPP of shrubland showed a marked increasing trend over time, while grassland exhibited a nearly significant decreasing trend (p = 0.066) (Figure 8). No significant changes were observed in the other ecosystems (Figure 8). Regarding S_ER, shrubland demonstrated a significant increasing trend (Figure 9), whereas grassland showed a declining trend over time (Figure 9). No significant temporal changes were observed in the other ecosystems. It is important to note that the extent of temporal variation in S_GPP and S_ER in forests was the smallest, which may be related to the high complexity of forest ecosystems (Figure 8 and Figure 9).

Moreover, even within the same ecosystem type, the direction and extent of temporal changes in S_GPP and S_ER exhibited certain differences. For example, although both S_GPP and S_ER in wetland showed non-significant temporal trends, the opposite directions of their changes over time still partially reflect the distinct responses of S_GPP and S_ER to precipitation variability (Figure 8 and Figure 9).

3.5. Dominant Factors of S_GPP and S_ER in Different Ecosystems

Overall, compared to other ecosystem types, S_GPP and S_ER in grassland were more readily influenced by temperature and precipitation changes. Specifically, the S_ER and S_GPP in grassland showed a significant declining trend with increasing MAP (Figure 10A,B), while S_ER significantly increased with rising MAT and VPD (Figure 10C,E). Aside from grassland, only the S_GPP in wetland significantly decreased with higher MAT (Figure 10D).

4. Discussion

4.1. Differences and Variations Between S_GPP and S_ER

Our study found no significant differences between S_GPP and S_ER (Figure 3A,B), indicating a comparable level of responsiveness to changes in precipitation. A precipitation manipulation experiment conducted on the Tibetan Plateau demonstrated that under extreme drought conditions, GPP and ER decreased by 17.26% and 19.05%, respectively, relative to the control, thereby providing indirect support for our findings [44].

The similarity in the responses of GPP and ER to precipitation changes may arise from their intrinsic connection. Physiologically, photosynthesis and respiration in plants are interrelated processes [45,46]. The energy required for photosynthesis, derived from the respiratory breakdown of organic matter, leads to the accumulation of stable compounds when photosynthesis exceeds respiration, thereby providing a consistent substrate for respiration [46,47,48]. On the other hand, stomatal conductance regulates the exchange of CO₂, O₂, and H₂O. Under drought stress, stomata close to reduce transpiration [49,50,51], impacting gas exchange and consequently affecting both photosynthesis and respiration. Ecologically, studies have confirmed a close relationship between GPP and ER at regional and global scales, primarily because productivity serves as the main and direct substrate supplier for respiration, fundamentally constraining ER [52,53].

However, the overall variation in S_GPP was greater than that of S_ER (Figure 3C), indicating that GPP was more sensitive to changes in precipitation compared to ER, a conclusion that was further supported by research on the impact of water availability on carbon dioxide flux in grassland ecosystems [44,54,55]. This may be attributed to several factors. First, the composition of ecosystem ER is complex, consisting of both autotrophic respiration (R_a) and heterotrophic respiration (R_h), which may exhibit different sensitivities to changes in precipitation. This is because R_a is more dependent on photosynthetic products [23] and photosynthesis is easily affected by precipitation [56,57]. Furthermore, existing research indicates that the sensitivity of aboveground plant respiration to precipitation exceeds that of belowground respiration, which is in turn more sensitive than microbial respiration [44]. This suggests that the greater stability of ER in response to water variations, compared to gross primary GPP, was primarily due to the relative stability of soil or heterotrophic respiration. Secondly, multiple researchers have indicated that GPP exhibits a greater response amplitude to precipitation fluctuations. For instance, a controlled experiment conducted in temperate grassland demonstrated that GPP increased more than ER in response to increased precipitation [54]. Additionally, research based on eddy covariance measurements has shown that GPP was more sensitive to precipitation, contributing more to NEE responses than ER [53,58].

GPP is more susceptible to changes in precipitation, which may be attributed to the role of H₂O as a substrate for light reactions. Successful photolysis of water is essential for the proper functioning of electron transport and carbon dioxide fixation during photosynthesis. Existing research indicates that photosynthesis is highly sensitive to drought; when facing drought stress, vegetation may respond by reducing stomatal conductance and the rate of electron transport [59], as well as inhibiting the activity of PS-II and Rubisco [60]. In contrast, ER was influenced not only by aboveground plant activity but also by root systems, soil microorganisms, and enzymatic activities, which are generally less affected by variations in precipitation [44,61].

The variation in S_GPP and S_ER in cropland and shrubland was the most similar (Figure 3D), likely due to the high sensitivity of R_h to drought in these arid and semi-arid regions [62], where unstable carbon and soil microbes are concentrated in surface layers, causing R_h to react strongly to slight moisture changes [63]. Across ecosystems, shrubland, grassland, and needleleaf forest exhibited higher S_GPP and S_ER (Figure 4A,B), likely because these ecosystems are globally water-limited [64] and the impact of global warming on ER is more pronounced in colder regions [65]. In arid ecosystems, higher water-use efficiency [23] likely increases the sensitivity of productivity to precipitation changes compared to humid regions, and the close link between GPP and ER contributes to the elevated S_ER.

4.2. The Response of S_GPP and S_ER to Changes in Water Availability

In our study, a positive sensitivity of carbon dioxide flux to precipitation signifies a synchronous relationship, whereas a negative value indicates an inverse correlation. When S_GPP and S_ER are positive, S_GPP shows a marked non-linear response to changes in AI and MAP (Figure 6A and Figure 7A). Specifically, S_GPP and S_ER are highest in arid and semi-arid regions. S_GPP and S_ER decline rapidly with increasing AI when AI < 0.65 and 0.85, but they stabilize when AI > 0.65 and 0.85. It is widely recognized that S_GPP is typically higher in arid regions and lower in more mesic ecosystems [11,25,66]. This may be attributed to the fact that water availability is a key limiting factor for vegetation growth in these areas [67,68], coupled with the higher water-use efficiency of local vegetation [69]. In contrast, in humid regions, factors such as light and soil nutrients may play a more significant role in limiting vegetation growth [70].

Regarding non-linearity, Knapp and Smith (2001) were the first to find that vegetation productivity responds to precipitation in a non-linear manner [71]. With further research, an increasing body of evidence supports the idea that grassland ecosystems exhibit positive asymmetry in their response to precipitation, meaning that productivity increases in wet years exceed declines in dry years [39,72], which is beneficial for the ecosystem’s carbon sink [73]. Conversely, negative asymmetry is observed in dry years [74], which is detrimental to the carbon sink. On the other hand, experiments in grassland have also shown that ER and its components exhibit asymmetric responses to changes in precipitation [39,44].

The asymmetry between carbon dioxide flux and precipitation may be attributed to two main factors: firstly, the asymmetric response of ANPP to precipitation was a key reason for the non-linear nature of carbon dioxide flux in response to precipitation changes [40,44]; secondly, the asymmetric relationship between soil moisture and precipitation exacerbates the non-linear relationship of carbon dioxide flux. Experimental studies have shown that reduced precipitation significantly decreases surface soil water content (SWC), while increased precipitation has limited effects, further intensifying the non-linear response of carbon dioxide flux across different precipitation gradients [44]. Moreover, overall, S_ER exhibits a significant decreasing trend over a wider range of AI and MAP changes (Figure 6B and Figure 7B), indicating that ER is more sensitive to variations in moisture while possessing some regulatory capacity. This phenomenon may be attributed to the soil’s ability to store and buffer water [75] and to the effective utilization of deep moisture by plant roots [76,77], which together enhance the ecosystem’s adaptability to changes in moisture.

4.3. Temporal Trend Pattern of S_GPP and S_ER Across Ecosystems

For S_GPP, only shrubland showed a significant increase, while wetland exhibited a non-significant decrease (Figure 8), and grassland showed a non-significant downward trend. In contrast, for S_ER, grassland exhibited a highly significant downward trend, while shrubland showed a significant increase over time (Figure 9). The temporal variations in S_GPP and S_ER between grassland and shrubland were notably different, which is a noteworthy observation. As global warming advances, total precipitation has increased [78]. Coupled with the CO₂ fertilization effect, this has contributed to the rise in global vegetation productivity [79,80,81]. Concurrently, the expansion of leaf area in vegetation has led to higher water consumption [82], establishing a balance between the increased availability of water and the elevated demand for it.

Studies have shown that grassland exhibits a broadly positive, symmetric response to precipitation changes [39], indicating a reduced water limitation on carbon dioxide flux. In contrast, the temporal trend of S_GPP in forest remains more stable, likely due to multiple factors. First, rising atmospheric CO₂ concentrations have enhanced forest water-use efficiency [83]. Additionally, the complex vertical structure of forest provides resilience to environmental changes [84]. These factors collectively contribute to the stability of forest S_GPP trends. For S_ER, the increased precipitation enhances soil respiration due to the strong water retention capacity of the soil [28,85]. Studies show that moderate increases in precipitation significantly boost microbial respiration, while root and microbial respiration only decline significantly under extreme precipitation increases [86]. Therefore, in ecosystems with relatively mild water resource limitations, an increase in precipitation results in a more pronounced declining trend in S_ER. It is noteworthy that while both S_GPP and S_ER in shrubland exhibit a consistent upward trend over time, the increase in S_GPP is markedly more pronounced than that of S_ER. This discrepancy may be attributed to the higher water use efficiency of vegetation in water-limited areas [87], as GPP typically responds more rapidly to changes in precipitation compared to ER [58].

4.4. Uncertainty Analysis

Eddy covariance offers a direct technique for assessing trace gas and energy fluxes between the surface and the atmosphere at an ecosystem level, usually within a 1 km radius of the measurement site [36]. This makes it difficult to directly compare it with other methods. Campioli et al. assessed eddy covariance (EC) data by comparing them with other methods, such as inventory and chamber techniques, and found that EC biases are not evident across sites, which indicates the effectiveness of the standard post-processing procedures and enhances confidence in the reliability of EC data [88]. Pastorello and the eddy covariance site teams further compared and technically validated their measurements based on their in-depth knowledge of the respective sites [36]. While measurement and processing uncertainties are inevitable and can be significant for certain sites and ecosystem conditions, the flux values in this dataset generally align with expectations. Overall, eddy covariance continues to be one of the most dependable methods for evaluating land–atmosphere exchanges at the ecosystem scale. The widespread use of eddy covariance data in many scientific studies indirectly confirms its reliability and value.

On the other hand, CO₂ flux (net ecosystem exchange, NEE) is primarily partitioned into GPP and ER using two methods: the nighttime flux method and the daytime flux method [36]. The nighttime flux method estimates ER by constructing a respiration–temperature model using nighttime data and calculates GPP based on NEE. The parameterization process accounts for dynamic factors such as water availability, substrate, and phenology [89]. The daytime flux method, which combines both daytime and nighttime data, estimates GPP through a light-response curve and vapor pressure deficit, while ER is estimated using a respiration–temperature relationship [90]. Compared to the nighttime flux method, the daytime flux method is more widely used in research, which is consistent with this study.

5. Conclusions

This study, using the FLUXNET2015 dataset, examined the spatiotemporal variations in S_GPP and S_ER and identified the primary environmental factors driving these variations. Our results show that S_GPP and S_ER were generally strongly correlated spatially. The correlation was weakest in mixed forests and strongest in broadleaf and needleleaf forests. While S_GPP and S_ER displayed notable variation within the same ecosystem, no significant differences were observed. In contrast, significant differences were found across ecosystems, with marked variability in their magnitudes. For instance, shrubland had the highest S_GPP values, while needleleaf forest had the highest S_ER values. Temporally, S_ER showed more pronounced changes than S_GPP. Ecosystem temporal trends varied: shrubland displayed an increasing trend for both metrics, grassland showed a decreasing trend in both S_GPP and S_ER, and forest maintained stable S_GPP but showed a downward trend in S_ER. Moreover, S_GPP exhibited a clear non-linear response to changes in the aridity index (AI), with both metrics rapidly declining before stabilizing. However, S_ER had a broader adaptive range to precipitation changes. Overall, this study improves our understanding of the spatiotemporal variations in ecosystem carbon dioxide fluxes under changing precipitation patterns.