We collected gross primary productivity (GPP) and net primary productivity (NEP) data measured via the eddy covariance method from literature published over the past 20 years (2002-2019) in regard to EVAs in China. Based on Web of Science database (http://apps.webofknowledge.com) and CNKI database (http://www.cnki.net), data were retrieved by using “eddy covariance”, “carbon flux” “carbon exchange”, “carbon budget”, “productivity”, “gross primary productivity”, “net ecosystem productivity” and “net ecosystem exchange (NEE)” as keywords. The data were filtered and corrected by researchers at each site, using coordinate rotation, WPL correction, storage flux calculation, outlier filtering, nighttime flux correction, NEE gap filling and partitioning. Additionally, the data were continuously measured for at least an entire year. At the same time, the geographic location, ecosystem and vegetation type at each observation site were extracted. The CUEe value was estimated as CUEe=NEP/GPP.

Through the above standard screening approach, carbon fluxes observation data of 55 ecosystems covering forestlands, grasslands, croplands and wetlands were obtained ( Figure 1 ). The data covered the temperate zone, warm temperate zone, Tibetan Plateau and other climatic regions. The latitude range of the selected flux stations was 30.47°N-49.35°N, and the longitude range was 83.57°E-122.65°E. There were 37 grassland sites, 3 forestland sites, 9 wetland sites and 6 cropland sites ( Table 1 ).

Climatic variables including the mean annual temperature (MAT) and mean annual precipitation (MAP) were also collected. The data were derived from the same studies as the carbon fluxes data. Mean values of the air temperature and precipitation in the observation year were calculated as the MAT and MAP, respectively.

The leaf area index (LAI) was derived from the satellite-borne Moderate Resolution Imaging Spectroradiometer (MODIS) data product (MOD13Q1) with a spatial resolution of 1 km and a temporal resolution of 8 days from 2000 to 2018. Soil data including the soil pH and soil organic carbon content (SOC) were retrieved from the global normalized soil dataset of the Harmonized World Soil Database (version 1.2) (https://daac.ornl.gov/cgi-bin/dsviewer.pl?ds_id=1247).

First, we compared differences in the CUEe among the different EVAs and vegetation types in China. The relationship between the GPP and NEP, and the relationships between the CUEe and longitude and latitude were analyzed via linear regression.

Linear and quadratic regression analyses were performed to examine the correlation between the MAT, MAP, soil pH and SOC and the CUEe with a significance level of α = 0.05. The hierarchical partitioning method was employed to determine the contributions of the longitude, latitude, MAT, MAP, soil pH, SOC, GPP and NEP to the CUEe via the “rdacca.hp” package in R (Lai et al., 2022).

We further established a structural equation model (SEM) to evaluate the direct and indirect factors regulating the CUEe, and assessed their contributions to the total effects of standardization (direct effects plus indirect effects). The causal relationship between the predicted variables was based on a priori knowledge of the effects of climatic variables (MAT and MAP), geographic location (longitude and latitude), soil parameters (soil pH and SOC), LAI, GPP and NEP on the CUEe. Since the variables of climate, geographic location and soil groups were closely related, principal component analysis (PCA) was conducted to create a multivariate index representing each group (Wang et al., 2017). The first principal component (PC1) explained 61-80% of the total variance of each group and was subsequently used for SEM analysis, in which the data were fitted to the model using the maximum likelihood estimation method. The model’s adequacy was determined using the χ² test method, goodness of fit (GFI) index, and root mean squared error of approximation (RMSEA) index. Favorable model fits were indicated by no significant difference when using the χ²-test method (P > 0.05), a high GFI value (>0.9), and a low RMSEA value (<0.08) (Liu et al., 2017). SEM analysis was conducted in Amos 21.0 (Amos Development Corporation, Chicago, IL).

All analyses were conducted in R software (version 3.5.1, R Development Core Team, Vienna, Austria). ArcGis 10.1 and R were used for plotting.

The results showed that the GPP ranged from 91.25 g C m^-2 yr^-1 in the Bajitan grassland to 1657.9 g C m^-2 yr^-1 in the Linze cropland ( Figure 2 ). The NEP and GPP of the different ecosystems were linearly correlated ( Figure 2 ). The CUEe varied greatly among the different ecosystems, such as -0.39 in the Xilinhot Stipa grassland and 0.67 in the Ansai grassland. Based on the site average, the estimated average value of the CUEe of EVAs in China was 0.20 ( Figure 2 ).

We found that there were significant differences in the CUEe among the different EVAs in China (p<0.05). The average values of the CUEe in arid and semiarid areas, Loess Plateau, agro-pastoral ecotones, and Tibetan Plateau were 0.34, 0.46, 0.07, and 0.14, respectively. Among them, the CUEe on the Loess Plateau was the highest, and that in the agro-pastoral ecotones was the lowest ( Figure 3 ). There was no significant difference in the CUEe among the different vegetation types (p>0.05), in which the CUEe values in grassland, cropland, forestland and wetland areas were 0.17, 0.37, 0.41 and 0.16, respectively ( Figure 3 ). The CUEe significantly decreased with increasing longitude, while it showed no trend with increasing latitude ( Figure 4 ).

We mainly analyzed the impact of climate factors (MAT and MAP) and soil factors (soil pH and SOC) on the CUEe. The CUEe first decreased and then increased with increasing MAT, soil pH and SOC, and decreased with increasing MAP ( Figure 5 ). Among the four environmental factors, MAT exerted the largest impact on the CUEe, which could explain nearly 41% of the variation in the CUEe.

Hierarchical partitioning analysis showed that the NEP, LAI and GPP were the most important factors influencing the CUEe, followed by the longitude ( Figure 6 ). It is not difficult to determine that compared to the soil factors (SOC and soil pH), the climate factors (MAT and MAP) exerted a greater impact on the CUEe.

SEM analysis showed that the GPP, NEP and LAI directly affected the CUEe, while climate and soil factors exerted indirect effects. Jointly considering the direct and indirect effects, biotic factors (GPP, NEP and LAI) were the most important predictors determining the regional variation in the CUEe ( Figure 7 ). Whether through hierarchical partitioning analysis or SEM, the results showed that these variables could explain approximately 70% of the total variation in the CUEe. Regarding the CUEe, the NEP exerted a greater impact on the CUEe than the GPP ( Figures 6 , 7 ).

Many studies have found that the CUE of plants is a constant (Waring et al., 1998; Delucia et al., 2007). We provided a reference for the basic status of the CUEe in EVAs in China, and suggested that the CUEe value ranged from -0.39 to 0.67 ( Figure 2 ). The variation range of the CUEe in this study was larger than that in other studies (Chen et al., 2018); the variation range of the CUEe values in EVAs remained reasonable and was smaller than the variation range from −1 to 0.6 for global ecosystems (Chen et al., 2015b). The increased variability of the CUEe may be due to the significant deviation in heterotrophic respiration (Rh) and its ratio to the net primary productivity (NPP) (Chen et al., 2018). The lowest CUEe value was found in the Xilinhot Stipa grassland, where the large amount of autotrophic respiration (Ra) and Rh release exceeded the low GPP. The highest CUEe value was found in the Ansai grassland, indicating that the carbon consumption of ecosystem respiration was low on the Loess Plateau. According to the site average value, the average CUEe value of EVAs in China was estimated at 0.2, which indicated that an average productivity of 20% was fixed in ecosystem biomass and soil organic matter (Hursh et al., 2017). This efficiency was higher than the average CUEe value in other Asian countries and global ecosystems (Kato and Tang, 2008; Chen et al., 2015b).

The CUEe varied with the different ecosystem vegetation composition and structure. Gilmanov et al. (2010) found that the CUEe in European grasslands was lower than that in croplands and wetlands. Similarly, the average CUEe in global grasslands was lower than that in other ecosystem types (Chen et al., 2015b). In contrast, it was reported that the CUEe in grassland was higher than that in deciduous broad-leaved forestland and coniferous forestland, which contributes to the plant tissue in grassland yielding a lower investment in ecosystem respiration (Re) than that in forestland (Law et al., 2002). Our results demonstrated that the CUEe value in grassland was lower than that in forestland and cropland ( Figure 3 ). This likely occurs because under the control of environmental conditions, grasslands are mainly distributed in semiarid and alpine areas, where the plant biomass is low and the active growth period is short. Compared to Re, the GPP was more significantly restricted by a low temperature and drought, which led to a higher Re/GPP ratio and thus a lower CUEe (1-Re/GPP) (Reichstein et al., 2007; Kato and Tang, 2008; Prescher et al., 2010; Chen et al., 2015b). Previous studies showed that there was a comparative Re in grassland to that in forestland and cropland, while the GPP was far lower than that in forestland and cropland (Chen et al., 2014; Chen et al., 2015a).

There were significant differences in the CUEe among the different EVAs. The CUEe on the Tibetan Plateau and agro-pastoral ecotones were significantly lower than those in the other EVAs ( Figure 3 ), which may be related to the impacts of rainfall and temperature on productivity (Kato and Tang, 2008; Yu et al., 2013). In our study, the CUEe decreased with increasing rainfall, and it was also lower within the low-temperature range ( Figure 5 ). Generally, rainfall on the Tibetan Plateau and agro-pastoral ecotones was higher than that in the other EVAs, but the temperature was lower (Liu et al., 2021), resulting in a low CUEe.

We found that the latitudinal distribution of the CUEe in EVAs in China was not significant ( Figure 4 ), which was consistent with previous results (An et al., 2017). However, we found that the CUEe showed a decreasing trend with increasing longitude ( Figure 4 ). In terms of longitude, the vegetation CUE usually decreased from west to east, which was closely related to ecosystem elevation (Chen and Yu, 2019). In EVAs in China, from west to east, the terrain transforms from plateau to plain terrains. Our results also showed that the CUEe increased with the elevation. This is consistent with previous studies on the vegetation CUE on a global scale (Zhang et al., 2009).

Climate factors such as the MAT and MAP are the two most important factors affecting the GPP, Re and NEP (Kato and Tang, 2008; Yu et al., 2013). Generally, with increaseing MAT and MAP, the GPP and Re increase, respectively, while the increase in the GPP is greater than that in Re (Yi et al., 2010; Tang et al., 2016). Therefore, the CUEe is expected to increase with increasing MAT and MAP. However, our results showed that the CUEe first decreased and then increased with increased MAT, and was negatively correlated with MAP (p<0.05) ( Figure 5 ). The temperature is the most important factor affecting the carbon fluxes on the Tibetan Plateau (Kato et al., 2006; Saito et al., 2009), which is positively correlated with its GPP and Re. With increasing temperature, the rate of increase of Re was higher than that of the GPP; thus, the CUEe decreased instead. In plateau areas limited by water, a high temperature could inhibit the GPP and Re, and the rate of decrease of Re was higher than that of the GPP, so the CUEe increased with increasing temperature (Wang et al., 2014). Chen et al. (2019) found that the MAT could explain nearly 47% of the variation in the CUEe, and our results were similar ( Figure 5 ). In contrast, the MAP explained the smaller differences in the CUEe ( Figure 6 ). On a global scale, when the MAP is below 2300 mm, the CUEe shows a downward trend with increasing MAP (Zhang et al., 2009). However, in EVAs in China, rainfall is the main limiting factor (Hu et al., 2010; Dong et al., 2011). With increased rainfall, the GPP and Re increased, while the rate of increase of the Re was higher than that of the GPP, so the CUEe showed a decreasing trend with increasing rainfall.

Compared to the climate factors, the soil factors (soil pH and SOC) imposed a less notable impact on the CUEe ( Figure 6 ). The variation range of soil conditions in EVAs in China may be limited (Zhang et al., 2019). Different ecosystems have different soil pH values, such as grasslands with high soil pH values and forestlands with low soil pH values (Chen and Yu, 2019). In our study, the CUEe was the lowest under almost neutral conditions ( Figure 5 ). This suggests that an alkaline or acidic environment is not enough to yield the hightest CUEe value in EVAs. The variation range of the CUE of plants growing in poor-soil, low-temperature, drought-prone and other high-stress environments is generally larger than that of plants growing in suitable environments. Overall, among ecosystems, especially EVAs, the higher the SOC content is, the lower the CUEe ( Figure 5 ).

The cycle of carbon between the Earth’s surface and the atmosphere is controlled by biotic and abiotic processes that regulate the storage of carbon in the biogeochemical cycle and release carbon into the atmosphere. The GPP and NEP are mainly determined by climate, soil and biotic factors. Therefore, it is not difficult for us to understand the regulatory mechanism of the CUEe. Through hierarchical partitioning analysis, we found that biotic factors such as the GPP, NEP and LAI exerted a greater impact on the CUEe in EVAs of China. However, the direct effects of climate (MAT and MAP) and soil factors (soil pH and SOC) were very limted in our study ( Figure 6 ). We further explored the relationship among them through SEM ( Figure 7 ). We propose that the geographical pattern shapes the climate and soil factors that influence vegetation factors such as ecosystem LAI and further determines the GPP and NEP, thus affecting the CUEe. Climate factors and soil factors mainly play an indirect role in determining the CUEe, while biotic factors play a more direct role in determining the CUEe ( Figure 7 ). Based on our research, we found that the variation in the CUEe was mainly affected by climate, soil and biotic factors.