Transpiration (Tc) is the main source of water flux from terrestrial evapotranspiration, accounting for 80%-90% of terrestrial evapotranspiration (Jasechko et al., 2013; Good et al., 2015; Zhang et al., 2021). Transpiration recirculates 62000±8000 km³ of water annually into the atmosphere (Jasechko et al., 2013). Soil water is the primary source of water for transpiration. Deficiencies in soil moisture (SM) can affect the stomatal conductance of vegetation, limiting productivity and threatening vegetation survival (Phillips et al., 2009; Peng et al., 2011; McDowell and Allen, 2015). Another major driver of Tc is vapor pressure deficit (VPD). On the one hand, VPD is an important driver of the atmospheric water demand for plants (Rawson et al., 1977; Yuan et al., 2019). On the other hand, when VPD increases to a certain threshold, a further increase in VPD leads to a decrease in vegetation stomatal conductance until it closes which in turn reduces Tc (Oren et al., 1999; Konings and Gentine, 2017). Therefore, VPD is a key regulator that determines the role of Tc (Novick et al., 2016; Yuan et al., 2019; Grossiord et al., 2020). In the context of current climate warming, VPD is gradually increasing (Yuan et al., 2019), whereas SM is decreasing worldwide which may reduce Tc (Deng et al., 2020). Therefore, quantifying the relative contributions of SM and VPD on Tc is beneficial to understand mechanisms of terrestrial water cycle.

Owing to the complex mechanisms of interactions between SM and VPD and the high spatial and temporal heterogeneity, the relative role of Tc remains unclear (Liu et al., 2020; Yu et al., 2022). Current studies on the effects of SM and VPD on Tc have focused on the site scale (Ray et al., 2002; Song et al., 2020). For example, Song et al. (2020) investigated the independent effects of SM and VPD on Tc in subtropical coniferous plantations in Jiangxi province, and their results showed that the limitation of VPD on Tc was greater than that of SM. However, the independent effects of SM and VPD on Tc have not been well-explored at regional and global scales which restrict our understanding of the ecosystem water cycling response to climate change (Seneviratne et al., 2010; Novick et al., 2016). In recent years, numerous studies on the effects of SM and VPD on photosynthesis have been conducted at regional and global scales (Liu et al., 2020; Yu et al., 2022). Considering the coupling of photosynthesis and transpiration (Beer et al., 2007; Jasechko et al., 2013; Zhang et al., 2019), this study would be a valuable reference for Tc studies. Liu et al. (2020) demonstrated that SM, rather than VPD, dominates ecosystem production in most vegetated land areas. Lu et al. (2022) showed that VPD has a great influence on ecosystem production efficiency compared to SM across larger regions. Dang et al. (2022) concluded that temperature is more important than SM in ecosystem productivity in global vegetation cover and that the importance of SM and temperature varies among vegetation types.

As SM and VPD are coupled (Seneviratne et al., 2010; Liu et al., 2020; Dang et al., 2022), their effects on transpiration and photosynthesis remain controversial. Therefore, disentangling the strong coupling between SM and VPD and exploring their independent effects on Tc is essential for understanding the climate response mechanisms of water cycling in terrestrial ecosystems. To explore the effects of SM and VPD on ecosystems, Liu et al. (2020) proposed a binning approach to successfully decouple the correlation between SM and VPD. This method divides the data into bins according to the SM and VPD percentiles, and the SM and VPD are largely decoupled in each SM or VPD interval which performs better in multi-factor independent contribution analysis. Yu et al. (2022) successfully disentangled the relative effects of SM and VPD on ecosystems in Central Asia using a binning approach. This method provides an effective way to disentangle the relative contributions of VPD and SM to Tc.

In recent years, the southern subtropical region of China has experienced significant climate change, especially a decline in SM and an increase in VPD which has had a significant impact on the regional Tc and thus poses a potential threat to the regional ecological environment and social economy (Deng et al., 2020; Denissen et al., 2022; Wang et al., 2022). However, the independent contributions of SM and VPD to Tc remain controversial, which seriously limits our understanding of regional water cycling climate change response patterns and the formulation of response strategies. In this study, we selected the subtropical region of China as the study area and explored the respective effects of VPD and SM changes on Tc with the help of correlation analysis and binning approach. The sensitivity differences in Tc response to VPD and SM changes in different vegetation types were further investigated. We hypothesized that if SM plays a dominant role in Tc, then Tc would not be very sensitive to changes in VPD, and high SM would promote Tc. Conversely, if VPD plays a dominant role in Tc, then Tc is not very sensitive to changes in SM and VPD promotes Tc.

Spearman’s correlation analysis can describe both linear and nonlinear relationships and requires no assumption of normality for the data. This method is widely used owing to its low sensitivity to outliers (Jin et al., 2020). In this study, Spearman’s correlation analysis was applied to Tc, VPD, SM, and SIF data in the study area. Because of the coarse spatial resolution of GLEAM Tc data, downscaling is required for accurate analysis of the driving forces of variabilities of transpiration in Chinese subtropical vegetation and to ensure the consistency of the spatial resolution of the data. Furthermore, the linear regression equations of SIF and GLEAM Tc data were established pixel-by-pixel using GOSIF data. The basic process was as follows: First, the GOSIF data were upscaled to 0.25°, and a statistical relationship was established with the GLEAM Tc data. Second, the regression equation was applied to the original high-resolution SIF data to predict Tc data for downscaling. To prevent the interference of other factors, data on temperatures less than 15℃ and photosynthetic photon flux density less than 500 µmol m^-2 s^-1 in the study area were removed (Liu et al., 2020). Considering the strong coupling between the SM and VPD, it is impossible to quantify the effects of the SM and VPD on Tc using conventional methods. Therefore, the “binning approach” proposed by Liu et al. (2020) was chosen to decouple the correlation between SM and VPD and then quantify the effects of SM and VPD on Tc. First, the SM and VPD data in the study area were sorted from small to large according to the value, and then, the sorted SM and VPD data were divided into 10 intervals (i.e., 0-10%, 10%-20%,..., 90%-100%) by percentile. The SM and VPD intervals were intersected to derive 10×10 subintervals which were called “bin” in this paper (Figure 2). SM and VPD were decoupled in each interval; therefore, it was convenient to quantify the effects of SM and VPD on Tc. A bin was deleted when the number of data points was less than 10, and the mean value of the bin was used to quantify the effects of SM and VPD on Tc. After disentangling the SM-VPD correlation, Tc (VPD|SM) was used to quantify the effect of VPD on Tc. ΔTc (VPD|SM) denotes the amount of change in Tc (mm month^-1) as VPD goes from the lowest to the highest value while keeping SM constant or changing less. Similarly, Tc (SM|VPD) was used to quantify the effect of the SM on Tc. The following formulae were used for the calculations (Liu et al., 2020): Formulaes (4) and (5)

where ΔTc(VPD|SM) is the difference between the Tc at the highest VPD bin and the lowest VPD bin in each SM interval; I is the total number of SM intervals; i is the specific SM interval number (i=1, 2,..., 10); n_i,max and n_i,min are the numbers of the largest and smallest VPD bins at SM interval i, respectively. Equally, where ΔTc(SM|VPD) is the difference between the Tc at the highest SM bin and the lowest SM bin in each VPD interval; J is the total number of VPD intervals; j is the specific VPD interval number (j=1, 2,...., 10); m_j,max and m_j,min are the numbers of the largest and smallest SM bins at VPD interval j, respectively.

Understanding the driving role of VPD and SM on transpiration changes in the context of climate warming remains to be improved mainly because SM and VPD are coupled through land-atmosphere interactions, restricting the quantification of the independent effects of SM and VPD on Tc. The “binning method” decouples the strong coupling between SM and VPD at regional and global scales (Liu et al., 2020; Cheng et al., 2022; Yu et al., 2022). In contrast, traditional methods such as correlation analysis and multiple linear regression do not sufficiently decouple the covariance between SM and VPD (Liu et al., 2020). The response of Tc to the SM and VPD may be nonlinear (Sage and Kubien, 2007; Green et al., 2019), and this method can capture this nonlinearity (Dang et al., 2022; Fu et al., 2022a). The SM-VPD correlation in this study showed a significant decrease after using this method (Figure 4), which provided a basis for exploring the independent effects of SM and VPD on Tc. In this study, we analyzed the independent effects of SM and VPD on Tc in subtropical China and found that the independent effects of SM on Tc were greater than those of VPD. This supports the results of Liu et al. (2020) and Yu et al. (2022) and emphasizes the importance of SM on Tc. Tc is more susceptible to SM in subtropical China, where droughts are frequent and SM highly fluctuates annually (Yuan et al., 2016). However, recent studies on the independent effects of SM and VPD on Tc in subtropical China have shown that VPD and SM limit Tc by 90.8% and 9.2% (Song et al., 2020), respectively; the reason for this difference might be the different study timescales. Because solar radiation, VPD, and temperature affect transpiration (Granier et al., 2000; Bai et al., 2015), we excluded pixels with temperatures less than 15℃, VPD less than 5 hPa, and photosynthetic photon flux densities less than 500 µmol m^-2 s^-1 from the study area to control the interference of other factors in the study (Liu et al., 2020). Therefore, this study focused on the relative contributions of changes in VPD and SM to Tc during the growing season (mainly summer). However, Song considered the ranges of VPD and SM seasonal variations in Tc throughout the year. Considering the similarity in the seasonal patterns of VPD and Tc in the subtropical regions of China, the correlation between VPD and Tc is higher on the intra-annual scale which shows the dominant role of VPD on Tc to some extent. Recent studies have demonstrated the existence of critical thresholds for plant water stress in terrestrial ecosystems (Fu et al., 2022b, 2022c). The results of this study showed that the response of Tc to SM in the study area was below the 35% threshold, where its increase contributed significantly to Tc. However, above the 35% threshold, changes in SM had a weaker effect on Tc. This indicates that there is a threshold effect on the response of vegetation to SM. Notably, although the threshold response of the VPD was not obvious, a high VPD promoted Tc when the SM was greater than the 10% threshold. Conversely, when the SM was less than the 10% threshold, a high VPD restricted the increase in Tc, indicating that the response of Tc to VPD was regulated by the SM (Liu and Biondi, 2020; Zhang et al., 2021). The high VPD in other forests was not always limited to Tc (Figure 7d), which might be due to the deep, vigorous root systems of forests that could still utilize deep soil water under atmospheric drought (McDowell et al., 2008; Zhang et al., 2020). In contrast, croplands have a greater tendency to close their stomata to reduce water consumption under water stress, because of their shallow root systems (Chen et al., 2021) (Figure 7h). High VPD and SM values contributed to Tc in forests under the same SM or VPD conditions (Figure 8a). However, croplands showed the opposite trend (Figure 8g), which may be related to environmental conditions, soil properties, and root systems across forests, short woody vegetation, and croplands (Chen et al., 2021). In future studies, it should distinguish the sensitivity of different vegetation types to SM and VPD in ecosystem water cycle simulations. In addition, numerous recent studies have shown that SIF is closely related to Tc and can be used to represent changes in Tc (Pagán et al., 2019; Feng et al., 2021; Shan et al., 2021). Figures 3d and 8 show that using SIF is feasible, verifying the feasibility of our analysis and the reasonableness of the results, respectively.

Although this study quantified the relative effects of VPD and SM on Tc in the subtropical regions of China, there are still some limitations. First, the current study shows that SM and VPD are the most important factors affecting Tc (Liu et al., 2020; Song et al., 2020; Yu et al., 2022); however, temperature, radiation, and wind speed can also affect Tc. Therefore, in this study, we attempted to set the control conditions. For example, we excluded study areas of temperatures less than 15℃, VPD less than 5 hPa, and photosynthetic photon flux density less than 500 µmol m^-2 s^-1 to minimize the interference of other factors with the results. However, factors such as wind speed and vegetation growth structure were not considered because of a lack of data, and a quantitative analysis of these factors is needed in future studies. Second, despite differences in influences such as seasonal and interannual dynamics, this study focused on the average state of vegetation to represent plant response characteristics in conventional years. In the future, we will focus on the effects of interannual variability and extreme events on the driving forces of Tc. Third, the study area is mainly located in a subtropical monsoon climate zone, and the climate type is relatively single, which causes the study results to be under-representative. In the future, the sensitivity of Tc to SM and VPD under different climatic, dry, and wet conditions should be explored. This study provides insight into the relative roles of SM and VPD in Tc in humid regions. The effect of the SM on Tc in this region may increase in the future in the context of climate change.

This study explored the effects of VPD and SM, as independent driving forces, on Tc changes in subtropical China based on multi-source remote sensing data and meteorological reanalysis data using the “binning method.” The sensitivity of the Tc response to SM and VPD was further analyzed by combining different vegetation types. The results show that the effect of SM on Tc in the study area was greater than that of VPD. Overall, the relative contribution of SM to Tc was approximately five times that of VPD. The sensitivities of SM and VPD to different vegetation types were different, and the sensitivity of the four vegetation types to SM was greater than that to VPD. Different vegetation types had different thresholds for Tc response to SM with the lowest thresholds for other forests. In addition, based on the strong coupling of photosynthesis and transpiration, this study repeated the above analysis using SIF remote sensing data and derived results consistent with Tc which proved the reliability of the results of this study. These results suggest that the role of SM should be considered in future investigations on the cognition of regional water cycle climate change response patterns and the development of response strategies. Finally, attention should be paid to response differences among different land cover types.