How Does Plant CO2 Physiological Forcing Amplify Amazon Warming in CMIP6 Earth System Models?

The physiological response to increasing CO2 concentrations will lead to land surface warming through a redistribution of the energy balance. As the Amazon is one of the most plant‐rich regions, the increase in surface temperature, caused by plant CO2 physiological forcing, is particularly large compared to other regions. In this study, we analyze the outputs of the 11 models in the Coupled Model Intercomparison Project Phase 6 to find out how CO2 physiological forcing amplifies Amazonian warming under elevated CO2 levels. With the CO2 concentration increase from 285 to 823 ppm, the Amazon temperature increased by 0.48 ± 0.42 K as a result of plant physiological forcing. Moreover, we assess the contributions of each climate feedback to the surface warming due to physiological forcing by quantifying climate feedbacks based on radiative kernels. Lapse rate feedback and cloud feedback, analyzed as the primary contributors, accounted for 53% and 37% of Amazon warming, respectively. The warming contributions of these two feedbacks also exhibit a significant spread, which contributes to the predictive uncertainty. The surface warming due to reduced evapotranspiration is larger than the upper tropospheric warming in the Amazon, resulting in surface warming by lapse rate feedback. In addition, cloud cover in the Amazon region decreases due to the reduced evapotranspiration. Decreased cloud cover amplifies surface warming through the shortwave cloud feedback. Furthermore, differences in circulation and local convection caused by physiological effect contribute to the inter‐model spread of the cloud feedback.


Introduction
Since the start of the Industrial Revolution, the atmospheric CO 2 concentration has steadily increased (Portner et al., 2022), leading to changes in the global climate.It is well known that the greenhouse effect causes global warming (CO 2 radiative forcing).In addition, the physiological response of plants to the rising CO 2 also affects global temperature (CO 2 physiological forcing) (Park et al., 2021;Zarakas et al., 2020).Plants exhibit two major physiological responses to rising atmospheric CO 2 : the CO 2 fertilization effect and the stomatal closure effect.The former refers to an increase in photosynthetic rates due to an increase in CO 2 and the latter refers to the partial closing of plant stomata under elevated CO 2 level (Ainsworth & Long, 2005;Drake et al., 1997;Gunderson & Wullschleger, 1994).These two effects have opposing influences on surface temperature.Fertilization effect enhances photosynthesis and transpiration of plants, which causes surface cooling.In contrast, decreasing stomatal conductance suppresses transpiration per unit of leaf area (Field et al., 1998) and reduces the moisture from the land to the atmosphere (Sellers et al., 1997).Consequently, surface warming occurs due to increasing the ratio of sensible heat flux to latent heat flux (Betts et al., 2004).
Previous studies using model simulations have shown that warming due to the stomatal closure effect will exceed cooling due to the fertilization effect in the future climate and this could have a significant impact on global warming (Park et al., 2020;Skinner et al., 2017;Zarakas et al., 2020).Cao et al. (2010) showed that CO 2 physiological forcing increases terrestrial temperature by 0.42 K in response to a doubling of CO 2 .Zarakas et al. (2020) found that for a doubling of CO 2 , changes due to physiological forcing contribute about 0.12 K to global warming and 6.1% of the temperature change due to total CO 2 forcing (radiative and physiological forcing).In addition, the plant CO 2 physiological forcing significantly affects the tropical atmospheric circulation through local and non-local processes, which suggests that physiological forcing can also influence surface temperature through these processes (Kooperman et al., 2018).
As the magnitude of surface warming due to physiological forcing varies considerably between models, CO 2 physiological forcing also plays an important role in increasing uncertainty in climate prediction.Zarakas et al. (2020) reported that the physiological effect is a significant source of uncertainty in climate projection, accounting for 14% of the total cross-model spread over land.To reduce the uncertainty, it is necessary to understand the mechanisms of physiologically driven warming and the sources of the model diversity.To address uncertainties of physiologically driven terrestrial warming, Park et al. (2021) quantified the warming contribution of climate feedback (e.g., temperature, water vapor, albedo, and cloud feedback) triggered by physiological forcing based on the radiative kernel method.This study revealed that cloud feedback mainly amplifies the warming and uncertainty in mid-high latitudes (40°-70°N).However, this study mainly focused on mid-high latitude regions, not tropical regions, even though the changes in evapotranspiration (ET) by physiological forcing are maximized in tropics (Zarakas et al., 2020).
Among the tropic regions, physiological forcing-driven temperature changes and the inter-model spread are the largest in the Amazon (Figure 1, Figure S1a in Supporting Information S1).Several previous studies have highlighted the significant global impact of the Amazon region.Amazon rainforest contributes significantly to the absorption of atmospheric CO 2 , accounting for 15% of the total photosynthesis on the planet and storing over 123 ± 23 Pg of carbon (Dirzo & Raven, 2003).Field et al. (1998) found that the Amazon serves as a habitat for a significant proportion of the world's biodiversity, estimated to be around 25%. Due to the strong influence of the Amazon region, even small temperature changes in Amazon would have a huge impact on carbon cycle and ecosystem.However, the onset of temperature change in the Amazon region is already evident.Several studies have shown the temperature variations in the Amazon region are attributed to physiological forcing.Li et al. (2023) reported physiological forcing-driven temperature changes of 0.13 K, with a range spanning from 0.06 to 0.28 K in response to a 100 ppm increase in CO 2 levels in the Amazon.Thus, it is necessary to understand why such large surface warming will occur in the Amazon region due to physiological effect and where the uncertainties within the models come from.To identify the mechanism of physiological forcing-induced Amazon warming and the cause of its significant inter-model spread, we apply the contribution analysis method used in Park et al. (2020) to the Amazon region (see Section 2).For this analysis, we used the outputs of the 11 Coupled Model Intercomparison Project 6 (CMIP6) models that performed the 1pctCO2-bgc experiment.In this study, we defined CO 2 physiological forcing as the net effect of physiological responses of plants to the rising CO 2 concentration, including both CO 2 fertilization effect and stomatal effect, following Zarakas et al. (2020).The rest of this paper is organized as follows.Methods part describes the CMIP6 data sets, the experimental design, and the method for quantifying the contribution of climate feedback.In results part, we examine the major feedback inducing Amazonian warming, describe the warming process in detail, and identify the cause of inter-model spread.Finally, we summarize the paper and discuss the relationship between cloud feedback and atmospheric energy transport (AET).

CMIP6 Data and Experimental Design
To evaluate the impacts of CO 2 physiological forcing, we used 11 Earth system models (ESMs) (Table 1), which participate in the Coupled Climate-Carbon Cycle Model Intercomparison Project (C4MIP) from the CMIP6.The project provides an opportunity to isolate the CO 2 radiative forcing and physiological forcing.As shown in Table 2, 1pctCO2 has increasing CO 2 concentrations for both radiative and biogeochemical processes (full; radiative and physiological effects), 1pctCO2-rad includes increasing atmospheric CO 2 only for the radiative response (rad; radiative effect), 1pctCO2-bgc prescribes increasing CO 2 only for the biogeochemical process (phy; physiological effect), and piControl maintains pre-industrial CO 2 concentrations (con; control).The total duration of the experiments is 140 years, with CO 2 concentrations increasing by 1% per year up to the quadrupling of pre-industrial CO 2 level (285 ppm).To assess the impact of physiological responses to increasing CO 2 , we calculated differences between the average of 71 and 140 years of the physiological experiment (1pctCO2-bgc) and the average of the whole period of detrended control experiment (piControl) as a baseline.In the 1pctCO2, 1pctCO2-rad, and 1pctCO2-bgc experiments, the CO 2 level between years 71 and 140 changes from 572 to 1,136 ppm, representing an increase of 538 ppm compared to the pre-industrial CO 2 levels.

Quantification of Climate Feedback as Warming Contributions
According to the classic radiative-forcing framework, CO 2 radiative forcing is defined as an initial perturbation of top-of-atmosphere (TOA) radiation due to increasing CO 2, and the radiative feedback is defined as TOA radiation change by initial radiative forcing-induced changes in variable changes (Hansen et al., 1984;Roe, 2009).Based on this, many studies have developed methods to analyze the contribution of each feedback to warming by radiative forcing (Goosse et al., 2018;Kim et al., 2021;Pithan & Mauritsen, 2014;Shell et al., 2008;Soden & Held, 2006;Soden et al., 2008).These studies mainly divided the contribution to CO 2 radiative forcing into temperature, albedo, water vapor, and cloud feedbacks (Goosse et al., 2018;Park et al., 2021;Pithan & Mauritsen, 2014).Temperature feedback is the impact of changing temperature on outgoing longwave radiation, which can be decomposed into a Planck feedback triggered by radiation changes from uniformly warming surface and troposphere, and a lapse rate feedback resulting from vertically non-uniform warming.Water vapor feedback and albedo feedback represent the changes in radiation resulting from variation in water vapor and surface albedo, respectively.Cloud feedback refers to the change in radiative forcing due to cloud, influencing both longwave and shortwave radiation.As the concept of climate feedbacks is well established in the previous studies, we use this methodology to calculate the warming contribution of physiologically driven changes in climate feedback components.
The radiative imbalance at the TOA (∆R) and the initial forcing (F) can be expressed as where λ is the total feedback parameter and ΔT s is the surface temperature change (Gregory et al., 2015;Knutti & Hegerl, 2008;Winton et al., 2010;Yoshimori et al., 2016).Equation 1illustrates that the total radiation change at the TOA is represented by the sum of the initial radiative forcing and the feedback effect.In the case of CO 2 physiological forcing, there is no initial forcing since the 1pctCO2-bgc experiment excluded the radiative effect of CO 2 itself.Hence, it's crucial to scrutinize the radiative effects resulting from each climate feedback induced by physiological forcing.The total feedback parameter λ can be expressed as the sum of the individual feedback components as where λ T , λ A , λ W , and λ C are the feedback parameters of the temperature, albedo, water vapor, and cloud feedback, respectively (Park et al., 2021;Pithan & Mauritsen, 2014;Soden & Held, 2006).The temperature feedback is decomposed into Planck feedback and lapse rate feedback, so it is essential to take into account the vertical temperature distribution when calculating temperature feedback parameters.The parameter of Planck feedback (λ p ) is calculated from vertically uniform warming that is obtained by expanding the surface temperature change to the entire troposphere, while the parameter of lapse rate feedback (λ LR ) is calculated from the difference between vertical warming distribution and surface warming degree.
To quantify the contribution of each feedback to the local surface warming, we use the extended radiative kernel method (Goosse et al., 2018;Park et al., 2021;Pithan & Mauritsen, 2014), based on the local energy budget equation: In this equation, F is the initial radiative forcing, which is zero in this experiment (1pctCO2-bgc The third term on the right-hand side in Equation 4 represents the contribution of each feedback to the total local warming.For example, the contribution of the water vapor feedback to surface warming is the third term where λ i is λ w .The contributions of AET and surface heat flux to total warming are the fourth and fifth terms on the righthand side in Equation 4, respectively. Many previous studies have used the radiative kernel technique to calculate the feedback parameters for each variable (Park et al., 2021;Shell et al., 2008;Soden & Held, 2006;Soden et al., 2008).We can express the feedback parameter of variable X as where k i is the radiative kernel and X i is the climate feedback variable that is affected by changes in surface temperature (such as temperature, albedo, and water vapor).The radiative kernel k i is defined as the change in TOA radiation divided by a change in the variable X i and is obtained as the amount of radiation change that occurs when all other variables are fixed and only variable X i is changed in the model.The radiative kernel can be used to decompose the influence of each climate feedback on the surface temperature change.In this study, we applied the kernels generated by using CAM5 from Pendergrass et al. (2018).To consider the climate feedbacks only in the troposphere, we excluded the stratospheric effect by separating the troposphere and stratosphere.We assumed a tropopause height of 100 hPa in the tropics (30°S-30°N) and 300 hPa at the pole, decreasing linearly with latitude.
Unlike the other feedback, the cloud feedback cannot be quantified directly by the kernel method because of its strong nonlinearity.Soden et al. (2008) designed a method to calculate cloud feedback by using the cloud masking effect.In this method, cloud radiative forcing (C RF ) is defined as the difference in net radiation flux at TOA between all-sky (R(T,w,c,a)) and clear-sky (R(T,w,0,a)) conditions: The subsequent equation can be derived by applying differentiation on both sides of Equation 6 and organizing it: According to Equation 1, the change in radiation can be expressed as changes attributed to forcing and feedback.Thus, the change in radiation under clear-sky conditions equals the sum of the clear-sky forcing and the changes due to clear-sky feedbacks.Therefore, the radiation flux change at TOA can be expressed as the change in cloud radiative forcing (C RF ) and radiation fluxes in cloud-free conditions: where K 0 are the clear-sky kernels and G 0 is the clear-sky forcing.Plus, we can also express the change in radiation flux by the sum of the total-sky radiation flux changes: where δ C R is the cloud feedback, K is the total-sky kernels and G is the total-sky forcing.By subtracting Equation 8 from Equation 9, a new equation about the cloud feedback can be derived as shown below.
Earth's Future 10.1029/2023EF004223 Finally, we can use Equation 10to calculate the warming contribution of cloud feedback.

Physiological Forcing Driven Climate Change in Amazon Region
When CO 2 increased from 285 to 823 ppm, the ET decreased over most land areas due to CO 2 physiological forcing (Figure 1a).This is consistent with the previous study that the stomata effect outweighs the fertilization effect at high CO 2 concentrations (Skinner et al., 2017).However, the magnitude of the decrease in ET varies by region, with a larger decrease in tropical regions.In particular, the Amazon rainforest region shows the largest decrease in ET ( 0.22 ± 0.08 mm/day).In addition, the surface warming driven by CO 2 physiological forcing is greatest in Amazon region (Figure 1b).The physiological forcing-driven warming in the Amazon region (0.48 ± 0.42 K) is about three times greater than its global mean value (0.16 ± 0.19 K) and it is the largest temperature change compared to other tropical regions with similar ET changes.We also see the largest warming contribution of CO 2 physiological forcing to total CO 2 forcing in the Amazon region (Figure S1b in Supporting Information S1).Furthermore, the uncertainty in temperature change due to the physiological forcing is also the largest in the Amazon region (Figure S1a in Supporting Information S1).Therefore, we need to understand the mechanisms of the strong warming in the Amazon and the sources of uncertainty, which will be discussed in more detail in the following sections.

Contributions of Climate Feedbacks to Amazon Warming
To address the mechanism of surface warming in Amazon region, we quantified the contribution of each climate feedback to Amazonian warming by CO 2 physiological forcing.Figure 2 shows the relationship between the total surface temperature change due to physiological forcing and the sum of the warming contribution due to climate feedbacks in each model.It is clear that the sum of warming contribution of feedback is in good agreement with the simulated surface warming not only for the global mean but also for the Amazon.Furthermore, the correlation between the two are significant (0.99 for the global mean and 0.91 for the Amazon region).This result suggests that the surface warming of each model due to the physiological forcing can be reliably decomposed into the warming contributions of the climate feedbacks.Earth's Future 10.1029/2023EF004223 Figure 3 shows the contributions of each feedback to surface warming in the Amazon for each model.It is evident that the lapse rate and cloud feedbacks have larger contributions than the other feedbacks.The multi-model mean (MMM) contribution of the lapse rate feedback to the surface temperature change is 0.25 ± 0.36 K (53% to total warming), and that of the cloud feedback is 0.18 ± 0.42 K (37%) in the Amazon region (Table S1).Note that the sum of warming contribution of these two climate feedbacks is almost similar to the physiologically driven surface temperature change based on the MMM values.It suggests that these two feedbacks play an important role in amplifying surface warming by CO 2 physiological forcing.
It is also interesting that inter-model spread of lapse rate feedback explains much of the inter-model spread in total land warming, indicating that the lapse rate feedback is a key process in explaining the model diversity of physiological forcing-driven Amazonian warming.On the other hand, the cloud feedback shows the greatest cross-model spread, but the cloud feedback cannot explain the cross-model spread in surface warming.This is because the warming due to the cloud feedback tends to be partially offset by the cooling due to AET (Figure S2a in Supporting Information S1), which will be discussed further in the last section.In the following sections, we will further examine how the lapse rate and cloud feedbacks amplify the Amazonian warming and how they induce the uncertainties within the 11 models.Through this process, it will be possible to find out why some models have a large temperature change and some have a small temperature change.

Amazonian Warming by Lapse Rate Feedback and Its Inter-Model Spread
The lapse rate feedback depends on the vertical distribution of the temperature response, which determines the magnitude of outgoing longwave radiation for a given surface warming (Stone & Carlson, 1979).If the surface temperature rises more than the upper air temperature, the increase in surface radiation exceeds the change in longwave radiation emitted from the upper troposphere out of the Earth.It makes the greenhouse effect more efficient, providing positive feedback.Conversely, if the upper layer temperature increases more than the surface temperature, a surface cooling effect occurs by lapse rate, resulting in negative feedback.Accordingly, the significant enhancement of the Amazonian warming by the lapse rate feedback can be explained by a vertical structure of temperature change.The spatial pattern of changes in upper air temperature (700-300 hPa) is quite different from the pattern of surface warming (Figures 1b and 4).In tropical land areas including the Amazon, the upper-tropospheric warming is weaker than the extratropic.Due to this bottom-heavy warming structure, tropics lose less longwave radiation, and it increases the surface temperature (positive lapse rate feedback).This result is Earth's Future 10.1029/2023EF004223 quite interesting as lapse rate feedback by CO 2 physiological forcing is in the opposite direction to lapse rate feedback by total CO 2 forcing in the tropics (Pithan & Mauritsen, 2014).
The warmer and drier condition formed by physiological response suppresses convective activity resulting from reduced convective available potential energy (Langenbrunner et al., 2019).The weakening of deep convection can lead to cooling in the upper troposphere due to reduced latent heat release.Due to this effect, in the tropical land area, the upper tropospheric warming is relatively weak compared to the surface warming.This process is more pronounced in the Amazon, where deep convection is climatologically active and suppression of the convection due to physiological effect is the strongest among the tropical regions (Kooperman et al., 2018).Therefore, the surface warming due to physiological effect is intensified by the lapse rate feedback due to the bottom-heavy warming structure in the Amazon region.
As noted above, the different strength of the lapse rate feedback between models well explains the inter-model spread of Amazonian warming (Figure 3).This relationship is also clearly shown in Figure 5a.The correlation between the surface warming and the strength of lapse rate feedback is 0.86, which is significant at the 99% confidence level.Furthermore, although the strength of lapse rate feedback is affected by the difference in temperature between the upper and lower atmosphere, the upper atmospheric warming in tropics is so weak that the inter-model spread of warming contributions of the lapse rate feedback is well explained by the spread of surface temperature (Figure S3 in Supporting Information S1).Therefore, the strength of the lapse rate feedback is closely related to the amount of reduced surface ET, which is directly associated with surface warming and reduced convection (Figure 5b).The correlation coefficient is 0.72, which is significant at the 95% confidence level.

Amazonian Warming by Cloud Feedback and Its Inter-Model Spread
Physiological forcing can lead to changes in cloud fractions, which alters shortwave (SW) and longwave (LW) radiation, resulting in temperature change.The reflection effect caused by the high reflectivity of clouds controls the shortwave (SW) cloud feedback, and the greenhouse effect caused by the low temperature of clouds is related to the longwave (LW) cloud feedback.The net effect of clouds depends on their shape or height, and it is known that low clouds tend to have a net cooling effect due to the dominant reflection effect, and high clouds tend to have a net warming effect due to the prevailing greenhouse effect in general (Hartmann, 2015).
The cloud feedback was the second largest feedback in the context of MMM, which makes a significant contribution to the surface warming caused by physiological forcing.We further analyzed the mechanism of cloud feedback and its impact on surface temperature.Figure 6a shows the changes in total cloud cover due to the physiological forcing.It is clear that total cloud cover decreases over most land areas.These results are consistent  Earth's Future 10.1029/2023EF004223 forcing within the boundary layer.According to these studies, decreased ET makes drier and warmer boundary layer, leading to the vertical dipole pattern of humidity and temperature.In addition, the physiological forcing induces strengthened easterly in the boundary layer and this forcing-driven wind flow at the surface creates an updraft over the Andes, which in turn creates a downdraft over the Amazon forest region.Consistent with the previous studies, we confirm that the circulation change triggered by surface heating due to physiological forcing causes updrafts over the Andes and downdrafts over the Amazon forest, leading to an overall reduction in clouds in the Amazon region (Figure S4 in Supporting Information S1).Furthermore, we can also see the decrease in moisture and clouds due to the decrease in ET in the boundary layer and relative increase in lower free atmosphere in Figure S4 in Supporting Information S1.
The reflection effect of clouds is related to the SW cloud feedback, and the greenhouse effect of clouds is associated with the LW cloud feedback (Hartmann, 2015).The two feedbacks have opposite effects on surface temperature.It is generally accepted that low clouds tend to have a net cooling effect due to the dominant SW cloud effect and high clouds tend to have a net warming effect due to the prevailing LW cloud effect (Hartmann, 2015).To unravel the warming mechanism of cloud feedback in the Amazon region, we decomposed the cloud feedback into SW and LW components (Figure 6b).Out of 10 models, 9 models show surface warming by SW cloud feedback, and 8 models show surface cooling by LW feedback.The magnitude of the SW cloud feedback (MMM: 0.40 ± 0.35 K) is larger than the LW cloud feedback (MMM: 0.22 ± 0.38 K), and thus the net effect of cloud feedback is positive.That is, decrease in low cloud cover due to the physiological effect intensifies surface warming (positive SW cloud feedback).
As shown in Figures 3 and 6b, the inter-model spread of cloud feedback is considerable, which may contribute to the uncertainty in climate projections.The inter-model spread is somewhat larger for the LW cloud feedback ( 1.02 to 0.37 K) than for the SW cloud feedback ( 0.04 to 0.98 K).We further examined the cause of the crossmodel spread of SW and LW cloud feedbacks separately.As mentioned above, the SW cloud feedback is more closely related to the low clouds, so we suggested that the lower tropospheric relative humidity (LTRH) is important for the SW cloud feedback.To support this argument, we performed the correlation analysis and found a linear relationship between the SW cloud feedback and the LTRH (Figure 7a).The correlation coefficient between them is 0.78, which is significant at the 99% confidence level.This suggests that the LTRH significantly influences the diversity of the SW cloud feedback between climate models.In addition, the LTRH is strongly related to changes in ET triggered by physiological effect (Figure 7b).The correlation is 0.82, which is significant at the 99% confidence level.As a result, the differences in reduction in ET between models induce the uncertainties within the models in the SW cloud feedback, which ultimately intensifies the difference in surface air temperature responses.
On the other hand, the LW cloud feedback, which represents the greenhouse effect, is closely related to changes in high clouds.The formation of high clouds is influenced by the moisture advection from the surface, so both the surface moisture flux and the ascending motion are involved.The upper tropospheric relative humidity (UTRH) and the LW cloud feedback have a strong linear relationship (Figure 8a).Surface humidity and updrafts that induce vertical moisture advection can cause the changes in UTRH in the Amazon where water vapor recycling is strong (Liang et al., 2020;Spracklen et al., 2012).That is, tropical deep convection can affect the UTLH and the formation of upper-level clouds.For example, the high clouds increase with water vapor supply from the lower layer when there is an active deep convection.Out of 10 models, 7 simulate increases in omega velocity, indicating reduced deep convection, which induces decrease in UTRH.Thus, omega and UTRH show a strong correlation ( 0.69), which is significant at the 95% confidence level.Interestingly, the omega velocity and LW cloud feedback have a weaker relationship with surface ET in the inter-model space, with correlation coefficients of 0.13 and 0.57, respectively (Figure S5 in Supporting Information S1).This may indicate that the strength of the LW cloud feedback is more controlled by the physical parameterization of each model and the following circulation change than the surface moisture flux change.

Summary and Discussion
In this study, we investigate the impact of plant physiological processes on Amazonian warming and its uncertainty within the models based on the quantification of climate feedback.We found that the lapse rate and cloud feedbacks are the main contributors to the surface warming in Amazon region and its inter-model spread.
The bottom-heavy warming profile caused by physiological forcing induces the positive lapse rate feedback.This is because of strong surface warming driven by reduced ET and relatively weaker upper troposphere warming due to reduced convection.The inter-model spread of lapse rate feedback arises from differences in simulated changes in surface warming driven by reduced ET.The decrease in cloud cover due to reduced ET leads to increase in temperature through SW cloud feedback in the Amazon region.The cross-model spread of SW and LW cloud Earth's Future 10.1029/2023EF004223 feedback is caused by differences in ET and vertical wind changes between ESMs, respectively.In summary, the main factor influencing the degree of warming in the Amazon projected by models is linked to the response of ET to elevated CO 2 concentrations and its subsequent impact on convection processes.
It is interesting to note that the cloud feedback has a weaker relationship with the total surface temperature in inter-model space, although the contribution of the cloud feedback is substantial in each model (Figure 3).This is because the cloud feedback and AET contribute to opposite directions.The correlation between total cloud feedback and AET is 0.86, which is significant at the 99% confidence level (Figure S2a in Supporting Information S1).The results show a weaker correlation between SW cloud feedback (R = 0.28) and AET, and a higher correlation between LW cloud feedback and AET (R = 0.69, 95% confidence level; Figures S2b and S2c in Supporting Information S1).As the Amazon region is located in the tropics and surrounded by oceans, the land surface temperature is higher than the surrounding area (Figure S6 in Supporting Information S1).In addition, the surface warming due to physiological forcing is also greater than in the surrounding oceanic region.When the surface convergence increases, surrounding cold air may converge, resulting in surface cooling by the AET.Conversely, as deep convection and upper clouds increase, the LW cloud feedback causes surface warming.Therefore, the greater the warming effect of LW cloud feedback, the greater the cooling effect of AET.As the LW cloud feedback and AET can cancel each other out, SW cloud feedback makes a significant contribution to the total surface temperature (R = 0.79, 99% confidence level; Figure S7 in Supporting Information S1).
Finally, we note that this study has several caveats.First, we used the radiative kernel taken from CAM5 and applied to all CMIP6 models.However, as the CMIP6 models use different atmospheric components, they have different climate sensitivities.Therefore, further research is needed to consider different climate sensitivity of each model, although it requires a lot of computational resources and time.Second, this study focuses only on biogeophysical processes: plant physiological responses and albedo changes.Previous studies have shown that there are multiple ways in which vegetation interacts with climate, including biogeophysical and biogeochemical effects (Bonan, 2008;He et al., 2022).In response to increasing CO 2 concentrations, the biogeophysical response of plants leads to an increase in surface temperature through the stomatal closure effect, while the biogeochemical process leads to surface cooling through increased photosynthesis and CO 2 uptake (He et al., 2022).He et al. (2022) considered both effects on global warming and found that the net change in surface temperature, even the direction, differs from region to region.Therefore, for a comprehensive understanding of temperature changes by vegetation and their uncertainty, we need to consider biogeochemical feedback.Instead of using concentration-based experiments, conducting emission-based experiments could allow understanding the role of vegetation considering biogeochemical and biogeophysical feedback.

Figure 1 .
Figure 1.Changes in evapotranspiration and temperature caused by CO 2 physiological forcing.Global patterns of multi-model mean differences in the (a) evapotranspiration and (b) near-surface air temperature between 1pctCO2-bgc (71-140 years) and piControl (whole period) experiments.The black box (16°S-8°N, 79°-45°W) represents the Amazon region that is used for the feedback analysis.The significant values at the 95% confidence level are displayed with hatch.

Figure 2 .
Figure 2. Temperature change driven by CO 2 physiological forcing versus sum of warming from individual feedbacks.Relationship between near-surface temperature change and the sum of the warming contribution of climate feedbacks for the 11 Coupled Model Intercomparison Project Phase 6 models in (a) global and (b) Amazon region.Temperature changes are calculated by subtracting the mean value of piControl from the 1pctCO2-bgc experiment over 71-140 years.The gray dashed line is the identity line indicating the coincidence of the two components.

Figure 3 .
Figure 3. Inter-model spread of total land warming due to CO 2 physiological forcing and the contribution of each feedback to the warming in Amazon region.The x-axis represents the Amazon warming of each model and the y-axis represents the changes in surface temperature due to physiologically driven climate feedbacks.Colored lines are linear regressions of each feedback on the total land surface warming.Box plots show the distribution of the warming contribution from the feedbacks.The boxes indicate the median, 25th and 75th percentiles of the warming contributions, and the whiskers represent the full inter-model spread of the warming contributions.

Figure 4 .Figure 5 .
Figure 4. Atmospheric temperature change due to CO 2 physiological forcing.(a) Global pattern of multi-model mean atmospheric temperature change at 700-300 hPa atm.The black box (16°S-8°N, 79°-45°W) represents the Amazon region that is used for the feedback analysis.(b) Zonal-mean continental warming caused by physiological forcing as a function of pressure and latitude (30°S-80°N).The black dashed square represents the area (20°S-20°N, 700-300 hPa).The significant values at the 95% confidence level are displayed with hatch.

Figure 6 .
Figure 6.Cloud cover change due to CO 2 physiological forcing and decomposition of cloud feedback.(a) Global pattern of multi-model mean (MMM) differences in total cloud cover between 1pctCO2-bgc (71-140 years) and piControl (whole period) experiments.The box (16°S-8°N, 79°-45°W) represents the Amazon region that is used for feedback analysis.(b) Scatter plot of changes in near surface air temperature due to shortwave and longwave cloud feedbacks.Circles represent near-surface air temperature changes due to cloud feedbacks in each model, and black stars indicate the MMM of temperature changes due to each cloud feedback.The significant values at the 95% confidence level are displayed with hatch.

Figure 7 .
Figure 7. Inter-model spread of shortwave (SW) cloud feedback.Relationship between SW cloud feedback, lower tropospheric (925-850 hPa) relative humidity (lower tropospheric relative humidity (LTRH)) and evapotranspiration.Panel (a) is a scatterplot of temperature changes due to SW cloud feedback versus LTRH.Panel (b) is a scatterplot of changes in LTRH versus evapotranspiration.The black dashed lines represent the least squares regression fit.The one (*) and two asterisks (**) indicate that the p-value is less than 0.05 and 0.01, respectively.

Figure 8 .
Figure 8. Inter-model spread of longwave (LW) cloud feedback.Relationship between LW cloud feedback, upper tropospheric (400-150 hPa) relative humidity (upper tropospheric relative humidity (UTRH)), and vertical wind speed.Panel (a) is a scatterplot of temperature changes due to LW cloud feedback versus UTRH and panel (b) is a scatterplot of changes in UTRH versus upper tropospheric omega (dp/dt).The black dashed lines represent the least squares regression fit.The one (*) and two asterisks (**) indicate that p-value is less than 0.05 and 0.01, respectively.

Table 1
List of Coupled Model Intercomparison Project Phase 6 Earth System Models (Land Models) Used in This Study ∆R was ignored because the magnitude of ∆R in Equation 1 was relatively small.Furthermore, the warming contribution of each climate feedback can be calculated by dividing Equation 3 by the global mean of the Planck feedback parameter (λ p ) as shown below.
). λ p is the Planck feedback parameter, which can be decomposed into the global mean Planck feedback parameter (λ p ) and the local Planck feedback parameter (λ p ′). λ i is the other parameter for temperature, albedo, water vapor and cloud feedback.ΔAET is the change in AET and ΔSHF is the change in total surface energy flux.When Equation 1 was Earth's Future