Resilience of Amazon rainfall to CO2 removal forcing

Over the Amazon region, rainfall-induced changes to CO2 pathways significantly impact humans and multiple ecosystems. Its resilience is of vital importance, and idealized CO2 removal experiments indicate that declining trends in rainfall amounts are irreversible and exhibit a deficiency when the CO2 concentration returns to the pre-industrial level. The irreversible decline in Amazon rainfall is mainly due to the weakened ascent, further led by two main causes. (1) Enhanced tropospheric warming and a wetter atmospheric boundary layer over the tropics during CO2 removal generate a strong meridional gradient of temperature and specific humidity; driven by prevailing northeasterly winds, negative moist enthalpy advection occurs, which in turn weakens the ascent over the Amazon and results in anomalous drought. (2) The enhanced radiative cooling of atmospheric column. Driven by the negative lapse-rate feedback, the outgoing longwave radiative flux increases in the clear-sky atmosphere. As a result, the anomalous diabatic descent generates to maintain the energy balance of the atmospheric column. This result implies that the symmetric removal of CO2 does not guarantee full recovery of regional precipitation.


Introduction
The Amazon rainforest, the world's largest tropical rainforest, stores approximately 100 Pg carbon and is home to 10%-15% of terrestrial biodiversity (Langenbrunner et al 2019, Jiang et al 2021, Boulton et al 2022, Wang and Huang 2022).It is an important regulator of the global climate and is recognized as a potential tipping element in the Earth system (Lenton et al 2009, Boulton et al 2013, Wang and Huang 2022).However, anthropogenic global warming is increasing tree mortality and propelling rainforest loss (Cox et al 2000(Cox et al , 2008, Gatti et al 2021, Bauman et al 2022, Boulton et al 2022), which is causing the resilience of the rainforest, typically defined as the return rate from perturbation to diminish (Verbesselt et al 2016, Zemp et al 2017, Yi and Jackson 2021, Boulton et al 2022).Since the early 2000s, resilience loss has been found in more than 75% of the Amazon rainforest (Boulton et al 2022, Forzieri et al 2022).The Amazon rainforest cover observed in 2019 was already reduced by almost 20% when compared to the coverage area that existed in 1988 (Davidson et al 2012, Aguiar et al 2016, Zemp et al 2017).Under a business-asusual scenario, deforestation would reach approximately 45% of the original extent of the forest by 2050, while longer-term deforestation would likely reach up to 100% (Zemp et al 2017).Climate change accelerates the loss of the Amazon rainforest by increasing the length of the dry season (Fu et al 2013, Adams et al 2017), drought frequency (Boulton et al 2022), and fire activity (Brando et al 2020, Wang and Huang 2022).Such loss leads to weaker carbon uptake and an acceleration of the atmospheric CO 2 growth rate, which forms positive feedback between the biosphere and atmosphere, and further exacerbates global warming.
Among the climate factors influencing the resilience of the Amazon rainforest, rainfall is particularly important.Several studies have suggested that the loss of resilience is faster in regions that receive less rainfall (Brando et al 2014, Verbesselt et al 2016, Bauman et al 2022, Boulton et al 2022, Forzieri et al 2022).During several once-in-a-century drought events (e.g.2015), the Amazon experienced a higher rate of tree mortality under drought stress, exhibiting a more severe loss of resilience (Tomasella et al 2008, Phillips et al 2009, Allen et al 2010, Lewis et al 2011, Saatchi et al 2013, Brando et al 2014, Bi et al 2016, Verbesselt et al 2016, McDowell et al 2018, Aleixo et al 2019).Drought not only reduces ecosystem respiration (Meir et al 2008, Doughty et al 2015, Thakur et al 2018) and causes tree mortality through 'hydraulic failure' (Bréda et al 2006) or 'carbon starvation' (McDowell et al 2008), but also indirectly affects tree death by increasing pest and disease disturbance (Kurz et al 2008, Seidl et al 2018), as well as the occurrence of fires (Brando et al 2014, Gatti et al 2014).Tree mortality weakens forest-rainfall feedback and promotes further drought occurrence (Staal et al 2020).Consequently, the resilience of the Amazon rainforest will likely diminish (Forzieri et al 2022) under global warming scenarios.
Given the importance of rainfall to the Amazon rainforest, it is essential to understand changes in rainfall over the Amazon, especially the changes during the vegetation growing season (December-February) (Xie et al 2022).Previous studies on the Amazon rainfall mainly focused on the current climate and global warming scenario (Pascale et al 2019, Thome Sena and Magnusdottir 2020, Almazroui et al 2021, Liu et al 2022, Torres et al 2022), and various driving factors have been identified.For example, El Niño is typically accompanied by drought in the Amazon, which is modulated by the Walker circulation or extratropical Rossby wave train related atmospheric circulation anomalies (Cai et al 2020).Moreover, the impact of ENSO on Amazon rainfall is projected to intensify under global warming.Additionally, the altered Madden-Julian Oscillation could also impact Amazon rainfall through teleconnection (Liu et al 2022).However, the 2015 Paris Agreement proposed a target of, 'holding the increase in the global average temperature to well below 2 • C above pre-industrial (PI) levels and pursuing efforts to limit the temperature increase to 1.5 • C above PI' .Toward this target, it is necessary to remove CO 2 from the atmosphere (Peters et al 2013, Field and Mach 2017, Rogelj et al 2018, Wang et al 2021, Huang et al 2022).If CO 2 removal (CDR) methods are applied, the atmospheric CO 2 concentration will peak in this century and then continue to decline.Therefore, there has been great interest in the climate effects of CDR.Several studies have focused on understanding the hysteresis and reversibility of climate systems under a CDR scenario (Wu et al 2014, Kim et al 2022, Kug et al 2022, Park and Kug 2022, Schwinger et al 2022, Song et al 2022, Zhang et al 2023).One of the most important climate changes under the CDR scenario is that the surface temperature and precipitation in most areas of the world experience irreversible and hysteretic responses (Kim et al 2022) due to heat accumulated in the deep oceans during the previous stage.This indicates that changes in the global hydrological cycle under current global warming cannot be reversed even if the CO 2 concentration is reduced.Considering this, we speculate that the Amazon, with drying trends under global warming (Almazroui et al 2021), may have been drier during the CDR period.The rainfall in the Amazon region can remarkably influence the Amazon rainforest, so studies on how Amazon rainfall would respond under the CDR scenario are strongly desired.How does the resilience of Amazon rainfall behave in the vegetation growth season?That is, whether Amazon rainfall is able to recover to the initial condition when CO 2 returns to its PI concentration.Furthermore, possible causes of the changes in Amazon rainfall need to be resolved in more detail.

Data
In this study, the observational and reanalysis products employed are (1) Global Precipitation Climatology Project (GPCP) precipitation data (Adler et al 2003), and (2) National Centers for Environmental Prediction-Department of Energy (NCEP-DOE) reanalysis data, including wind variables (Kanamitsu et al 2002).The monthly mean variables are adopted.All the above datasets cover a common period of 1979-2014, with a horizontal resolution of 2.5 • × 2.5 • .
As listed in table S1, four experiments from the Coupled Model Intercomparison Project six (CMIP6) are used, including historical experiment, piControl experiment, 1%CO 2 experiment, and 1%CO 2 -CDR experiment (Eyring et al 2016, Keller et al 2018).Both the 1%CO 2 and 1%CO 2 -CDR experiments are 140 years in length, and are referred to as the CO 2 ramp-up and ramp-down experiments.The 1st year in the 1%CO 2 experiment is defined as Year 1. Eight available models, including ACCESS-ESM1-5, CanESM5, CESM2, CNRM-ESM2-1, GFDL-ESM4, MIROC-ES2L, NorESM1-LM, and UKESM1-0-LL, were used in this work.Vertical velocity data, surface upwelling radiation data, and top of atmosphere (TOA) incident shortwave radiation data in UKESM1-0-LL are not available.The surface upwelling radiation data of CESM2 is not available.
To obtain the Multi-Model Ensemble (MME), we interpolated the outputs onto a 1 • × 1 • grid.The ensemble spread range among the eight models was defined as 1.96 × standard deviation.The results were deemed robust if more than 75% of the members (6 out of 8) display the same sign as the MME.The anomalies are the outputs of the CO 2 ramp-up and ramp-down experiments subtracted by the climatology of the last 100 years in the piControl experiment.The statistical significance of the mean difference was assessed by the two-sided Student's t test.The 21 year running mean was employed to remove interannual variability.The boreal winter (December-February) average was analyzed when the vegetation growth over the Amazon was mainly concentrated.

The resilience of Amazon rainfall
The resilience of Amazon rainfall is defined as whether the Amazon rainfall can recover to the preindustrial level by removing the atmospheric CO 2 concentration.During the CO 2 removal period, the extent to which Amazon rainfall recovers to preindustrial level, AR (t), is expressed as the ratio of the change in rainfall relative to the CO 2 peaks to the change in rainfall during the CO 2 ramp-up period.Therefore, the equation is written as follows: R 140 , R PI represent the Amazon rainfall at the CO 2 peaks and pre-industrial levels.R t is the Amazon rainfall at a given year t during the CO 2 removal period.A stable AR (t) ⩽ 0 indicates that the Amazon rainfall is recovering to the pre-industrial level.Figure 1 illustrates the related variables in formula (1).

The moisture budget
To explore the reasons for the rainfall responses over the Amazon, a simplified moisture budget equation was employed (Huang et al 2013): where ∆ indicates the anomalies relative to the PI level.P, q s , and ω 500 are precipitation, specific humidity at 850 hPa, and pressure velocity at 500 hPa, respectively.The terms on the right-hand side of the equation ( 2) represent the thermodynamic term related to the moisture changes (−∆q 850 • ω 500 ) and the dynamic term related to the circulation changes (−∆ω 500 • q 850 ).The nonlinear processes and horizontal advection term of water vapor were omitted here because they are tiny in the tropics (Held and Soden 2006, Seager et al 2010).1)).R140, RPI represent the Amazon rainfall at the CO2 peaks and pre-industrial levels, respectively.Rt is the Amazon rainfall at a given year t during the CO2 removal period.A stable AR (t) ⩽ 0 indicates that the Amazon rainfall is recovering to the pre-industrial level.

The moist static energy (MSE) budget
Vertical velocity is constrained by the MSE balance in the tropics (Neelin and Held 1987).Because the seasonal mean time tendency of anomalous MSE can be neglected, the simplified anomalous MSE equation was employed: where MSE is h = C p T + L v q + Φ and moist enthalpy is represented by k = C p T + L v q.The superscript ′ , overbar −, and angle bracket ⟨⟩ represent the anomaly, the mean, and the mass-weighted vertical integration from surface to 100 hPa, respectively.ω, C p , T, q, L v , Φ , and NL are the pressure velocity, specific heat at constant pressure, air temperature, specific humidity, latent heat of vaporization, geopotential height, and nonlinear term, respectively.Moreover, F net is the net energy flux into the atmospheric column from the surface and TOA, which is equal to where S, R are the shortwave and longwave radiative flux.SH and LH are the sensible and latent flux.
The subscripts s and t denote the surface and TOA, respectively.The superscript arrows (↑ and ↓) represent the direction of radiative fluxes.For all these terms in equation ( 4), the positive value implied that the atmospheric column gained heat. Namely, the downward direction at the TOA and the upward direction at the surface were positive.The sum of the first seven terms on the right-hand side of equation ( 4) is the radiative flux, of which the first four terms can be combined as the net shortwave radiation flux (S ′ net ), and the last three can be combined as the net longwave radiation (R ′ net ).The net longwave radiation can be further divided into clear-sky (R ′ clear ) and cloud-related (R ′ cloud ) terms.Therefore, equation ( 4) can be simplified to (5) In the Amazon, where active deep convection prevails, the vertical profile of anomalous vertical velocity (ω ′ ) shows a top-heavy structure with a maximum at approximately 400 hPa (figure S1), which is classified as the deep mode of tropical vertical motion (Back and Bretherton 2009).In contrast, the vertical profile of climatological MSE (h) exhibits a bottom-heavy structure, and its maximum value is at approximately 800 hPa (figure S1).According to previous studies, in the tropics, the ascending motion of the deep mode typically exports the MSE out of the atmospheric column, causing an increase in gross moisture stability (Back and Bretherton 2009).If the summation of the terms on the right-hand side of equation ( 3) is negative (positive), corresponding to the reduced (increased) MSE in the column, the anomalous descending (ascending) would be generated to decrease (increase) the MSE exported out of the column and thus maintain the balance of the MSE budget.

Evaluation of the CMIP6 rainfall
Before investigating the resilience of Amazon rainfall to idealized CO 2 forcing, the reproducibility of the CMIP6 models on Amazon rainfall is evaluated.Here, we compared the climatological observed rainfall with the MME of historical experiments in CMIP6.
The observed result exhibited heavy rainfall in the Amazon, with a maximum of approximately 11 mm d −1 (figure S2).The CMIP6 MME can grasp this spatial distribution, although the rainfall center was slightly southeastward and was stronger.The spatial correlations between individual models and observation all exceeded 0.9, with the highest correlation at 0.96 for CMIP6 MME (figure S3).Moreover, the northeast wind prevails in the Amazon in winter, which can be well reproduced by the CMIP6 MME.The above evaluation suggested that the CMIP6 MME is sufficient to investigate the resilience of Amazon rainfall to symmetric CO 2 removal.

Irreversible change in Amazon rainfall under the CDR scenario
As shown in figure 1(a), the Amazon rainfall change is asymmetric and lags behind the evolution of the CO 2 concentration in the CO 2 ramp-up and rampdown experiments.Amazon rainfall is defined as the regional average of land rainfall over the key domain (17.5 • S-7.5 • N, 70 • W-55 • W; red rectangles in figure 2).It decreases as the CO 2 concentration increases; after CO 2 peaks, this decline persists for approximately 25 years, up to 18% of the rainfall as that during the PI period.Thereafter, Amazon rainfall begins to recover at 40th year during the CO 2 ramp-down period (figure S4).However, this recovery is gradual and only about 71.1% (25-75 percentile: 62.86%-83.35%) of increased rainfall has been recovered when the CO 2 concentration returns to the PI level (figure S4), with anomalous drought compared to that of the PI level.To show the spatial distribution associated with the regionally averaged rainfall anomalies, two periods of the same CO 2 concentration were selected to analyze the resilience of Amazon rainfall: years 1-40 in the CO 2 rampup period (hereafter RU) and years 240-279 in the CO 2 ramp-down period (hereafter RD; gray bands in figure 2(a)).Precipitation change is insignificant in the Amazon during the RU period (figure 2(b)); however, during the RD period, it shows a significant deficiency compared to PI levels (figure 2(c)).Even with the same CO 2 concentration, the Amazon is significantly drier during the RD period (figure 2(d)).Overall, the simulation results imply that if the symmetrical CDR is implemented, Amazon rainfall is irreversible and may exhibit drier conditions when the CO 2 concentration returns to the unperturbed PI level.
To explore the cause of the irreversible response of Amazon rainfall, a simplified moisture budget equation, which divides the precipitation into thermodynamic (−∆q 850 • ω 500 ) and dynamic (−∆ω 500 • q 850 ) terms, is applied (equation ( 2)).The differences in these two terms between the RD and RU periods are shown in figure 3. The summation of the thermodynamic and dynamic terms reproduces the equivalent changes in rainfall (figure 3(c); figure 2(d)), further indicating a good approximation of equation ( 2) for the tropics (Zhou et al 2022).The thermodynamic component (figure 3(a)) represents the 'wetget-wetter' mechanism, with the increased precipitation over the climatological rain band where upward motion prevails.In contrast, the dynamic component (figure 3(b)) is generally opposite to the thermodynamic component, and the amplitude is much larger.Moreover, the spatial pattern of the dynamic term bears a close resemblance to the change in precipitation (figure 2(d)).As a result, the anomalous drought conditions in the Amazon during the RD period relative to the RU period are mainly led by the dynamic component.The weakening of the ascent in the Amazon during the RD period relative to the RU period leads to the irreversible response of Amazon rainfall, as well as anomalous drier conditions locally.

Physical mechanisms behind the irreversible response
Moisture budget analysis suggests that the drier condition in the Amazon is caused by the weakened ascent during the RD period.Below, the cause of the weakened vertical velocity is explored further from    S1).The related physical processes are analyzed as follows.
Among the terms of the MSE budget, the MEA term contributes the most to the weakened ascending motion in the Amazon.The moisture enthalpy (k) is a function of temperature and specific humidity, so both anomalous temperature advection (figure 5(a)) and moisture advection (figure 5(b)) may lead to a weakened vertical velocity.Although the CO 2 concentrations during the RD and RU periods are the same, the global surface temperature during the latter period is warmer than that in the former period due to the ocean's large thermal inertia (Wu et al 2014, Sun et al 2021, Yeh et al 2021).Following an approximately moist adiabat lapse rate, the troposphere over the whole tropics shows larger warming during the RD period relative to the RU period (figure 5(c)).Meanwhile, the equatorial centraleastern Pacific exhibits enhanced El Niño-like warming (Kug et al 2022, Oh et al 2022, Song et al 2022, Zhou et al 2022), which further heats the free troposphere by equatorial wave adjustments (Su andNeelin 2002, Neelin andSu 2005).According to the Clausius-Clapeyron expression, as the troposphere warms, the atmospheric boundary layer must become moister (Held andSoden 2006, Hu et al 2021).Thus, in the tropics, there is a strong meridional gradient of tropospheric temperature and specific humidity at 850 hPa (Shading in figures 5(c) and (d)).Driven by the climatological northeasterly wind (figures 5(c) and (d)), cold and dry air is advected into the Amazon.Consequently, it leads to a negative MEA and a weakened ascent in the Amazon, contributing to the irreversible precipitation response under the CO 2 removal scenario.
The term F ′ net plays a secondary role (figure 4).The Amazon is dominated by negative F ′ net , which is mainly led by the anomalous clear-sky longwave radiation R ′ clear and latent heat (table S1). Figure 6 shows the spatial pattern of F ′ net , R ′ clear , and LH ′ .The term R ′ clear weakens during the RD period, which suggests an intensification of radiative cooling in the cloud-free atmosphere.To maintain the energy balance of the atmospheric column, an anomalous diabatic descent occurs (Rodwell and Hoskins 1996), which leads to drier conditions over the Amazon region.The strengthened radiative cooling of the atmospheric column can be attributed to the negative lapse-rate feedback (Colman 2003, Soden andHeld 2006).The global surface warming is enhanced during the RD period compared to the RU period.Following an approximately moist adiabat lapse rate, the warming is greater in the upper troposphere than in the lower one over the Amazon, resulting in a more stable atmosphere.The enhanced warming in the upper troposphere contributes to the emission of outgoing longwave radiation, thus cooling the atmospheric column.Although this effect can be offset by the positive water vapor feedback (Held and Soden 2000, Beer and Eisenman 2022), i.e. the enhanced global surface warming allows the atmosphere to hold more water vapor, which heats the atmosphere by blocking more outgoing longwave.Eventually, the net feedback is negative, leading to the atmospheric column cooled by clear-sky longwave radiation, and thus the anomalous descent.In addition, the anomalous latent heat flux also contributes, while it is partially offset by the positive anomalies

Summary and discussion
The resilience of Amazon rainfall in an idealized CDR simulation is investigated in this paper.It showed that if symmetric CDR is applied, Amazon rainfall decline is largely irreversible when CO 2 concentration returns to its pre-industrial level, with the Amazon incurring anomalous drought conditions.If the rainfall change is divided into thermodynamic and dynamic components, they exhibit opposite-sign contributions to the irreversible precipitation change in the Amazon.It is the weakened dynamic component associated with the vertical velocity that dominates the irreversible Amazon rainfall change.
Moreover, what are the driving mechanisms behind the weakening of upward motion in the Amazon during the RD period?We attempt to resolve this issue through MSE budget analysis.The MSE budget reveals that there are two reasons for the weakened ascent in the Amazon: (1) the negative The climatological northeasterly advects dry and cold air (negative MEA anomalies) to the Amazon, which reduces the MSE in the atmospheric column.Hence, the upward motion in the Amazon weakens, making it drier.(2) The negative anomalous clear-sky longwave radiative flux R ′ clear .In the Amazon, the radiative cooling in the cloud-free atmosphere strengthens during the RD period due to the negative lapse-rate feedback.Correspondingly, an anomalous diabatic descent occurs.These two processes jointly leads to the weakened ascendance and the irreversible response of the Amazon rainfall.
In this study, we analyze the direct causes of the asymmetric response of Amazon rainfall, i.e. the negative MEA and negative radiative flux of the atmospheric column result in anomalous descent during the RD period.Essentially, the negative MEA is linked to the SST anomalies over the tropical Pacific Ocean, i.e. the enhanced El Niño-like warming.As outlined in section 3.2, our analysis reveals that this anomalous SST warming causes tropical tropospheric warming and contributes to the negative MSE.For the formation of enhanced El Niño-like warming during the CO 2 ramp-down period, previous studies attributed it largely to the lagged recovery of subthermocline ocean temperature and the Walker circulation (Song et al 2022, Zhou et al 2022, Zhang et al 2023).During the CO 2 removal period, the subthermocline ocean warms stronger than the upper one, which generally weakens the vertical temperature gradient and reduces the cooling effect led by the climatological upwelling in the central and eastern Pacific.Moreover, the Walker circulation weakens due to temperature and precipitation constraints, accompanied by a decrease in upwelling intensity.In combination with the processes of the Walker circulation and ocean thermal stratification, the enhanced El Niño-like warming becomes more apparent during the RD period.However, the quantitative relative importance of these two processes in the formation of enhanced SST patterns remains unclear, which calls for further analysis in future work.
Our study provides implications for the impacts of CDR on Amazon rainfall.The symmetric CDR does not result in a fully recovery of Amazon rainfall, which may threaten the growth of rainforests.This indicates that the persistent impacts of CO 2 should be considered when developing climate mitigation pathways.In addition, the environmental conditions in the Amazon rainforest are intricate and are regulated by various climate factors, such as heat, wildfire, and El Niño-Southern Oscillation, among others (Brando et al 2014, Aleixo et al 2019).In particular, there is a strong interaction between climate and intrinsic physiological processes (Leite-Filho et al 2021, He et al 2022).Hence, more comprehensive components need to be considered when understanding the impacts of CDR on the Amazon rainforest in further studies.

Figure 1 .
Figure1.Illustration of the definition of Amazon rainfall resilience (formula (1)).R140, RPI represent the Amazon rainfall at the CO2 peaks and pre-industrial levels, respectively.Rt is the Amazon rainfall at a given year t during the CO2 removal period.A stable AR (t) ⩽ 0 indicates that the Amazon rainfall is recovering to the pre-industrial level.

Figure 2 .
Figure 2. Evolution (a) and spatial pattern (b)-(d) of the Amazon rainfall response in CO2 ramp-up and ramp-down experiments.(a) Time series of the CO2 concentration (blue line; multiplied by −1 for better display; unit: ppm) and anomalous rainfall (red line; unit: mm d −1 ).The vertical dashed line denotes the CO2 peak year.Pink shadings indicates the 1.96 × standard deviation for the eight model members.The interannual variability is removed by 21 year running means.Two vertical gray bands in Years 1-40 (RU) and Years 240-279 (RD) are the two 40 year periods with the same average CO2 concentration.(b) and (c) are the anomalous precipitation patterns during the RU and RD periods, respectively.(d) Is the difference between (c) and (b).Black dots in (b), (c) denote that the sign agreement test is satisfied.Black dots in (d) indicate the regions where the 75% sign agreement test is satisfied and the average difference between the RU and RD periods is statistically significant at a 90% confidence level.The Amazon region is marked by red boxes.

Figure 3 .
Figure 3.The difference in the thermodynamic component (a), dynamic component (b), and their sum (c) between the RU and RD periods.The units are: mm d −1 .The dots indicate the grids at which more than 75% of the models have the same sign as those of MME.The red boxes are the same as in figure 2(a).

Figure 4 .
Figure 4.The difference in the terms in the MSE budget equation (unit: W m −2 ) over the Amazon (red rectangles in figure 2) between RD and RU periods.The error bars reflect the ensemble spread of the eight models.

Figure 5 .
Figure 5. (a) The difference in the horizontal advection of anomalous temperature by climatological wind between the RD and RU periods (unit: W m −2 ).(b), (c), and (d) are same as (a) but for the horizontal advection of anomalous moisture driven by climatological wind (b; unit: W m −2 ), tropospheric-averaged (850-200 hPa) temperature anomalies (c; shading; unit: K) and 850 hPa climatological wind (c; vectors; unit: m s −1 ), anomalous specific humidity (d; shading; unit: g kg −1 ) and wind at 850 hPa (d; vectors; unit: m s −1 ).The black dots and red boxes are the same as in figure 1(a).Vectors only exceeding 75% of sign consistency are shown.

Figure 6 .
Figure 6.(a) The difference in energy flux into the atmospheric column between RD and RU periods (unit: W m −2 ).(b), and (c) are the same as (a) but for the clear-sky longwave radiative flux, and latent heat flux, respectively (unit: W m −2 ).The black dots and red boxes are the same as in figure 1(a).
Beer E and Eisenman I 2022 Revisiting the role of the water vapor and lapse rate feedbacks in the Arctic amplification of climate change J. Clim.35 2975-88 Bi J, Myneni R, Lyapustin A, Wang Y J, Park T, Chi C and Knyazikhin Y 2016 Amazon forests' response to droughts: a perspective from the MAIAC product Remote Sens. 8 356 Boulton C A, Good P and Lenton T M 2013 Early warning signals of simulated Amazon rainforest dieback Theor.Ecol.6 373-84 Boulton C A, Lenton T M and Boers N 2022 Pronounced loss of Amazon rainforest resilience since the early 2000s Nat.Clim.Change 12 271-8 Brando P M, Balch J K, Nepstad D C, Morton D C, Putz F E, Coe M T and Soares-Filho B S 2014 Abrupt increases in Amazonian tree mortality due to drought-fire interactions Proc.Natl Acad.Sci.USA 111 6347-52 Brando P M, Soares-Filho B, Rodrigues L, Assuncao A, Morton D, Tuchschneider D and Coe M T 2020 The gathering firestorm in southern Amazonia Sci.Adv. 6 eaay1632 Bréda N, Huc R, Granier A and Dreyer E 2006 Temperate forest trees and stands under severe drought: a review of ecophysiological responses, adaptation processes and long-term consequences Ann.For.Sci.63 625-44 Cai W J et al 2020 Climate impacts of the El Niño-Southern Oscillation on South America Nat.Rev. Earth Environ. 1 215-31 Colman R 2003 A comparison of climate feedbacks in general circulation models Clim.Dyn.20 865-73 Cox P M, Betts R A, Jones C D, Spall S A and Totterdell I J 2000 Acceleration of global warming due to carbon-cycle feedbacks in a coupled climate model Nature 408 184-7 Cox P M, Harris P P, Huntingford C, Betts R A, Collins M, Jones C D and Nobre C A 2008 Increasing risk of Amazonian drought due to decreasing aerosol pollution Nature 453 212-5 Davidson E A, de Araujo A C, Artaxo P, Balch J K, Brown I F, MM C B and Wofsy S C 2012 The Amazon basin in transition Nature 481 321-8 Doughty C E, Metcalfe D B, Girardin C A, Amezquita F F, Cabrera D G, Huasco W H and Malhi Y 2015 Drought impact on forest carbon dynamics and fluxes in Amazonia Nature 519 78-82 Eyring V, Bony S, Meehl G A, Senior C A, Stevens B, Stouffer R J and Taylor K E 2016 Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization Geosci.Model Dev. 9 1937-58 Field C B and Mach K J 2017 Rightsizing carbon dioxide removal Science 356 706-7 Forzieri G, Dakos V, McDowell N G, Ramdane A and Cescatti A 2022 Emerging signals of declining forest resilience under climate change Nature 608 534-9 Fu R, Yin L, Li W, Arias P A, Dickinson R E, Huang L and Myneni R B 2013 Increased dry-season length over southern Amazonia in recent decades and its implication for future climate projection Proc.Natl Acad.Sci.USA 110 18110-5 Gatti L V, Basso L S, Miller J B, Gloor M, Gatti Domingues L, Cassol H L G and Neves R A L 2021 Amazonia as a carbon source linked to deforestation and climate change Nature 595 388-93 Gatti L V, Gloor M, Miller J B, Doughty C E, Malhi Y, Domingues L G and Lloyd J 2014 Drought sensitivity of Amazonian carbon balance revealed by atmospheric measurements Nature 506 76-80 He M Z, Piao S L, Huntingford C, Xu H, Wang X H, Bastos A and Gasser T 2022 Amplified warming from physiological responses to carbon dioxide reduces the potential of vegetation for climate change mitigation Commun.Earth Environ. 3 160 Held I M and Soden B J 2000 Water vapor feedback and global warming Annu.Rev. Energy Environ. 25 441-75 Held I M and Soden B J 2006 Robust responses of the hydrological cycle to global warming J. Clim.19 5686-99