Asymmetrical response of summer rainfall in East Asia to CO2 forcing

Understanding the regional hydrological response to varying CO2 concentration is critical for cost-benet analysis of mitigation and adaptation polices in the near future. To characterize summer monsoon rainfall change in East Asia due to a change in the CO2 pathway, we used the Community Earth System Model (CESM) with 28 ensemble members in which the CO2 concentration increases at a rate of 1% per year until its quadrupling peak, i.e., 1,468 ppm (ramp-up period), followed by a decrease of 1% per year until the present-day climate conditions, i.e., 367 ppm (ramp-down period). Although the CO2 concentration change is symmetric in time, the rainfall response is not symmetric. The amount of summer rainfall in East Asia is much larger during a ramp-down period than during a ramp-up period when the two periods of the same CO2 concentration are compared. This asymmetrical rainfall response is mainly due to an enhanced El Niño-like warming pattern as well as an increase in the meridional sea surface temperature gradient in the western North Pacic during a ramp-down period. These sea surface temperature patterns enhance the atmospheric teleconnections to East Asia and the local meridional circulations around East Asia, resulting in more rainfall over East Asia during the ramp-down period. This result implies that the removal of CO2 does not guarantee the return of regional rainfall to the previous climate state with the same CO2 concentration.


Introduction
Continued anthropogenic greenhouse gas emissions pose a threat that could change the climate system.
The rapid increase in CO 2 concentration has perturbed radiative forcing in the atmosphere, resulting in an increase in the global mean surface temperature (GMST) of approximately 1°C compared to the preindustrial period 1,2 . These changes may also lead to a change in the climate system to a severe, pervasive, and irreversible state, causing devastating changes to the biogeochemical and hydrological cycles, ecosystems, and biodiversity 3,4 . To reduce such risks, the removal of CO 2 from the atmosphere is considered to be important for achieving climate change mitigation goals 5,6 . In light of these challenges, understanding the changes to the climate system in response to a carbon dioxide removal (CDR) scenario is critical for climate mitigation and adaption actions in the near future 7 .
The climate system response in a CDR scenario is often characterized by hysteresis effects and irreversible changes [7][8][9][10] . Recently, the CDR-induced climate change has been investigated using state-ofthe-art climate models [9][10][11][12][13] . Among these, the response of the global hydrological cycle has been widely examined due to the fact that it signi cantly in uences the properties of extreme weather and climate events including droughts, oods, and effects on water supplies. For instance, the intensi ed El Niño-like warming pattern induced by a CDR could lead to an asymmetrical response of the rainfall pattern across the globe [12][13][14] . In addition, the South Asian monsoon season has also shown an asymmetrical rainfall response to changes in the CO 2 pathway, which has primarily originated from the slow ocean response 12 .
These results indicate the need for a wide understanding of the regional rainfall change to the CDR scenario.
The East Asian Summer Monsoon (EASM) is one of the most distinct components of the Asian monsoon system 15 . Its variability has signi cant impacts on weather and climate conditions with substantial social and economic in uence on the local and global community [16][17][18] . Thus, understanding its future changes is of fundamental societal and scienti c importance. The possible changes in the EASM in a warmer climate have been broadly examined using the Coupled Model Intercomparison Projection Phase 5 (CMIP5) climate models [19][20][21] . It is particularly reported that the EASM rainfall has been projected to increase by 6.4%/K marked by an increase in summer monsoon rainfall 19 . However, there have been fewer studies focusing on the EASM's response to the CDR scenario.
The purpose of this paper is to investigate the rainfall changes in the EASM in response to a change in the CO 2 pathway based on idealized CO 2 experiments (Methods section) and their related physical mechanisms. The results may be helpful for providing useful information on water mitigation and adaptation in this region in a changing climate.

Results
Global mean surface temperature and rainfall response We rst show the time series of GMST and rainfall with a change in the CO 2 pathway (Fig. 1b). The GMST continues to increase for several years after a quadrupling peak of CO 2. This phenomenon could be interpreted as a period when the GMST initially responds quickly to a decrease in the CO 2 concentration as the mixed ocean layers cool, followed by a slow decline period due to the release of heat previously accumulated in the ocean 10,12 . This trend is also seen in the global mean rainfall that continues to increase for almost two decades after the CO 2 quadrupling peak, which is likely due to the increase in the moisture convergence and surface evaporation under global warming conditions as shown in previous studies 19,22,23 . Such global rainfall hysteresis could be explained by accumulated heat storage in the ocean 9 and the fast rainfall response to direct CO 2 forcing 24 .
During the CO 2 stabilization period, the GMST and rainfall remains higher than the PD climate state when the CO 2 concentration is the same (Figs. 1a,b). The equilibrium state can be reached when CO 2 stabilization forcing is used for more than 1,000 years 25 . These results further emphasized the ocean's large thermal inertia 9,25 , implying that the response of the GMST and rainfall in a changing CO 2 pathway are conditionally irreversible.
Asymmetrical summer rainfall response in East Asia to a change in the CO 2 pathway Hereafter, we analyze the regional rainfall pattern over East Asia between the ramp-up period (2090-2139, herein referred to as the RU period) and ramp-down period (2141-2190, hereafter referred to as the RD period) during boreal summer (June-July-August) ( Fig. 2a and see also Fig. 1a). It should be noted that we select two periods when the CO 2 concentration is the same in both the RU period and RD period. Despite having the same CO 2 concentration in the RU and RD periods, we nd that the amount of rainfall signi cantly increases in the EASM region from the RU period to the RD period (Fig. 2a). In contrast, it decreases in the western North Paci c, which will be discussed later.  (Fig. 1b), EASM rainfall also continues to increase for almost two decades after the CO 2 quadrupling peak. The climatological amount of rainfall in the PD simulation is 5.54 mm per day during summer; therefore, it increases by almost 20% during the quadrupling peak of the CO 2 concentration. Then, it gradually decreases until the stabilization period. In addition, the decreasing rate of EASM rainfall during the ramp-down period is smaller than its increasing rate during the ramp-up period. This results in the enhancement of summer rainfall in East Asia despite the same CO 2 concentration in the RU and RD periods (Fig. 2a).
The tropical Paci c SST response to the change in the CO 2 pathway To understand these asymmetric changes in the EASM rainfall, it is essential to examine the response of SST during summer with a change in the CO 2 pathway. In particular, the changes in the tropical SST in response to CO 2 forcing could account for a large portion of the regional rainfall pattern via atmospheric teleconnection as well as the response of the tropical rainfall pattern 13,14,[26][27][28] . It is found that the tropical Paci c SST response is characterized by an El Niño-like warming pattern during both the RU and RD periods (Figs. 3a,b). Note that the SST pattern is obtained from the deviation in the ensemble mean climatological SST from a PD simulation (Methods). Furthermore, an El Niño-like SST warming pattern is strengthened during the RD period, resulting in an asymmetrical response of the SST in the tropical Paci c between the RU period and the RD period despite having the same CO 2 concentrations (Fig. 3c), which is consistent with the results of previous studies based on ramp-up and ramp-down CO 2 experiments using different climate models 13,14,29 . These researchers inferred that both a reduction in ocean strati cation and a continuous weakening of the Walker circulation since the peak of the CO 2 concentration would serve to produce an enhanced El Niño-like warming pattern during the RD period. They further speculated that the ocean changes are the driving mechanism that then feed back into the changes in the Walker circulation. However, it is still necessary to understand the detailed processes leading to an enhanced El Niño-like warming during the RD period when a different climate model is used.  30 . The magnitude of the Walker circulation in a changing CO 2 pathway exhibits an asymmetrical response (Fig. 4a). While the intensity of the Walker circulation gradually weakens until the quadrupling peak of the CO 2 concentration, its intensity slowly recovers during the entire period of rampdown and it continues to recover until the end of the stabilization period. The slow recovery of the Walker circulation intensity during the RD period is associated with an enhanced El Niño-like SST warming pattern via atmosphere-ocean coupled processes 31 compared to that during the RU period.
To explain this phenomenon, we emphasize the reduced ocean thermal strati cation in the central-toeastern tropical Paci c where the climatological mean upwelling is the most dominant ( gure not shown) during the RD period. The reduced ocean thermal strati cation induces warming through the vertical advection of warm subsurface water by the climatological-mean upwelling in the central-to-eastern tropical Paci c. This results in an enhanced El Niño-like SST response as well as its associated slow recovery of Walker circulation via atmosphere-ocean coupled processes during the RD period. To examine this, we show the time series of ocean thermal strati cation, which is de ned as the difference between the mixed-layer (30-100m) and sub-thermocline layer (150-230m) ocean temperature in the central-toeastern tropical Paci c (5 S-5 N, 190 -250 E) (Fig. 4b). During the ramp-up period, both the mixed-layer and sub-thermocline layer ocean temperatures tend to warm; however, the mixed-layer temperature warms faster than that in the sub-thermocline layer. This results in a strengthening of the ocean thermal strati cation during the RU period. Meanwhile, after the CO 2 quadrupling peak, the mixed-layer temperature starts to cool, while the lower ocean temperature continues to warm, leading to a rapid decline in ocean strati cation. Therefore, the continuous ocean warming in the sub-thermocline layer after the CO 2 quadrupling peak leads to an asymmetrical response of ocean strati cation from the RU period to the RD period. This process contributes to an enhanced El Niño-like SST warming as well as its associated slow recovery of Walker circulation during the RD period.
In addition to the mechanism noted above, we suggest that the equatorward migration of the Intertropical Convergence Zone (ITCZ) leads to an enhanced El Niño-like SST warming pattern during the RD period. There is asymmetrical hemispheric warming in the Southern Hemisphere relative to the Northern Hemisphere from the RU period to the RD period ( Supplementary Fig. 1), which originates from the large heat capacity of the Southern Ocean. This inter-hemispheric asymmetrical warming pattern may cause an equatorward shift of ITCZ during the RD period compared to that during the RU period, because the ITCZ tends to migrate toward the warmer hemisphere 32, 33 (Fig. 5). Figure 5a clearly shows that the latitudinal position of ITCZ shifted more equatorward after the quadrupling peak of the CO 2 concentration, which is consistent with the changes in rainfall in the tropical Paci c basin between the RU period and the RD period (Figs. 5b,c). An equatorward shift of ITCZ causes a weakening of the trade winds and thus acts to favor an El Niño-like warming via reducing evaporative cooling and upwelling [33][34][35] , resulting in enhanced El Niño-like warming during the RD period.
Atmospheric circulation response and its impacts on rainfall in East Asia We argue that the asymmetrical response of rainfall in East Asia between the RU and the RD period (see Fig. 2a) is partly due to the enhanced El Niño-like SST pattern (Fig. 3c) through atmospheric teleconnection 28,[36][37][38] . To support this notion, we analyze the large-scale zonal and meridional overturning atmospheric circulation pattern.
There is an asymmetrical response of the zonal overturning atmospheric circulation averaged for 20 S-20 N between the RU and the RD period (Fig. 6a). There are distinct strong upward and downward motions in the eastern tropical Paci c and the western tropical Paci c, respectively, which is consistent with a weakening of Walker circulation during the RD period. This result indicates that the anomalous downward motion over the western North Paci c region contributes to the more suppressed rainfall anomalies over the western North Paci c region during the RD period compared with those during the RU period (see also Fig. 2a). Furthermore, these large-scale atmospheric vertical motions in the western tropical Paci c are responsible for the meridional overturning atmospheric circulation to East Asia 39,40 . Figure 6b displays the cross-sections of the differences in the zonally (110-145°E) averaged overturning circulation patterns from the tropics to East Asia between the RU and RD periods. The downward motions and low-level divergences over the western tropical Paci c led to strong updrafts and northward water vapor transports over 25 -35 N where the EASM rainfall is signi cantly enhanced during the RD period compared to that during the RU period. In other words, the large-scale atmospheric overturning circulation pattern, which is related to the enhanced El Niño-like SST pattern, primarily contributes to the enhanced (suppressed) rainfall anomalies over the EASM (western North Paci c) region between the RU and RD periods.
On the other hand, the meridional thermal gradients may also contribute to the asymmetrical response of summer rainfall in East Asia through the modulation of monsoon circulation [41][42][43] . It should be noted that the SST is warmer in the western North Paci c (15 -25 N, 120 -150 E) during the RD period than that during the RU period (see Fig. 3c). This SST warming pattern is induced by the downward motion related to the strengthening of the western North Paci c subtropical high ( Fig. 7a and see also Fig. 6b), which is due to the intensi ed El Niño-like warming pattern as mentioned above. Concurrently, atmospheric warming with a barotropic structure is prominent over the subtropical Paci c (Fig. 7b). This distinct warming leads to an increase in the meridional temperature gradient from the tropics to East Asia where the climatological baroclinicity is strong. Thus, there is strengthened zonal wind anomalies in the upper troposphere where the westerly jet axis is located via the thermal wind balance 44,45 (Fig. 7c). This acceleration of the jet may promote the convection to its south, forming the rain band in association with the ageostrophic secondary circulation 46,47 . It leads to anomalous rainfall over East Asia, contributing to increasing EASM rainfall.
We argue that the continued warming in the western North Paci c, which is associated with the intensi ed El Niño-like warming pattern, contributes to the asymmetrical response of EASM rainfall through the modulation of the atmospheric circulation pattern. It is also noteworthy that the El Niño-like warming is related to the enhanced (suppressed) rainfall anomalies over the EASM (western North Paci c) region during summer in the PD simulation ( Supplementary Fig. 2), supporting the notion noted above that the enhanced El Niño-like warming causes the asymmetrical response of rainfall in East Asia between the RU and RD periods.

Discussion
The idealized CO 2 forcing experiments are designed to provide an understanding of the regional hydrological response to direct atmospheric CO 2 removal. We found an asymmetrical response of the East Asian summer monsoon rainfall pattern with a change in the CO 2 pathway. The model results show that the rainfall over the EASM region gradually increased during the RU period. However, it slowly decreased following the CO 2 decrease during the RD period. This asymmetrical response is mainly due to the intensi ed El Niño-like warming pattern during the RD period. This SST warming pattern is largely due to the slow warming response of sub-thermocline ocean temperature in the central-to-eastern tropical Paci c, leading to contrasting ocean thermal strati cation from the RU period to the RD period. Furthermore, the migration of ITCZ due to the contrast in the asymmetrical hemispheric warming is also associated with the enhanced El Niño-like SST warming pattern during the RD period. Subsequently, anomalous large-scale overturning circulation patterns due to the intensi ed El Niño-like warming pattern lead to an anomalous upward (downward) motion over EASM (western North Paci c), leading to the enhanced (suppressed) rainfall anomaly. In addition, continued warming over the western North Paci c strengthens the meridional temperature gradient over East Asia where the climatological baroclinicity is strong. This leads to an accelerated westerly jet, which leads to increased rainfall in East Asia by altering the ageostrophic secondary circulation.
It is noteworthy that the EASM rainfall during the CO 2 stabilization period remains higher than that during the PD climate state when the CO 2 concentration is the same (Fig. 2b). This irreversible feature of EASM rainfall may be due to the continued warming over the eastern tropical Paci c and western North Paci c region during the CO 2 stabilization periods ( Supplementary Fig. 3), which is associated with the large thermal inertia of the ocean. These results suggest that the asymmetrical response and irreversible change in EASM rainfall are closely linked through oceanic memory. We rst conduct the present-day climate simulation with a xed CO 2 concentration (367 ppm) for the simulation period of 900 years. This reference simulation is referred to as a PD simulation. Then, the changing CO 2 pathway with 28 ensemble members is employed. In this study, the atmospheric CO 2 concentration with a 1% increase per year peaks with its quadrupling (i.e., 1,468 ppm) (referred to as the ramp-up period), followed by a decrease of 1% per year until the CO 2 concentration in the present-day climate (referred to as the ramp-down period) as shown in Fig. 1a. The CO 2 stabilization with a ∘ ∘ ∘ ∘ ∘ concentration of 367 ppm is continued for 120 years. The 28 ensemble members are identical except for the oceanic initial conditions, which are taken arbitrarily from a PD simulation.

Methods
Anomaly calculation We analyze the ensemble mean atmospheric and oceanic variables simulated from the 28 ensemble members to emphasize the role of atmospheric CO 2 forcing by excluding the role of internal variability in the climate system. All of the analyses are based on deviation from the PD climatology from the ensemble mean variables. To test the statistical signi cance of the difference in the atmospheric and oceanic elds between the ramp-up and -down periods, we use a two-sided Student's ttest.

Declarations
Code availability All the NCL and Fortran 90 codes used to generate the results of this study are available from the authors upon request.    with the climatological rainfall averaged for 90°W-180° (black), which was obtained from the PD simulation.