Global Cooling Hiatus Driven by an AMOC Overshoot in a Carbon Dioxide Removal Scenario

The reversibility of global mean surface temperature was examined by a transient CO2 reversibility experiment using an Earth system model. The results showed that after CO2 ramp‐up toward CO2 quadrupling and ramp‐down returned to the present‐day level, the global mean surface temperature kept decreasing but stopped to change for ∼40 years in the early net‐zero CO2 emission period. This period, referred to a cooling hiatus, resulted from a compensation between Southern Hemisphere cooling and Northern Hemisphere warming. The Northern Hemisphere warming was centered over the North Atlantic. This localized warming was caused by an excessive heat advection by a delayed and surpassed Atlantic Meridional Overturning Circulation (AMOC) to CO2 forcing. During the progression of CO2 change, the meridional salinity gradient between subtropic and subpolar regions was enhanced, and the oceanic stratification in subpolar North Atlantic was reduced due to accumulated heat and reduced vertical salt import in the deeper ocean. As AMOC started to recover, consequently, the enhanced salt advection feedback and the relaxed buoyant force resulted in AMOC overshoot.

• In CO 2 removal experiment, change of global mean surface temperature was halted for 40 years during the early net-zero CO 2 emission period • Cooling hiatus was driven by an excessive heat advection by a delayed and surpassed Atlantic Meridional Overturning Circulation (AMOC) to CO 2 forcing • AMOC overshoot is due to the salt advection feedback, amplified by enhanced salinity gradient and reduced oceanic stratification

Supporting Information:
Supporting Information may be found in the online version of this article.
tipping of climate change (van Vuuren et al., 2013). In this regard, the prospective climate state after the stabilization of the CO 2 concentration and associated driving mechanisms are of interest to the scientific community, and have socioeconomic merit. Obviously, the answer on how the climate state will be in the end is directly linked to the reversibility or irreversibility of a climate system. Here, climate irreversibility means that the system cannot be restored to its initial state or can be restored only on a long timescale (Boucher et al., 2012). The climate system is compose of various climate elements, and each climate element must have different reversibility depending on its intrinsic response time scale, for example, spanning from hours-to-weeks of atmosphere to 100-10,000 years of ice sheet (e.g., Ruddiman, 2013), possibility for tipping (Lenton et al., 2008), and the time scale of forcing (e.g., An et al., 2021).
One way to measure the reversibility and/or irreversibility of a climate system is to simulate a climate system model under the following scenarios: overshoot (Huntingford & Lowe, 2007;Nusbaumer & Matsumoto, 2008), peak-and-decline (Schaeffer et al., 2015), CO 2 reversibility (Andrews & Ringer, 2014), rampup and -down (Armour et al., 2011;Boucher et al., 2012;Wu et al., 2011), or zero emissions commitment (MacDougall et al., 2020;Matthews & Weaver, 2010). These scenarios are quite similar to each other. For example, one scenario requires that CO 2 atmospheric concentration increases by a certain rate (e.g., 1% per year) from an initial level to a maximum level, which is two, three, or four times the initial level. Subsequently, the CO 2 concentration either decreases after keeping its CO 2 level constant for a certain period or immediately decreases to the initial level without keeping the CO 2 level constant at any time. One may also allow free evolution of atmospheric CO 2 concentration under net-zero emission after increasing of CO 2 concentration to a certain level (MacDougall et al., 2020). States from further model integration with a fixed initial level of CO 2 concentration, that is, recovering with net-zero emission toward the initial climate condition, can be compared with the initial state to measure the irreversibility of a climate system.
Irreversibility of a variety of physical phenomena was examined under a ramp-up and -down scenario. For example, it was demonstrated that both low-level clouds and ocean stratification in the Southern Ocean exhibit hysteresis, implying a possibility of irreversibility (Boucher et al., 2012). It has been also argued that the irreversibility or multiple ice-cover states over the full range of simulated sea ice conditions and an annually ice-free Arctic Ocean hardly existed (Armour et al., 2011). Another study found significant hysteresis behavior in the precipitation response due to previously accumulated heat in the ocean (Wu et al., 2011). The hysteresis behavior of Atlantic Meridional Overturning Circulation (AMOC) in these kinds of experiments is highly interesting (Haskins et al., 2019;Jackson et al., 2014;Sgubin et al., 2015) because AMOC is one of the major tipping elements that can abruptly shift to a qualitatively different state as the climate forcings change (Lenton, 2011;Swingedouw et al., 2020).
Based on multimodel experiments under various types of ramp-up and ramp-down scenarios, it was shown that AMOC became anomalously stronger during the early recovering period keeping a constant CO 2 concentration, referred to as "overshoot" (Jackson et al., 2014;Wu et al., 2011). The study argued that the magnitude of this overshoot was related to the accumulated salinity in the subtropical Atlantic Ocean. The hydrological cycle change during the previous period of high CO 2 levels, partly through hydrological cycle change, and so enhancing a salt advection feedback during the recovering period. A hysteresis-like response of AMOC was also reported by a subsequent study (Sgubin et al., 2015), in which climate models yielded either stronger or weaker AMOC in the recovering phase compared to their initial states.
Asymmetry between the weakening and recovery of AMOC in its hysteresis loop is a dynamic nature of the AMOC system (An et al., 2021;Haskins et al., 2019). Therefore, the asymmetric changes in AMOC are expected under the CO 2 ramp-up and -down scenario. In this study, we investigated the changes in AMOC and its impact on global climate by performing 28 ensemble experiments of CESM 1.2 (Yu & Pritchard, 2019) under CO 2 ramp-up and -down scenario (Methods). The main focus has been given to the following aspects: (a) How does the North Atlantic Ocean respond to the CO 2 ramp-up and -down scenario? (b) What drives the sea surface temperature (SST) change in the subarctic Atlantic Ocean? (c) What are the major processes responsible for changes in SST in subarctic Atlantic? (d) What is the driving mechanism for the excessive response of SST to the changing CO 2 forcing?

Model Configuration
In this study, we used the Community Earth System Model version 1.2 (CESM1.2) (Hurrell et al., 2013), which is composed of the atmosphere (The Community Atmospheric Model version 5, CAM5), ocean (The Parallel Ocean Program version 2, POP2), sea ice (The Community Ice Code version 4, CICE4), and land models (Community Land Model version 4, CLM4). The atmospheric model had a horizontal resolution of ∼1° × 1° and 30 vertical levels (Neale et al., 2012). The ocean model had 60 vertical levels, with a longitudinal resolution of 1° and a latitudinal resolution of 0.3° near the equator that gradually increases to 0.5° near the poles (Smith et al., 2010). The land model comprised the carbon-nitrogen cycle (Lawrence et al., 2011).

Experimental Design
Two idealized CO 2 pathways were applied ( Figure 1a): the constant and varying scenarios. These scenarios were similar to the protocol of the carbon cycle reversibility experiment (CDR-reversibility) except for the initial CO 2 level (Keller et al., 2018). The constant CO 2 scenario was simulated for 900 years with 367 ppm as the present-day CO 2 content, referred to as the present-day period (PD). Subsequently, the varying CO 2 scenario with 28 ensemble members was applied, in which the CO 2 concentration increased 1% per year for 140 years until it quadrupled (i.e., 1,468 ppm; ramp-up period), immediately followed by a symmetric decrease (∼1% per year) for 140 years until the CO 2 concentration reached the initial level (367 ppm; rampdown period). Subsequently, the constant CO 2 (367 ppm) scenario, that is, net-zero emission, was simulated for 220 years (restoring period). The 28 ensemble members were identical except the atmospheric and oceanic initial conditions with different phases of Pacific Decadal Oscillation and Atlantic Multidecadal Oscillation at the PD period. The 28 ensemble members are likely enough to identify a climate signal driven the CO 2 forcing.

Mixed Layer Temperature and Salinity Equations
The upper-ocean temperature budget was computed as follows (Chemke et al., 2020;Kim et al., 2006): where the square bracket indicates the depth average within the mixed layer. In Equation 1, T represents the mixed layer temperature and Q net is the net air-sea heat flux (i.e., radiative shortwave [SW] and longwave [LW] fluxes, sensible , latent [LE], and sea ice melting heat flux [MELTH]). Here, the positive fluxes denote the heat transport from the atmosphere to the ocean. However, all results in this study are shown as anomalies, and thus the positive (negative) anomaly in heat flux indicates anomalous warming (cooling) of ocean. ρ is the reference density (1,027 kg m −3 ), C p is the specific heat of seawater at a constant pressure (3985 J kg −1 K −1 ), and h is a mixed-layer depth that varies in time. Here, the mixed-layer depth is defined as the shallowest depth where the local, interpolated buoyancy gradient matches the maximum buoyancy gradient between the surface and any discrete depth within that water column (Large et al., 1997). u and w are the horizontal and vertical velocities, respectively. The first three terms on the right-hand side in Equation 1 are net surface heat flux, horizontal temperature advection, and vertical advection, respectively. The "ML mixing" refers to a mixing process in the mixed layer. The subsurface processes represent entrainment across the lateral induction and vertical diffusion. In this study, the last two terms of Equation 1 were calculated as residuals.
The salinity budget analysis was performed using the following simplified equation (Kim et al., 2006;Qu et al., 2011): This equation is the same as Equation 1, however, a mixed-layer salinity S was applied. The coefficient Q A represents surface salinity flux, which includes evaporation, precipitation, runoff, ice runoff, and ice melting fluxes. The last two terms of Equation 2 are residuals.

Cold Blob, Warm Blob, and Cooling Hiatus
During the CO 2 ramp-up phase, it was found that the global mean surface temperature (GMST) gradually increases up to 5°C compared to the GMST of PD. During the CO 2 ramp-down phase, GMST gradually decreases to ∼1.0°C above the PD GMST ( Figure 1a). After achieving a constant CO 2 level or net-zero CO 2 emission, the GMST slowly approaches toward the PD level, although it never reaches the exact PD level AN ET AL.
10.1029/2021EF002165 4 of 12 ( Figures 1a and 1b). Such irreversible or long delays in returning response was also reported in previous studies (Boucher et al., 2012;Chadwick et al., 2013;Ehlert & Zickfeld, 2018), and in particular, the long-delayed response is possibly due to the thermal inertia of the Earth climate system, especially associated with the heat uptake of the ocean. Although the irreversibility of GMST might be temporary, the ensemble mean temperature difference remain significant during the last 200 years of the simulations. The warming episode during the CO 2 ramp-up phase was not homogeneous over the globe, and there were regional differences as shown in the tendency of ensemble-mean annual-mean surface temperature (EMST) at 2°C of GMST that occurred around the year 2067 ( Figure 1c). Here, the tendency is defined as difference in EMST between 30-year average (2067-2096) after and 30-year average (2037-2066) before recording 2°C of GMST. The regional features of surface temperature changes were quite similar to the observed trends of temperature change in the 20th century ( Figure 2.21 of IPCC AR5 WG1) such as the greatest warming over the Arctic (so-called "Arctic Amplification"), stronger warming over high latitude continents, and weaker warming over the sea, especially in southern tropical oceans. Furthermore, an extremely weak warming tendency was observed over the northern subpolar Atlantic Ocean, similar to the currently observed trend (e.g., Chemke et al., 2020;Keil et al., 2020;Liu et al., 2015). This cold blob or warm hole has been suggested to be driven by the reduced oceanic heat transport associated with a slowing AMOC (Chemke et al., 2020;Kim & An, 2013;Liu et al., 2020;Rahmstorf et al., 2015), the enhanced export of oceanic heat to higher latitudes by subpolar gyre, and cooling by shortwave-cloud feedback (Keil et al., 2020). For example, the freshening over the North Atlantic deep water formation region due to reduced evaporation associated with CO 2 increases disrupts AMOC (Kim & An, 2013), and the increase in low-level clouds due to ocean surface cooling blocks the incoming shortwave radiation (Keil et al., 2020). This cold blob was later replaced by a warm blob during the CO 2 ramp-down phase as shown in the tendency of EMST at 2°C of GMST around the year 2246 ( Figure 1d). Again, the tendency is defined as difference in EMST between 30-year average (2246-2275) after and 30-year average (2216-2245) before recording 2°C of GMST. Specifically, this warming trend (warm blob or cooling hole) generally appears over the North Atlantic Ocean, with its maximum extent being located at the northeast direction of the maximum warm hole during the ramp-up period (Figure 1d). This indicates that a similar but opposite physical process as a cold blob might occur for a warm blob as well.
During the early recovery period (2280-2320), the warm blob shown in Figure 1d was further enhanced, especially over the Labrador and Irminger Seas (Figure 1e), indicating that the enhanced heat import into the subpolar gyre region where AMOC is connected. Here, the tendency of EMST during this early recovery period (Figure 1e) is defined as the difference between the average of (2301-2320) and that of (2281-2300). During this period, the global-averaged net radiative flux at the top of atmosphere was negative ( Figure S1), indicating that the ocean is likely a heat source (Andrews & Ringer, 2014). Moreover, a moderately warming trend occurs over the Arctic Ocean and North Pacific, and a weakly warming trend widely covers all Northern Hemisphere continents, except at the tropics. The GMST of the Northern Hemisphere also displayed a warming trend between 2280 and 2320, followed by a cooling trend after 2320. The cooling trend was observed over the entire Southern Hemisphere and tropics during the same period (Figure 1e). The strongest cooling was observed over the ocean surrounding Antarctica. The hemispheric contrast in surface temperature results in no change in GMST for ∼40 years (2280-2320), that is, a cooling hiatus, through compensation in opposite temperature changes between Northern and Southern hemispheres. In the following section, we mainly focus on this peculiar change in the surface temperature change.

Surface Heat Budget Over the North Atlantic Subpolar Ocean
To assess the cause of the North Atlantic subpolar ocean (ASPO) temperature change, heat budget analysis of the ocean mixed layer (OML) temperature was conducted over the ASPO (45°-70°N and 50°-10°W; boxed in Figure 1c; Figures 2b-2d). As shown in Figure 2a, the ASPO temperature gradually increased up to ∼2°C compared to PD during the ramp-up period, however, the rate of increase was relatively weaker compared to that at other same or higher latitude regions (Figure 1c). During the ramp-down period, ASPO temperature sharply decreased by ∼3°C compared to PD, and subsequently started to increase before the CO 2 concentration reached the PD level. During the restoring phase, the ASPO temperature kept increasing and reached its maximum, representing a 2°C warmer climate in 2320, which is almost equal to the temperature at maximum CO 2 concentration. This rapid warming from the 2220s-2320s was much stronger than that in the CO 2 ramp-up warming period (2000-2140), although the CO 2 concentration decreased and later reached a constant level similar to PD. After the peak, the ASPO temperature slowly decreased (Figure 2b) although warmer conditions compared to PD period were maintained for more than 100 years.
The increasing ASPO temperature during the ramp-up phase was mainly due to warming by surface heat flux, whereas cooling by the zonal and meridional oceanic heat advection along with the residual processes partly compensated the warming, causing a weak warming in ASPO (Kim & An, 2013) (Figure 2c). The residual is the sum of all other processes except net surface heat fluxes, and horizontal and vertical advection such as three-dimensional thermal diffusion and thermal advection by subgrid scale eddies. During the ramp-down phase, the surface heat flux kept increasing until the year 2200 despite a decrease in CO 2 concentration because although the evaporative cooling gets reduced as the ocean surface cools (Figures 2a  and 2d), cooling by other processes overcompensates this. Therefore, ASPO temperature declined until approximately the year 2240, when its minimum temperature was observed. Subsequently, the zonal heat ad-AN ET AL.
10.1029/2021EF002165 6 of 12 vection started to increase, followed by an increase in meridional heat advection. The net surface heat flux decreased as the ocean surface warmed up, and the difference between ocean temperature and surface air temperature decreased (orange line in Figure 2a). The residual also increased during the ramp-down period, and became positive during the recovery period. The negative tendency of ASPO temperature due to the residual was mainly due to vertical diffusion, as oceanic stratification increased during the ramp-down period (see Figure 4f). The recovery toward the positive tendency was related to the reduced oceanic stratification as the AMOC recovered. The positive ASPO temperature during the cooling hiatus period and almost zero tendency in the remaining recovery period resulted in an ASPO temperature peak of 2°C at the end of the cooling hiatus period. The positive ASPO temperature tendency is likely to be a combination of the positive tendency due to surface heat flux and zonal heat advection, and the reduced negative tendency is probably associated with the meridional heat advection. Here, it is important to note that the positive tendency due to surface heat flux was a byproduct (or passive feedback) resulting from horizontal heat advection.
It was found that the net heat flux at the surface mainly follows the latent heat flux (LE), and both sensible heat flux (SE), and net long-wave radiation (LW) contribute to the surface heat flux in a similar way, although their contribution was lower than LE. The net short-wave radiation (SW) exhibited almost a mirror image of LW with a smaller magnitude. Climatologically, the evaporative cooling (LE) occurs over ASPO due to the colder air originating from the continent over the warmer ocean surface, especially during boreal winter. During the ramp-up period, the faster warming of near surface air than the ASPO surface (i.e., decreased temperature difference between atmosphere and ocean in Figure 2a) resulted in the reduction of evaporative cooling (Figure 2d). The reduced air-sea temperature difference was maintained during the ramp-down period because of the faster cooling of ASPO than that of the atmosphere, however, the magnitude of the difference gradually decreased because of the increase in ocean surface temperature until the end of the ramp-down phase (Figure 2a). Furthermore, the ocean mixed layer depth in ASPO reduced from 150 m in PD to 60 m during the ramp-down period, which increased the efficiency of ocean surface temperature and LE change (not shown). LE and SE were determined by the air-sea interaction at the surface. However, they relied more on the change in ocean surface temperature, driven by the oceanic heat advection associated with AMOC and ocean circulation such as subpolar gyre (Keil et al., 2020). Therefore, the warming tendency by surface heat flux (especially LE and SE) is likely associated with a passive feedback due to temperature changes in the upper ASPO by horizontal heat advection.

Role of Ocean Circulations in Cooling Hiatus
The horizontal oceanic heat transport over ASPO is directly linked to AMOC and subpolar gyre circulation. As shown in Figure 3a, the AMOC decreased during the ramp-up period. It decreased further until the year 2180 in the middle of the ramp-down period, with a recorded strength of ∼10 Sv (Sv = 1 million m 3 s −1 ), which is only half of PD. The strongest cooling tendency occurred at 2200 (Figures 2a and 2b), and thus, the temperature change lagged AMOC by ∼20 years (see Figure S2). After the year 2180, AMOC increased quickly up to ∼25 Sv until the cooling hiatus period, and subsequently, it decreased slowly. Therefore, the difference between the maximum and minimum AMOC strength was ∼15 Sv. AMOC change was almost in-phase with salinity change over ASPO, and was slightly ahead of the density change. Therefore, the local density change, especially due to salinity change over the oceanic convection region within ASPO, is strictly linked to AMOC. In this respect, we can expect that the feedback processes associated with surface hydrology change and oceanic salt advection control AMOC change.
A negative tendency of salinity started to appear right after the CO 2 increase, and its maximum negative tendency was marked at the highest CO 2 concentration (Figure 3b). Around 2200, the tendency changed its sign from negative to positive, which aligned with the minimum salinity over ASPO. The maximum positive salinity occurred during the cooling hiatus period, and the tendency sign changed again at the end of the cooling hiatus. Thus, the maximum salinity was observed at the end of the cooling hiatus. Afterward, the salinity tendency converged to zero. The negative salinity trend during the ramp-up was driven primarily by zonal salinity advection, and secondly by the increasing surface freshwater flux (i.e., mainly by the reduced evaporation, Liu et al., 2020;Ma et al., 2020) and meridional salinity advection (Figures 3c and 3d). During the ramp-down period, the zonal salinity advection started to increase, and thus, the phase change in the zonal salinity advection was similar to that of the total salinity tendency. While the meridional sa-linity advection kept decreasing by the end of the ramp-down phase, the negative salinity tendency was maintained by the reduced evaporation. The residual, which indicates vertical salinity diffusion, produced a substantial positive salinity trend during the ramp-up and -down periods. This is because the enhanced vertical stratification in the upper ocean prevented the diffusion of low salinity water of the subsurface to the surface. Note that vertical salinity advection was very weak. The change in zonal salinity advection was likely related to the change in the subpolar gyre. The surface wind stress curl over ASPO was reduced during the ramp-up phase, thereby suppressing the subpolar gyre circulation. The stress increased during the ramp-down phase, leading to enhanced subpolar gyre circulation (not shown). Therefore, the zonal salinity advection change was associated with the change in surface wind stress. Although the meridional salinity advection did not strictly follow changes in AMOC, it can be suggested that the decreasing trend in salinity induced by meridional salinity advection was related to AMOC. Interestingly the zonal salinity advection changed prior to changes in meridional advection and residual terms. All salinity budget terms changed AN ET AL. 10.1029/2021EF002165 8 of 12 their tendency during the hiatus period (Figure 3c), and subsequently, the total salinity tendency changed its sign from positive to negative (Figure 3b). Consequently, the salinity as well as AMOC were maximized at the end of the cooling hiatus period.
Positive freshwater flux, which induced a negative salinity trend was mainly driven by reduced surface evaporation (Figure 3d). The reduced sea ice melting, whose rate increased during the ramp-up phase and decreased during the ramp-down phase, and the reduced precipitation especially during the ramp-down, induced a negative freshwater flux leading to a positive salinity trend (see also Figure S3). As previously mentioned, each of the surface freshwater flux terms changed its tendency during the hiatus period ( Figure 3d).
AN ET AL.

Overshoot of AMOC
As the AMOC started to recover, anomalously higher salinity in the subtropical region was advected to the subpolar region ( Figure 4b) (Haskins et al., 2019;Wu et al., 2011). During the slowing down period of AMOC, the suppressed salt exported from the subtropics (southward current) led to an increase in salinity in the subtropics (Figures 4b and 4c). The increase in subtropical salinity was also driven by the reduced atmospheric freshwater flux (i.e., precipitation minus evaporation; e.g., Figure S3) during the ramp-up to -down period, which is a general response induced by the CO 2 driven global warming (Allen & Ingram, 2002;Christensen et al., 2013). Therefore, as AMOC recovered, the enhanced salt advection feedback associated with stronger meridional salinity gradient and the recovered northward current induced stronger northward salt transport, and then the zonal advection of the salt by subpolar gyre into the region for deep convection promoted the overshoot in AMOC.
The surface warming of the subpolar region during the ramp-up period was slowly diffused to the deeper ocean during the ramp-down period (Figures 4a and 4d), and the lower salty surface water during the rampdown period was also slowly diffused to the deeper ocean ( Figure 4e). Furthermore, as a fresh buoyant cap formed at the ocean mixed layer during the ramp-down period (Figures 4e and 4f), the subsurface and deep ocean water became isolated from the surface fluxes, and thus remained warm ( Figure 4d) (Haskins et al., 2019). These effects led to a less stratified density profile and caused instability in the water column, especially around the end of the ramp-down. This could lead to an overshoot in AMOC by promoting convection during the cooling hiatus period (Figure 4f).

Conclusions
In this study, the transient CO 2 reversibility experiment, after returning to the initial CO 2 level, showed a pause in the GMST cooling trend for ∼40 years, that is, global cooling hiatus. This cooling hiatus was caused by a combination of the oceanic zonal heat advection associated with subpolar gyre (Keil et al., 2020), meridional heat advection by the overshoot of AMOC, and the passive feedback by the latent heat flux. The AMOC overshoot was also simulated by other climate models, and we found that the overshoot intensity of AMOC depends on how much salinity gradient between the North Atlantic deep water formation region and subtropical Atlantic builds up during CO 2 forcing period (Sgubin et al., 2015), as well as how much the oceanic stratification is developed as shown in this study. During the progression of CO 2 change, the salt advection feedback due to the increased meridional salinity gradient between subtropic and subpolar regions was enhanced, and the oceanic stratification due to accumulated heat and reduced vertical salt import in the deeper ocean was reduced. Consequently, as AMOC started to recover, these two effects drove AMOC overshoot.
This study indicates that GMST will not simply decrease after the net CO 2 emission becomes zero. A delayed and excessive oceanic circulation response including AMOC overshoot to CO 2 forcing might lead to a warm environment for several decades in the early mitigation phase including a global hydrological change ( Figure S3). Furthermore, the strong climate variability, especially over the Northern Hemisphere, is also expected in this period (see the large ensemble spread in Figure 1b).
One of the main oceanic circulation inside ASPO is the subpolar gyre (SPG), which plays an important role in transporting heat and salt (Moreno-Chamarro et al., 2017), especially into the Labrador Sea, where a core of the downwelling branch for the AMOC is located. For example, the heat and salt transport by Irminger current and the freshwater export from Baffin Bay to Labrador Sea mostly influence the strength of AMOC. However, we did not consider the internal processes of ASPO, while taking the areal average over ASPO. Therefore, the understanding on an interaction between AMOC and SPG will provide a clue on the detailed structure of SST change in North Atlantic Ocean, which will be our next project.