Climate Sensitivity and Feedback of a New Coupled Model (K-ACE) to Idealized CO 2 Forcing

: Climate sensitivity and feedback processes are important for understanding Earth’s system response to increased CO 2 concentration in the atmosphere. Many modelling groups that contribute to Coupled Model Intercomparison Project phase 6 (CMIP6) have reported a larger equilibrium climate sensitivity (ECS) with their models compared to CMIP5 models. This consistent result is also found in the Korea Meteorological Administration Advanced Community Earth System model (K-ACE). Idealized climate simulation is conducted as an entry card for CMIP6 to understand Earth’s system response in new coupled models and compared to CMIP5 models. The ECS in the K-ACE is 4.83 K, which is higher than the range (2.1–4.7 K) of CMIP5 models in sensitivity to CO 2 change and higher bound (1.8–5.6 K) of CMIP6 models. The radiative feedback consists of clear-sky and cloud radiative feedback. Clear-sky feedback of K-ACE is similar to CMIP5 models whereas cloud feedback of K-ACE is more positive. The result is attributable for strong positive shortwave cloud radiative e ﬀ ect (CRE) feedback associated with reduced low-level cloud cover at mid latitude in both hemispheres. Despite the cancellations in strong negative long wave CRE feedback with the changes in high-level clouds in the tropics, shortwave CRE has a dominant e ﬀ ect in net CRE. Detailed understanding of cloud feedback and cloud properties needs further study.


Introduction
Climate sensitivity (ECS; equilibrium climate sensitivity) is typically defined as the average global temperature rise following a doubling of CO 2 concentration in the atmosphere compared to pre-industrial levels. It results from a variety of feedback processes (e.g., Planck feedback, water vapor feedback, cloud feedback, and radiative feedback) within the global climate [1][2][3][4][5][6]. Thus, it is essential to identify and understand the specificity of climate system warming in response to increased CO 2 levels in climate projection.
The ECS is important for effective carbon reduction polices to achieve a specific warming target [7]. For example, pre-industrial CO 2 is approximately 260 parts per million (ppm), and hence, doubling would be approximately 520 ppm. Current levels of atmospheric CO 2 have exceeded 410 ppm, with the 520-ppm threshold expected to be surpassed in the next 50-100 years based on predicted future greenhouse gas emissions. Additionally, Matthews et al. [8] found that global warming depends mainly

Model Experiment and Methodology
The K-ACE is a coupled climate model (Atmosphere-Ocean-sea Ice-Land; AOIL), and detailed component models (the number of physical and biogeochemical processes included) with coupling approaches are described in Lee et al. [34]. Hence, limited details are given here. The Unified Model (UM) in the Global Atmosphere 7.1, the latest configuration [35], is the atmospheric component of the K-ACE. A new dynamics scheme, the Even Newer Dynamics for General atmospheric modeling of the environment (ENDGame) [36], is implemented for faster and more efficient model integrations. Atmospheric radiative transfer is calculated by Suite of Community Radiative Transfer codes based on Edwards and Slingo (SOCRATES) [37], and tropospheric aerosols are calculated using the GLObal Model of Aerosol Processes (GLOMAP) [38,39] that considers the number concentration, size distribution, composition, and optical properties of aerosols based on the aerosol microphysics and chemistry. The ocean component is the Modular Ocean Model of GFDL (MOM) [40] and the Sea Ice model of Los Alamos (CICE) [41] is used for the sea-ice component. These components are coupled using the OASIS3-MCT coupler [42,43]. The horizontal resolution is N96 (~135 km) with a regular latitude-longitude grid in the atmosphere and a tri-polar grid in the ocean. Additionally, the vertical resolution is 85 levels (L85) in the atmosphere and 50 levels (L50) in the ocean. There are increased vertical levels in the atmosphere compared to a previous version of UM (38 levels). Moreover, there are many changes to the atmospheric physics; however, the new mixed-phase cloud scheme (PC2) is most significant [44,45], which uses three prognostic variables for water mixing ratio (water vapor, liquid, and ice) and cloud fraction (liquid, ice, and mixed-phase). Compared to the previous cloud scheme [46], the PC2 cloud scheme has a direct physical link between condensate, cloud fraction, and the physical processes that lead to their production. Additionally, it does not contain a separate, diagnostic large-scale cloud scheme [44]. Thus, the radiative effects of cloud related to convection are represented in the large-scale fields. Wilson et al. [45] reports that this approach offers the advantage of a more physically realistic cloud process. Overall, the PC2 cloud scheme leads that the ice cloud fraction extends higher (decrease low-level cloud) and high cloud cover is increased in tropics and high latitudes [45].
The 23 members of CMIP5 models and 18 members of CMIP6 models are used in this study (available members from the Earth System Federation Grid (ESGF) nodes at the time of writing). In CMIP6, as in earlier CMIP phases, the pre-industrial control (hereafter referred to as the pi-Control) simulation is an attempt to produce a stable quasi-equilibrium for beginning state of CMIP abrupt 4 × CO 2 experiment (hereafter referred to as the abrupt experiment). Additionally, this experiment is now included in the standard Diagnostic, Evaluation, and Characterization of Klima (DECK) experiments, which is a requirement for participation in the CMIP6 [47]. Modeling groups routinely calculate climate sensitivity for each new model version. Considering these points, the ECS for K-ACE is calculated from the CMIP abrupt 4 × CO 2 experiment (hereafter referred to as the abrupt experiment), and the CO 2 concentration (1136.8 ppm) is abruptly quadrupled from the global annual mean from the year 1850 (284.2 ppm). A useful method to calculate ECS via abrupt experiment is proposed by Gregory et al. [48], which is estimated by comparing the response of the top of atmosphere (TOA) radiative flux and surface air temperature (ECS value as one-half of the x-intercept and the total climate feedback parameter as the slope of the regression). This method has been widely used to provide ECS of climate models [25][26][27]30]. Shortwave (SW) and Longwave (LW) feedback parameters are calculated using a similar concept; however, the TOA radiative fluxes are applied anomalies instead of total value. Figure 1 shows the global annual mean surface air temperature changes from the abrupt experiments. The simulated period is 150 years, of which we regard the first 20 years as the transient part and the remaining 130 years as the equilibrium state. During the simulated period, the surface temperature does not stabilize. Compared to the 23 models from CMIP5, the temperature changes in K-ACE warms more in response to increased CO 2 (5-95% confidence levels). This is similar result with other CMIP6 models (HadGEM3-GC3.1-LL and UKESM1 [22], CNRM-CM6 [20], CESM2 [23], EC-Earth [21], and E3SMv1 [24]).

Climate Sensitivity to Idealized CO 2 Change
Atmosphere 2020, 11, x FOR PEER REVIEW 3 of 13 represented in the large-scale fields. Wilson et al. [45] reports that this approach offers the advantage of a more physically realistic cloud process. Overall, the PC2 cloud scheme leads that the ice cloud fraction extends higher (decrease low-level cloud) and high cloud cover is increased in tropics and high latitudes [45]. The 23 members of CMIP5 models and 18 members of CMIP6 models are used in this study (available members from the Earth System Federation Grid (ESGF) nodes at the time of writing). In CMIP6, as in earlier CMIP phases, the pre-industrial control (hereafter referred to as the pi-Control) simulation is an attempt to produce a stable quasi-equilibrium for beginning state of CMIP abrupt 4×CO2 experiment (hereafter referred to as the abrupt experiment). Additionally, this experiment is now included in the standard Diagnostic, Evaluation, and Characterization of Klima (DECK) experiments, which is a requirement for participation in the CMIP6 [47]. Modeling groups routinely calculate climate sensitivity for each new model version. Considering these points, the ECS for K-ACE is calculated from the CMIP abrupt 4 × CO2 experiment (hereafter referred to as the abrupt experiment), and the CO2 concentration (1136.8 ppm) is abruptly quadrupled from the global annual mean from the year 1850 (284.2 ppm). A useful method to calculate ECS via abrupt experiment is proposed by Gregory et al. [48], which is estimated by comparing the response of the top of atmosphere (TOA) radiative flux and surface air temperature (ECS value as one-half of the x-intercept and the total climate feedback parameter as the slope of the regression). This method has been widely used to provide ECS of climate models [25][26][27]30]. Shortwave (SW) and Longwave (LW) feedback parameters are calculated using a similar concept; however, the TOA radiative fluxes are applied anomalies instead of total value. Figure 1 shows the global annual mean surface air temperature changes from the abrupt experiments. The simulated period is 150 years, of which we regard the first 20 years as the transient part and the remaining 130 years as the equilibrium state. During the simulated period, the surface temperature does not stabilize. Compared to the 23 models from CMIP5, the temperature changes in K-ACE warms more in response to increased CO2 (5-95% confidence levels). This is similar result with other CMIP6 models (HadGEM3-GC3.1-LL and UKESM1 [22], CNRM-CM6 [20], CESM2 [23], EC-Earth [21], and E3SMv1 [24]).   Figure 2a shows the global annual mean change in net TOA radiative flux as a function of global mean surface temperature change. This regression has been expressed by Gregory et al. [48] for a simple linear relationship between radiative forcing and climate response. This approach considers that global surface temperature is unchanged. Therefore, climate feedback processes have an impact on the TOA radiation balance, which is included in our forcing estimate. Based on the Gregory-style regression, net climate feedback λ (Figure 2a), the slope of fitting line, is approximately −0.69 Wm −2 K −1 , and ECS of K-ACE is 4.83 K. Figure 2b shows the ECS of CMIP5 and CMIP6 models. It is observed that the ECS of CMIP6 with values spanning 1.8-5.6 K is larger than the range of CMIP5 (2.1-4.7 K). Recently, many studies reported that the ECS from several CMIP6 models has increased substantially [20][21][22][23][24]. The ECS of K-ACE is higher than that of CMIP5 and near high bound of CMIP6 models (Figure 2b).

Climate Sensitivity to Idealized CO2 Change
Atmosphere 2020, 11, x FOR PEER REVIEW 4 of 13 Figure 2a shows the global annual mean change in net TOA radiative flux as a function of global mean surface temperature change. This regression has been expressed by Gregory et al. [48] for a simple linear relationship between radiative forcing and climate response. This approach considers that global surface temperature is unchanged. Therefore, climate feedback processes have an impact on the TOA radiation balance, which is included in our forcing estimate. Based on the Gregory-style regression, net climate feedback λ (Figure 2a), the slope of fitting line, is approximately −0.69 Wm −2 K −1 , and ECS of K-ACE is 4.83 K. Figure 2b shows the ECS of CMIP5 and CMIP6 models. It is observed that the ECS of CMIP6 with values spanning 1.8-5.6 K is larger than the range of CMIP5 (2.1-4.7 K). Recently, many studies reported that the ECS from several CMIP6 models has increased substantially [20][21][22][23][24]. The ECS of K-ACE is higher than that of CMIP5 and near high bound of CMIP6 models (Figure 2b). Previous studies report that a wide range of ECS values produced by global climate models and its uncertainty are mainly caused by radiative feedbacks [17,20,22,49]. To understand the difference in the climate sensitivity of K-ACE regarding the radiative feedback compared to CMIP5 and CMIP6 models, the global mean radiative feedback contribution of ECS are divided into feedbacks at clearsky and cloud sky. In Figure 3, radiative feedback components are calculated as the slope of the linear fit of radiative flux change against temperature change for 150 years [48]. Hereafter, net radiative feedback, clear-sky feedback, and cloud radiative effect feedback are represented by λNET, λCS, and λCRE, respectively. SW and LW components of clear-sky feedback and cloud radiative effect are indicated by λSWCS, λLWCS, λSWCRE, and λLWCRE, respectively. The CRE is defined as the difference between all-sky and clear-sky net radiative fluxes.
Taken individually, cloudy components of the net feedback are significant outliers from the CMIP5 average and clear-sky components mostly fall within the range of CMIP5 and CMIP6 models. Under clear sky, the λSWCS and λLWCS of K-ACE are 0.61 Wm −2 K −1 and −1.82 Wm −2 K −1 , respectively. The combination of λSWCS and λLWCS results in negative λCS of K-ACE showing similar magnitude with CMIP5 models. The cloud components of K-ACE are strong positive λSWCRE with 0.97 Wm −2 K −1 and negative λLWCRE with −0.46 Wm −2 K −1 , and both λSWCRE and λLWCRE are extreme in the range of CMIP5 model spread. Senior et al. [32] reports that λSWCRE and λLWCRE of HadGEM3-GC2 tend toward the higher and lower ends of the range of CMIP5 models. K-ACE also shows the similar results. Similarly, Previous studies report that a wide range of ECS values produced by global climate models and its uncertainty are mainly caused by radiative feedbacks [17,20,22,49]. To understand the difference in the climate sensitivity of K-ACE regarding the radiative feedback compared to CMIP5 and CMIP6 models, the global mean radiative feedback contribution of ECS are divided into feedbacks at clear-sky and cloud sky. In Figure 3, radiative feedback components are calculated as the slope of the linear fit of radiative flux change against temperature change for 150 years [48]. Hereafter, net radiative feedback, clear-sky feedback, and cloud radiative effect feedback are represented by λ NET , λ CS , and λ CRE , respectively. SW and LW components of clear-sky feedback and cloud radiative effect are indicated by λ SWCS , λ LWCS , λ SWCRE , and λ LWCRE , respectively. The CRE is defined as the difference between all-sky and clear-sky net radiative fluxes.
Taken individually, cloudy components of the net feedback are significant outliers from the CMIP5 average and clear-sky components mostly fall within the range of CMIP5 and CMIP6 models. Under clear sky, the λ SWCS and λ LWCS of K-ACE are 0.61 Wm −2 K −1 and −1.82 Wm −2 K −1 , respectively. The combination of λ SWCS and λ LWCS results in negative λ CS of K-ACE showing similar magnitude with CMIP5 models. The cloud components of K-ACE are strong positive λ SWCRE with 0.97 Wm −2 K −1 and negative λ LWCRE with −0.46 Wm −2 K −1 , and both λ SWCRE and λ LWCRE are extreme in the range of CMIP5 model spread. Senior et al. [32] reports that λ SWCRE and λ LWCRE of HadGEM3-GC2 tend toward the higher and lower ends of the range of CMIP5 models. K-ACE also shows the similar results. Similarly, the range of the λ SWCRE and λ LWCRE for CMIP6 models extend in a positive and negative direction, respectively, compared to CMIP5 models. Overall, positive λ CRE in K-ACE, even with cancellation by negative λ CS , contribute to high bound of λ NET (small absolute value), leading to high values of ECS. Andrews et al. [22] and Golaz et al. [24] suggest that an unusual combination of forcing and feedback leads to a higher ECS. This relationship between λ NET and ECS is also represented. Additionally, the spread of λ CRE in inter-model differences is larger than that of λ CS and model spread of λ NET is influenced by a wide spread of λ CRE . Considering this, λ CRE is the key contributor of uncertainty (Figure 3), which is comparable with many previous studies [7,17,20,22,32,[49][50][51][52].
Atmosphere 2020, 11, x FOR PEER REVIEW 5 of 13 the range of the λSWCRE and λLWCRE for CMIP6 models extend in a positive and negative direction, respectively, compared to CMIP5 models. Overall, positive λCRE in K-ACE, even with cancellation by negative λCS, contribute to high bound of λNET (small absolute value), leading to high values of ECS. Andrews et al. [22] and Golaz et al. [24] suggest that an unusual combination of forcing and feedback leads to a higher ECS. This relationship between λNET and ECS is also represented. Additionally, the spread of λCRE in inter-model differences is larger than that of λCS and model spread of λNET is influenced by a wide spread of λCRE. Considering this, λCRE is the key contributor of uncertainty (Figure 3), which is comparable with many previous studies [7,17,20,22,32,[49][50][51][52]. Many studies have focused on the analysis of radiative feedback in clear and cloudy sky conditions [7,32,53,54] and on the differences in the SW cloud feedback [49,[55][56][57]. Further, Ceppi et al. [6] reports that the contributions to λSWCRE and λLWCRE are far from being spatially homogeneous which is influenced by cloud distribution. Therefore, global patterns of corresponding feedback components are also investigated in this study for understanding the effect of CRE feedback to higher ECS.
The difference in λNET between the CMIP5 ensemble and K-ACE is significant in spatial distribution ( Figure 4). Unlike an El Niño-like pattern in the mean field of CMIP5 models, the λNET of K-ACE occurs positively over the whole Pacific region and strong negative feedback occurs in the maritime continent (Figure 4a,b). These areas are strongly related to where cloud processes are important, such as in the Madden-Julian oscillation and monsoon system, and may be attributed to the improvements in the cloud process simulations due to the implementation of new cloud microphysics scheme [32]. Over the Tibet area (high altitude), a stronger positive feedback compared to CMIP5, is due to the decrease in albedo. However, the λNET of CMIP6 ensemble shows similar distribution with CMIP5 except for a positive value in the high northern latitude (Figure 4c). In general, the difference between the K-ACE and CMIP5/CMIP6 ensemble is larger over the ocean than over land. Many studies have focused on the analysis of radiative feedback in clear and cloudy sky conditions [7,32,53,54] and on the differences in the SW cloud feedback [49,[55][56][57]. Further, Ceppi et al. [6] reports that the contributions to λ SWCRE and λ LWCRE are far from being spatially homogeneous which is influenced by cloud distribution. Therefore, global patterns of corresponding feedback components are also investigated in this study for understanding the effect of CRE feedback to higher ECS.
The difference in λ NET between the CMIP5 ensemble and K-ACE is significant in spatial distribution ( Figure 4). Unlike an El Niño-like pattern in the mean field of CMIP5 models, the λ NET of K-ACE occurs positively over the whole Pacific region and strong negative feedback occurs in the maritime continent (Figure 4a,b). These areas are strongly related to where cloud processes are important, such as in the Madden-Julian oscillation and monsoon system, and may be attributed to the improvements in the cloud process simulations due to the implementation of new cloud microphysics scheme [32]. Over the Tibet area (high altitude), a stronger positive feedback compared to CMIP5, is due to the decrease in albedo. However, the λ NET of CMIP6 ensemble shows similar distribution with CMIP5 except for a positive value in the high northern latitude (Figure 4c). In general, the difference between the K-ACE and CMIP5/CMIP6 ensemble is larger over the ocean than over land.

Clear-Sky Feedback
In Figure 3, λLWCS, λSWCS, and λCS of the K-ACE are within the range of CMIP5 models. The λLWCS is the most important negative feedback (Figure 5a-c) contributing to negative λCS to maintain the stability of the climate system after the appearance of a strong external perturbation [7]. Those clearsky components of K-ACE show similar spatial patterns to the CMIP5 and CMIP6 results whereas the magnitude in λSWCS of K-ACE is slightly weaker than of those models, which is due to the Arctic region (Figure 5d-f). There is large positive λCS of K-ACE in high latitude areas and high-altitude areas (Figure 5g-i), reflecting the warming effect of sea-ice melting [7]. These have been consistent with reduced clear-sky albedo due to a loss of sea-ice and snow cover with increased global temperature [34].

Clear-Sky Feedback
In Figure 3, λ LWCS , λ SWCS , and λ CS of the K-ACE are within the range of CMIP5 models. The λ LWCS is the most important negative feedback (Figure 5a-c) contributing to negative λ CS to maintain the stability of the climate system after the appearance of a strong external perturbation [7]. Those clear-sky components of K-ACE show similar spatial patterns to the CMIP5 and CMIP6 results whereas the magnitude in λ SWCS of K-ACE is slightly weaker than of those models, which is due to the Arctic region (Figure 5d-f). There is large positive λ CS of K-ACE in high latitude areas and high-altitude areas (Figure 5g-i), reflecting the warming effect of sea-ice melting [7]. These have been consistent with reduced clear-sky albedo due to a loss of sea-ice and snow cover with increased global temperature [34].

Clear-Sky Feedback
In Figure 3, λLWCS, λSWCS, and λCS of the K-ACE are within the range of CMIP5 models. The λLWCS is the most important negative feedback (Figure 5a-c) contributing to negative λCS to maintain the stability of the climate system after the appearance of a strong external perturbation [7]. Those clearsky components of K-ACE show similar spatial patterns to the CMIP5 and CMIP6 results whereas the magnitude in λSWCS of K-ACE is slightly weaker than of those models, which is due to the Arctic region (Figure 5d-f). There is large positive λCS of K-ACE in high latitude areas and high-altitude areas (Figure 5g-i), reflecting the warming effect of sea-ice melting [7]. These have been consistent with reduced clear-sky albedo due to a loss of sea-ice and snow cover with increased global temperature [34].

CRE Feedback
CRE feedback is positive in CMIP5, and CRE feedbacks play an important role in the higher sensitivity in GCM [54]. Additionally, CRE feedback is the main contributor to the uncertainty in climate sensitivity [53] and exhibits the largest amount of inter-model spread, originating primarily form the Atmosphere 2020, 11, 1218 7 of 13 SW effect. The λ CRE of CMIP5 and CMIP6 ensemble depends on strong positive λ SWCRE with the compensation of negative λ LWCRE across the tropical Pacific ( Figure 6). Moreover, opposite tendencies of λ SWCRE and λ LWCRE occur horse-shoe patterns over subtropical Pacific and maritime continent, as clouds reflects solar radiation into space and block LW radiation from the surface. This is related to the increased sea surface temperature in the East Pacific as the El Niño-like response.
tendencies of λSWCRE and λLWCRE occur horse-shoe patterns over subtropical Pacific and maritime continent, as clouds reflects solar radiation into space and block LW radiation from the surface. This is related to the increased sea surface temperature in the East Pacific as the El Niño-like response.
The spatial distribution of CMIP5 and CMIP6 CRE feedback and K-ACE is clearly different (Figure 6a-c). The opposite tendencies in λSWCRE and λLWCRE patterns are also found in K-ACE over the tropical Pacific. However, strongly positive λSWCRE and negative λLWCRE occurred in K-ACE over the tropical Western Pacific and Philippines region compared to the CMIP5 and CMIP6 ensemble means (Figure 6d-i). K-ACE also shows positive λSWCRE in the ocean and at mid-latitudes (approximately 30° N and S), enhancing positive λCRE. This is a different pattern of El Niño-like patterns represented in the CMIP5 and CMIP6 ensemble (Figure 6b,c). However, as mentioned in previous studies [7,32], the spatial pattern of λCRE is dominated by positive a λSWCRE, which is investigated in K-ACE (Figure 6a,d).

Low-Level Cloud Amount
Andrews et al. [22] reported that a new mixed-phase cloud scheme effects cloud feedback. Considering this, the relationship between simulated cloud properties based on the cloud scheme and λCRE are investigated in this study. This approach may provide a new hint further understanding of the model spread of λSWCRE and λLWCRE. In previous study, Senior et al. [32] reports that a new prognostic cloud fraction and condensation scheme [44,45] may contribute to a strong cloud radiative Figure 6. Global distribution of the net (first row), shortwave (SW; second row), and longwave (LW; third row) cloud radiative effect (CRE) feedback components in Wm −2 K −1 for the K-ACE (a,d,g), the CMIP5 (b,e,h), and the CMIP6 (c,f,i) models. These patterns are determined from the slope of the Gregory regression [48] in the abrupt experiment.
The spatial distribution of CMIP5 and CMIP6 CRE feedback and K-ACE is clearly different (Figure 6a-c). The opposite tendencies in λ SWCRE and λ LWCRE patterns are also found in K-ACE over the tropical Pacific. However, strongly positive λ SWCRE and negative λ LWCRE occurred in K-ACE over the tropical Western Pacific and Philippines region compared to the CMIP5 and CMIP6 ensemble means (Figure 6d-i). K-ACE also shows positive λ SWCRE in the ocean and at mid-latitudes (approximately 30 • N and S), enhancing positive λ CRE . This is a different pattern of El Niño-like patterns represented in the CMIP5 and CMIP6 ensemble (Figure 6b,c). However, as mentioned in previous studies [7,32], the spatial pattern of λ CRE is dominated by positive a λ SWCRE, which is investigated in K-ACE (Figure 6a,d).

Low-Level Cloud Amount
Andrews et al. [22] reported that a new mixed-phase cloud scheme effects cloud feedback. Considering this, the relationship between simulated cloud properties based on the cloud scheme and λ CRE are investigated in this study. This approach may provide a new hint further understanding of the model spread of λ SWCRE and λ LWCRE . In previous study, Senior et al. [32] reports that a new prognostic cloud fraction and condensation scheme [44,45] may contribute to a strong cloud radiative response. This new mixed-phase cloud scheme shows improved cloud formation (well-behaved in the very low and high cloud fraction) in the maritime continent as well as tropical and subtropical region [44,45] and has been implemented by UM in GA7.1. The family models of HadGEM3 include this physics scheme; K-ACE has also used this scheme.
Atmosphere 2020, 11, 1218 8 of 13 As shown in Figure 7a, low-level cloud tends to decrease during warming, which in turn affects the positive λ SWCRE compared to CMIP5 models (Figure 3; Figure 6a). Hence, the characteristic of the K-ACE cloud scheme (less low-level cloud) makes λ SWCRE stronger compared to CMIP5 models. This is similar to Chen et al. [7], where the λ SWCRE change of CAMS-CSM (one of CMIP6 model) is explained by low-level cloud change. Based on the characteristics of cloud scheme, the amount of low-level ice cloud of K-ACE is reduced in the mid-latitudes, decreasing albedo contributions to surface warming [44,45]. This process causes a positive λ SWCRE in the mid-latitude compared to CMIP5 models (Figure 7b,c). Figure 7b indicates low cloud feedback, and the zonal mean of low cloud feedback is calculated by the global distributions of the slope in Figure 7a. Additional evidence suggests that decreasing tropical low cloud (15 • S-15 • N; Figure 7b) with increasing temperature makes stronger λ SWCRE compared to CMIP5 (Figure 7c; increasing temperature produces less low clouds, more incoming radiation, and increased warming) that are directly related to cloud formations in K-ACE. Increased higher anvil cloud (less low-level cloud) in the tropical Pacific [45] contributes to a stronger λ SWCRE . In addition, λ SWCRE increases in mid-latitude, which is consistent with Andrews et al. [22]. This difference of λ SWCRE is at mid-latitudes due to mixed-phase cloud and their interactions [22]. Overall, less low cloud (stronger λ SWCRE ) from K-ACE implies a stronger response to CO 2 and this makes higher ECS.
Atmosphere 2020, 11, x FOR PEER REVIEW 8 of 13 response. This new mixed-phase cloud scheme shows improved cloud formation (well-behaved in the very low and high cloud fraction) in the maritime continent as well as tropical and subtropical region [44,45] and has been implemented by UM in GA7.1. The family models of HadGEM3 include this physics scheme; K-ACE has also used this scheme. As shown in Figure 7a, low-level cloud tends to decrease during warming, which in turn affects the positive λSWCRE compared to CMIP5 models (Figure 3; Figure 6a). Hence, the characteristic of the K-ACE cloud scheme (less low-level cloud) makes λSWCRE stronger compared to CMIP5 models. This is similar to Chen et al. [7], where the λSWCRE change of CAMS-CSM (one of CMIP6 model) is explained by low-level cloud change. Based on the characteristics of cloud scheme, the amount of low-level ice cloud of K-ACE is reduced in the mid-latitudes, decreasing albedo contributions to surface warming [44,45]. This process causes a positive λSWCRE in the mid-latitude compared to CMIP5 models ( Figure  7b,c). Figure 7b indicates low cloud feedback, and the zonal mean of low cloud feedback is calculated by the global distributions of the slope in Figure 7a. Additional evidence suggests that decreasing tropical low cloud (15° S-15° N; Figure 7b) with increasing temperature makes stronger λSWCRE compared to CMIP5 (Figure 7c; increasing temperature produces less low clouds, more incoming radiation, and increased warming) that are directly related to cloud formations in K-ACE. Increased higher anvil cloud (less low-level cloud) in the tropical Pacific [45] contributes to a stronger λSWCRE. In addition, λSWCRE increases in mid-latitude, which is consistent with Andrews et al. [22]. This difference of λSWCRE is at mid-latitudes due to mixed-phase cloud and their interactions [22]. Overall, less low cloud (stronger λSWCRE) from K-ACE implies a stronger response to CO2 and this makes higher ECS.  Figure 8a shows a global pattern of high cloud feedback (calculated by linear regression between change in cloud area (%) and surface air temperature (K); similar to the slope in Figure 7a). The red and blue shadings indicate increasing and decreasing cloud area with increasing temperature, respectively, and darker shading means a higher response of CO2. For high-level clouds, the cloud decreases over the maritime continent and increases over the tropical Pacific (Figure 8a). Moreover, high-level cloud increases over the subtropical areas. Based on the characteristics of cloud altitude in K-ACE, the mid-level cloud amount has effectively decreased due to the greater amount of high anvil cloud (increased high-level cloud due to the moistening in the upper tropical troposphere) [45]. Higher cloud tops are colder and emit less radiation to space. Therefore, an increase in the altitude of  Figure 8a shows a global pattern of high cloud feedback (calculated by linear regression between change in cloud area (%) and surface air temperature (K); similar to the slope in Figure 7a). The red and blue shadings indicate increasing and decreasing cloud area with increasing temperature, respectively, and darker shading means a higher response of CO 2 . For high-level clouds, the cloud decreases over the maritime continent and increases over the tropical Pacific (Figure 8a). Moreover, high-level cloud increases over the subtropical areas. Based on the characteristics of cloud altitude in K-ACE, the mid-level cloud amount has effectively decreased due to the greater amount of high anvil cloud (increased high-level cloud due to the moistening in the upper tropical troposphere) [45]. Higher cloud tops are colder and emit less radiation to space. Therefore, an increase in the altitude of cloud top leads to warming by reducing outgoing LW radiation. This result is consistent with Andrews et al. [22] who demonstrated that a large high cloud response exists, which leads to a higher net cloud feedback. Additionally, the global pattern of λ LWCRE (Figure 6c) and cloud feedback (Figure 8a) is spatially similar. Considering this, a higher cloud formation of K-ACE compared to CMIP5 models contribute to stronger λ LWCRE , which could result in higher ECS. net cloud feedback. Additionally, the global pattern of λLWCRE (Figure 6c) and cloud feedback ( Figure  8a) is spatially similar. Considering this, a higher cloud formation of K-ACE compared to CMIP5 models contribute to stronger λLWCRE, which could result in higher ECS.

Higher Altitude Cloud
Additionally, there is a cancellation of λSWCRE and λLWCRE in the tropical Pacific as expected for the increased amount of high-level cloud in this region (Figure 6a,c). Overall, λCRE is small in tropics (Figure 8b), which emphasizes that the amount changes of low-level cloud are important.

Conclusions
This study describes model characteristics for high climate sensitivity of K-ACE which has been developed by NIMS/KMA as part of participating in CMIP6. Abrupt experiments, as an entry card to CMIP6, are analyzed for climate sensitivity and feedback processes, and compared to 23 CMIP5 models and 18 CMIP6 models. Climate sensitivity is defined as the global temperature response to a doubling CO2 concentration in the atmosphere compared to pre-industrial levels. It is a basic metric of the climate research and essential for carbon reduction policies to achieve a specific warming target [7]. Recently, it is reported that the ECS in CMIP6 models has increased compared to CMIP5 and cloud feedback is considered the main cause [20][21][22]. Following this, we have analyzed the ECS and related feedbacks for K-ACE in a similar way and the results are summarized as follows.

•
In the idealized CO2 experiment, temperature rising in K-ACE is larger than in the CMIP5 ensemble mean. Based on the cloud characteristics of K-ACE, less ice cloud in mid-latitudes (decreasing albedo, more surface warming) and less low-level cloud in the tropics (increasing temperature produces less low clouds, more incoming radiation, and increased warming) contribute to a stronger λSWCRE and higher ECS.

•
For high-level cloud, cloud characteristics of K-ACE (more high cloud due to the moistening in the upper tropical troposphere) contribute to less outgoing LW. In addition, the global pattern of λLWCRE (Figure 6d) and cloud feedback (Figure 8a) is spatially similar. This infers that higher cloud formation of K-ACE compared to CMIP5 models contributes to stronger λLWCRE, which could cause higher ECS. Additionally, there is a cancellation of λ SWCRE and λ LWCRE in the tropical Pacific as expected for the increased amount of high-level cloud in this region (Figure 6a,c). Overall, λ CRE is small in tropics (Figure 8b), which emphasizes that the amount changes of low-level cloud are important.

Conclusions
This study describes model characteristics for high climate sensitivity of K-ACE which has been developed by NIMS/KMA as part of participating in CMIP6. Abrupt experiments, as an entry card to CMIP6, are analyzed for climate sensitivity and feedback processes, and compared to 23 CMIP5 models and 18 CMIP6 models. Climate sensitivity is defined as the global temperature response to a doubling CO 2 concentration in the atmosphere compared to pre-industrial levels. It is a basic metric of the climate research and essential for carbon reduction policies to achieve a specific warming target [7]. Recently, it is reported that the ECS in CMIP6 models has increased compared to CMIP5 and cloud feedback is considered the main cause [20][21][22]. Following this, we have analyzed the ECS and related feedbacks for K-ACE in a similar way and the results are summarized as follows.

•
In the idealized CO 2 experiment, temperature rising in K-ACE is larger than in the CMIP5 ensemble mean. The ECS of K-ACE calculated by Gregory et al. [48] is 4.83 K, substantially higher than the range of CMIP5 of 2.1-4.7 K and the higher bound of 1.8-5.6 K of CMIP6 models. • Radiative feedback is decomposed by clear-sky and cloud radiative feedback effects. Under clear skies, λ SWCS and λ LWCS are within the range of CMIP5 and CMIP6 models. Meanwhile, CRE feedback of K-ACE shows large amplitude. SW and LW components of CRE feedback (λ SWCRE , λ LWCRE ) tend toward the higher and lower end of the range of CMIP5 and CMIP6 models.

•
Based on the cloud characteristics of K-ACE, less ice cloud in mid-latitudes (decreasing albedo, more surface warming) and less low-level cloud in the tropics (increasing temperature produces less low clouds, more incoming radiation, and increased warming) contribute to a stronger λ SWCRE and higher ECS.

•
For high-level cloud, cloud characteristics of K-ACE (more high cloud due to the moistening in the upper tropical troposphere) contribute to less outgoing LW. In addition, the global pattern of λ LWCRE (Figure 6d) and cloud feedback (Figure 8a) is spatially similar. This infers that higher cloud formation of K-ACE compared to CMIP5 models contributes to stronger λ LWCRE , which could cause higher ECS.
The relationships between simulated cloud properties based on cloud scheme and λ CRE are presented in this study. This approach provides some understanding of the cloud feedback based on cloud properties (amount and altitude). However, there are several limitations in this approach. The various climate models participated in CMIP6 have implemented different cloud schemes and CRE feedback is too complex to decompose into contributions from several cloud properties. Therefore, further work is necessary to understand the criteria for determining the spatial pattern of a cloud formulation related cloud scheme and the influence of these patterns on cloud properties at regional and global scales.

Conflicts of Interest:
The authors declare no conflict of interest.