Positive Low Cloud Feedback Primarily Caused by Increasing Longwave Radiation From the Sea Surface in Two Versions of a Climate Model

Low cloud feedback in global warming projections by climate models is characterized by its positive sign, the mechanism of which is not well understood. Here we propose that the positive sign is primarily caused by the increase in upward longwave radiation from the sea surface. We devise numerical experiments that enable separation of the feedback into components coming from physically distinct causes. Results of these experiments with a climate model indicate that increases in upward longwave radiation from the sea surface cause warming and absolute drying in the boundary layer, leading to the positive low cloud feedback. The absolute drying results from decrease in surface evaporation, and also from decrease in inversion strength which enhances vertical mixing of drier free tropospheric air into the boundary layer. This mechanism is different from previously proposed understanding that positive low cloud feedback is caused by increases in surface evaporation or vertical moisture contrast.

• The increase in longwave radiation from the sea surface is a leading order cause of the positive low cloud feedback in a climate model • This increase in longwave radiation leads to warming and drying in the boundary layer, which contributes to the decrease in the low cloud • This mechanism is not associated with increases in surface evaporation or vertical moisture contrast

Supporting Information:
Supporting Information may be found in the online version of this article.
Several studies have been conducted to address this issue by attributing simulated changes in low cloud to changes in environmental factors (e.g., Brient & Schneider, 2016;Ceppi & Nowack, 2021;Cesana & Del Genio, 2021;Klein et al., 2017;McCoy et al., 2017;Myers & Norris, 2016;Qu, Hall, Klein, & DeAngelis, 2015;Qu et al., 2014;Zhai et al., 2015).Qu et al. (2014), among others, developed a heuristic model which interprets the positive low cloud feedback in the subtropical low cloud regions in GCMs.The model indicates that changes in low cloud amount mainly come from two factors: local SST warming and increase in the strength of the inversion capping the atmospheric boundary layer, which is measured by the Estimated Inversion Strength (EIS, Wood & Bretherton, 2006).The local SST warming tends to decrease low cloud, while the enhancement of EIS tends to increase the cloud.The net effect is a decrease in low cloud amount because the effect of the SST outweighs that of the EIS in most models.
The mechanism underlying the effect of EIS on low cloud is well understood (Klein & Hartmann, 1993;Wood & Bretherton, 2006).However, the mechanism of how the local SST warming influences the low cloud is still under debate.The following two mechanisms have been proposed, based on studies using Large Eddy Simulations.
First, SST warming leads to an increase in surface latent heat flux, which enhances vertical mixing by turbulence or convection in the lower troposphere.This enhances entrainment of drier air from the free troposphere into the moister boundary layer, desiccating low cloud (Rieck et al., 2012).Second, the increase in latent heat flux from the sea surface induces an increase in water vapor specific humidity in the atmosphere.The magnitude of the increase in humidity is more pronounced in the boundary layer than in free troposphere, increasing the vertical moisture contrast.This increase in moisture contrast enhances the efficiency with which vertical mixing dehydrates the boundary layer, reducing low cloud (Bretherton & Blossey, 2014;Sherwood et al., 2014;van der Dussen et al., 2015).
Recently, however, detailed examination of some GCM experiments gave results which are not consistent with the above understanding.For instance, Webb et al. (2018) explored the impact of surface latent heat flux on low cloud amount, forcing the latent heat flux to increase at different rates with SST warming in HadGEM2-A.They found that the magnitude of the low cloud decrease becomes smaller when the latent heat flux is forced to increase at higher rates.Similar results were obtained by Watanabe et al. (2018) using MIROC5.These findings suggest that mechanisms other than the increase in latent heat flux are needed to explain the decrease in low cloud with SST warming in climate models.However, such mechanisms are yet to be identified.Here we propose an alternative mechanism for the low cloud decrease with SST warming based on a new method for decomposing feedbacks in GCM experiments.We argue that the increase in upward longwave radiation from the sea surface is a leading order cause of the low cloud decrease.

Numerical Experiments
The low cloud feedback is investigated using an atmospheric GCM MIROC6 with the spatial resolution of T85 (∼1.4°) with 81 vertical levels (Tatebe et al., 2019).The simulation protocol follows that of the Atmospheric Model Intercomparison Project (AMIP), in which the atmosphere is forced by a historical SST (AMIP experiment) and the SST uniformly warmed by 4K (AMIP-p4K experiment).The SSTs are not affected by the changes in the atmosphere since they are prescribed as a boundary condition.These AMIP-type experiments provide a good approximation to the cloud feedbacks determined from coupled atmosphere-ocean CO 2 -forced simulations (Qin et al., 2022;Ringer et al., 2014).
In the AMIP-p4K run, the uniform SST warming of 4K compared to the AMIP run modifies the atmosphere via two causal pathways, first by increasing the upward longwave radiation from the sea surface, and second by changing the turbulent transport at the air-sea interface, such as the latent and sensible heat fluxes (Figure 1).The decrease in low cloud amount, and hence the positive low cloud feedback, is a result of these two causal factors.
We attempt to better understand the roles of the two factors by adding two experiments.In the first experiment, SST is raised by 4K only when calculating the upward longwave radiation from the sea surface using Planck function (AMIP-p4Krad experiment, Figure 1).In the second, SST is raised by 4K only when calculating the turbulent transport at the air-sea interface using bulk aerodynamic formulas (AMIP-p4Kturb experiment).More details of the two experiments are given in the Supporting Information (Text S1 in Supporting Information S1).
All of the experiments are integrated for 1979-2014 and the output is averaged for 36 years.
The differences of the SST warming experiments compared to the AMIP run are called "total response (AMIP-p4K minus AMIP)," "radiative component (AMIP-p4Krad minus AMIP)," and "turbulent component (AMIP-p4Kturb 10.1029/2023GL104786 3 of 10 minus AMIP)," respectively.As the total response, we focus on the low cloud feedback, and write it as a sum of the radiative component, the turbulent component, and a synergy term (Figure 1).Now the low cloud feedback is separated into components that originate from physically distinct causes, namely, the effect of increasing SST on upwelling surface longwave radiation and its effect on surface turbulent fluxes.The intention here is to see which component makes the low cloud feedback positive.The synergy is a residual term that is evaluated as the difference between the total response and the sum of the radiative and turbulent components.It represents the effect of the radiative and turbulent components working together.
All of the experiments, as outlined above, are repeated using another atmospheric GCM MIROC5 with the spatial resolution of T42 (∼2.8°) with 40 vertical levels (Ogura et al., 2017;Shiogama et al., 2012;Watanabe et al., 2010).MIROC5 is different from MIROC6 in terms of its representation of the atmospheric boundary layer.Specifically, MIROC5 does not include a shallow convection parameterization while MIROC6 does.Still, the results from both models are consistent with the main conclusions.For conciseness, we present results from MIROC6 in the main part, while those from MIROC5 are shown in the Supporting Information (Figures S1 and S2 in Supporting Information S1).

Results
We first present the low cloud feedback simulated by MIROC6 in Figure 2a.This is evaluated by multiplying changes in the ISCCP low cloud amount by the cloud radiative kernel, which gives the changes in radiation flux at the TOA induced by the low cloud changes (Bodas-Salcedo et al., 2011;Klein & Jakob, 1999;Webb et al., 2001;Zelinka et al., 2012).The ISCCP cloud amount with cloud top pressure greater than 680 hPa is used for the evaluation.In Figure 2a, we confirm that the global average low cloud feedback is positive.The positive signal is particularly evident in subtropical marine regions off the western coasts of continents, where low clouds prevail in both observations and model control climates.
The low cloud feedback is separated into the radiative component, turbulent component, and synergy as shown in Figures 2b, 2c, and 2e.The radiative component is characterized with positive contributions over the oceans, while the turbulent component is dominated by negative contributions (Figures 2b and 2c).If we add the two components together, as shown in Figure 2d, the result captures the geographical pattern (especially the sign) of the total low cloud feedback in Figure 2a.The pattern correlation between Figures 2a and 2d is 0.81.Therefore, the low cloud feedback can be approximated as a sum of the radiative and turbulent components, although the synergy effect is not negligible as shown in Figure 2e.
Focusing on the sum of the radiative and the turbulent components in Figure 2d, we find that the low cloud feedback becomes positive where the radiative component outweighs the turbulent component.Without the radiative component, the low cloud feedback would have been negative overall (Figure 2c).This means that the low cloud feedback becomes positive because of the radiative component.In other words, the positive sign of the feedback is mainly attributed to the increase in upward longwave radiation from the sea surface.This argument applies to MIROC5, too (Figure S1 in Supporting Information S1).
How does the longwave radiation cause the positive low cloud feedback?The mechanism is further examined, focusing on area averages over the five oceanic regions indicated by the black rectangles in Figure 2.These regions are chosen because the positive low cloud feedback stands out here in MIROC6 (Figure 2a), and also because they match the low cloud regions based on observations (Qu et al., 2014).Here, vertical profiles of cloud-related variables are examined in Figure 3.We focus on the cloud amount below the 680 hPa level because this is where the low cloud feedback originates (Figures 3a and 3e).Note also that the low cloud feedback is strongly correlated with the cloud amount, but less well with the cloud optical thickness or cloud top pressure (Figure S3 in Supporting Information S1).
The total response of the cloud amount below the 680 hPa level (Figure 3e, black) shows a characteristic dipole pattern, in which a cloud decrease above (σ-p level ≈ 0.85) is moderated by a cloud increase below (σ-p level ≳ 0.9).The dipole pattern reflects shallowing of the boundary layer cloud at σ-p level ≈ 0.9 (Figure 3a).As a comparison, we also plot the radiative and turbulent components in Figure 3e (red and blue).Clearly, the turbulent component (blue) fails to reproduce the total response (black) at the σ-p level ≳ 0.9, namely, the blue curve exceeds the black one.This explains how the turbulent component shows increase in low cloud, leading to the negative feedback.In contrast, the radiative component (red) shows a decrease in low cloud at σ-p level ≈ 0.9, which opposes the cloud increase in the turbulent component (blue).When added together, the radiative and turbulent components (green) roughly reproduce the dipole pattern in the total response (black), although the positive and negative maxima are exaggerated.Hence, the low cloud decrease in the radiative component (red) is the key to understanding the low cloud decrease in the total response (black).
The low cloud decrease in the radiative component (Figure 3e, red) is consistent with a decrease in relative humidity (Figure 3f, red), which comes from both a warming and a decrease in specific humidity (Figures 3g and 3h,red).This can be confirmed by looking at the geographical distribution (Figure S4 in Supporting Information S1).The warming is caused by the increase in upward longwave radiation from the sea surface, which is absorbed by the atmosphere (Figure 3i).The decrease in specific humidity can be explained by two mechanisms.First, the magnitude of the warming is larger in the boundary layer compared to the free troposphere, having a bottom-heavy vertical profile (Figure 3h, red).This decreases the strength of the inversion capping the boundary layer.As a result, vertical mixing across the inversion increases, making the boundary layer less humid (Klein & Hartmann, 1993).Second, the longwave-induced warming of the atmosphere increases the static stability at the air-sea interface.Note that the SST is kept the same as the AMIP experiment when calculating the turbulent transport at the air-sea interface in the AMIP-p4Krad experiment.The increase in the static stability suppresses the turbulent transport of water vapor from the sea surface, thereby contributing to the decrease in specific humidity (Text S2 and Figure S8 in Supporting Information S1).
The warming and the absolute drying in the boundary layer, as described above, leads to the low cloud decrease in the radiative component.The mechanism may be summarized as "Cloud Reduction due to Increased Surface The δq is defined as the specific humidity q at 1,000 hPa minus q at 700 hPa.The delta, Δ, denotes changes induced by the SST warming of 4K.The data are averages over the low cloud regions in Figure 2.
Temperature Longwave Emission (CRISTLE)."In addition, the decrease in the low cloud initiates a process that reduces the low cloud further.Namely, the decrease in the low cloud causes weakening of the downward longwave radiation from the cloud.As a result, divergence in the downward longwave radiation decreases, which leads to weakening of the radiative cooling of the boundary layer (Figures S7c,S7f,and S7i in Supporting Information S1,blue).This contributes to warming and a decrease in relative humidity, thereby reducing the low cloud further (Figure S6e in Supporting Information S1, green, Brient & Bony, 2012).We note that the low cloud decrease in the radiative component is not associated with an increase in specific humidity or surface evaporation (Figure 3g and Figure S8a in Supporting Information S1).We also considered a number of other possible explanations for the low cloud reductions in the radiative component (Table S1 in Supporting Information S1).
In the turbulent component, by contrast, the low cloud changes are associated with the increase in specific humidity and surface evaporation.We attribute the low cloud increases in the turbulent component to multiple processes that compete with each other (e.g., Narenpitak & Bretherton, 2019;Schneider et al., 2019;Tan et al., 2017;Vial et al., 2016;Wyant et al., 2009).For instance, the magnitude of the increase in specific humidity is larger at lower altitudes, which enhances the moisture contrast between the free troposphere and the boundary layer (Figure 3g, blue).As a result, the upward moisture flux by shallow convection increases, which tends to decrease the low cloud (Figures S5c and S5f in Supporting Information S1, red, Zhang et al., 2013;Brient et al., 2016).In contrast, we also note that the vertical temperature profile stabilizes with warming, which increases strength of the inversion capping the boundary layer (Figure 3h, blue).As a result, vertical mixing across the inversion reduces, which tends to keep the boundary layer more humid and increase the low cloud (Miller, 1997;Tan et al., 2016).Understanding the roles of different processes within the turbulent component will be a subject of future studies.More details of the competing processes are given in Table S1 in Supporting Information S1.
In the AMIP-p4Krad and the AMIP-p4Kturb experiments, the SST warming takes place uniformly, including both the low cloud regions and the convective regions such as the western tropical Pacific.Readers might be interested in whether the SST warming changes deep convection, and whether the change in the deep convection has remote effects on the low clouds.Our preliminary answer is "yes, to some extent." In the AMIP-p4Kturb experiment, for instance, the SST warming leads to increase in precipitation over the western tropical Pacific and the tropical Indian oceans (Figure S9c in Supporting Information S1), which is related to the enhanced latent heating by the deep convection.In the low cloud regions, at 700 hPa level, temperature warms up by 4.8 K (Figure S10c in Supporting Information S1), which increases the EIS, and subsidence weakens by 4.2 hPa/day (Figure S11c in Supporting Information S1).These are consistent with the understanding that deep convection affects low clouds by changing the tropical overturning circulation and temperature in the free troposphere (e.g., Andrews & Webb, 2018;Erfani & Burls, 2019;Schiro et al., 2022;Silvers & Robinson, 2021;Williams et al., 2023), and both the warming and the weakening of subsidence will tend to increase the low clouds (e.g., Myers & Norris, 2013;Qu, Hall, Klein, & Caldwell, 2015).Regarding the AMIP-p4Krad experiment, the SST warming leads to reduction of precipitation over the western tropical Pacific and the tropical Indian oceans (Figure S9b in Supporting Information S1), which is related to suppressed latent heating by the deep convection.However, remote effects of the changes in the deep convection are relatively small.In the low cloud regions, at 700 hPa level, temperature warms up by only 0.3 K (Figure S10b in Supporting Information S1) and subsidence weakens by only 1.0 hPa/day (Figure S11b in Supporting Information S1).Those changes do not explain the decrease in the low cloud amount (Figure 3e, red).
Then, what is the role of the local SST warming in the AMIP-p4Kturb experiment?Does it decrease the low clouds, as indicated by the LES experiments?Currently, we have no answer for this.Additional efforts are needed to separately quantify the local and remote effects of the SST warming, which is a subject of future studies.
The results obtained so far illustrate how the low cloud feedback originates from the sea surface warming.The processes involved in the feedback are classified into the radiative and the turbulent components.The two components are dissimilar to each other, with the former decreasing the ISCCP low cloud amount (LCA), while the latter increases it.However, the two components are both related to changes in the EIS, as follows.In the radiative component, the LCA decreases as the EIS decreases (Figures 3e and 3h, red).In the turbulent component, the LCA increases as the EIS increases (Figures 3e and 3h, blue).In the synergy component, also, the LCA increases as the EIS increases (not shown).The relationship between the LCA and the EIS is qualitatively consistent with observation (Wood & Bretherton, 2006).
If we add the three components together, however, the relation between the LCA and the EIS changes compared to that above.Namely, the LCA decreases as the EIS increases (Figures 3e and 3h, black), which may appear counter-intuitive.Why does the relation between the LCA and the EIS break down when the components are added together?This issue is examined in Figure 3j.
In Figure 3j, the changes induced by the SST warming of 4K are represented by 2-D vectors on the ΔEIS-ΔLCA plane.The radiative component is shown in red, with the coordinate values of (∆EIS rad ,∆LCA rad ), while the turbulent component is shown by blue, with the coordinate values of (∆EIS turb ,∆LCA turb ).The two vectors appear in the third and the first quadrants, indicating that the LCA decreases (increases) as the EIS decreases (increases).Adding the two components together, we obtain the sum shown by green, with the coordinate values of (∆EIS turb + ∆EIS rad ,∆LCA turb + ∆LCA rad ).Now the vector appears in the fourth quadrant, indicating that the LCA decreases as the EIS increases, which captures the sign of the total response shown in black.
Focusing on the sum of the two components, we find that the LCA decreases as the EIS increases under the following conditions: ∆EISturb + ∆EISrad > 0, and ∆ LCAturb + ∆LCArad < 0 (1) Namely, the change in the EIS is dominated by the turbulent component, while the change in the LCA is dominated by the radiative component.In other words, the total response to the SST warming includes two counter-acting components, and which component dominates depends on the variable we look at.This explains how the relation between the LCA and the EIS changes when adding the radiative and turbulent components together.
We also note that rate of change in the LCA with respect to the EIS is different between the radiative and turbulent components, as follows: The conditions (1) can be met only under the condition (2).The condition (2) indicates that LCA is less sensitive to EIS in the turbulent component than in the radiative component.This may be because, in the turbulent component, the EIS increase is accompanied by an increase in vertical moisture contrast, δq (Figures 3g and 3h, blue).The change in the EIS tends to increase the LCA, while the change in the δq tends to decrease it, making the LCA less sensitive to the EIS (Kawai et al., 2017).
Similar arguments hold, even if we replace the EIS with the surface latent heat flux or the vertical moisture contrast, δq (Figures 3k and 3l).Namely, in the total response shown in black, the LCA decrease is accompanied by an increase in latent heat flux or δq.This can be explained by the fact that the LCA decrease is dominated by the radiative component while the increase in latent heat flux or δq is driven by the turbulent component.

Conclusions
In order to understand the reason for the positive sign of the low cloud feedback simulated by GCMs, we devise numerical experiments which enable separation of the feedback into three components, namely, the effect of increasing SST on upwelling surface longwave radiation, its effect on surface turbulent fluxes, and the synergy between the two.The numerical experiments are conducted using MIROC5 and MIROC6.The results indicate that the positive sign of the low cloud feedback is mainly attributed to the increase in longwave radiation from the sea surface, which leads to a warming and a drying in the boundary layer, as well as a decrease in the low cloud amount (LCA).The mechanism involved is summarized as "Cloud Reduction due to Increased Surface Temperature Longwave Emission (CRISTLE)."It is not associated with increases in surface latent heat flux or vertical moisture contrast.The decomposition of the feedback also helps to explain how the LCA decrease is accompanied by increases in the EIS, the latent heat flux, and the vertical moisture contrast.
The present study mainly discusses the positive low cloud feedback over the subtropical oceans off the western coast of the continents.If we broaden the scope, however, we find other regions where the low cloud feedback becomes negative due to changes in the surface turbulent fluxes.There are also regions where the synergy exceeds sum of the radiative and the turbulent terms.Therefore, the geographical pattern of the low cloud feedback on the global scale is determined by the changes in the upward surface longwave radiation and the turbulent fluxes, as well as their interaction.We also note that the low cloud feedback simulated in the present study includes

Figure 1 .
Figure 1.Schematic showing the experimental design.Ts_rad indicates the SST used for calculating LW radiation from the sea surface.Ts_turb is the SST used for calculating turbulent transport from the sea surface, including latent heat (LH) and sensible heat (SH) fluxes.

Figure 2 .
Figure 2. Low cloud feedback induced by 4K increases in SST.(a) Total low cloud feedback, (b) radiative component, (c) turbulent component, (d) sum of the radiative and turbulent components, and (e) synergy.Global averages are indicated at the top right of each panel.The units can be converted to [W/m 2 /K] by dividing by the surface warming of 4.54 K in the AMIP-p4K run.Black rectangles indicate low cloud regions focused on in Figure 3.

Figure 3 .
Figure 3. (a-i) Vertical profiles of cloud-related variables averaged over the low cloud regions in Figure 2. (a-d) for AMIP and AMIP-p4K experiments, and (e-i) for changes due to +4K SST warming.The vertical coordinate is hybrid σ-p on model level, which is compared with pressure levels on the top-right corner.Horizontal lines at the σ-p level of 0.67 mark the boundary between low-top clouds and middle-top clouds at 680 hPa.Diamonds indicate values at the lowest level.The changes in upward longwave, (i), are evaluated assuming that the atmosphere remains fixed at the AMIP condition.(j-l) Relationships between changes in low cloud amount and changes in (j) EIS, (k) latent heat flux, and (l) vertical moisture contrast δq.The δq is defined as the specific humidity q at 1,000 hPa minus q at 700 hPa.The delta, Δ, denotes changes induced by the SST warming of 4K.The data are averages over the low cloud regions in Figure2.