The copyright line for this article was changed on 13 APR 2015 after original online publication.
 To understand mechanisms of shortwave cloud-radiative feedback to global warming in a general circulation model (GCM), we analyzed the response of tropical clouds to uniform increase of sea surface temperature in an atmospheric GCM with two different experimental designs: a single Atmospheric Model Intercomparison Project (AMIP) run for 30 years and a series of 10 day weather hindcasts following the Transpose AMIP II (TAMIP). Given the fast time scale of cloud processes, the hindcast ensemble can capture initial transient responses toward equilibrium obtained in the AMIP experiment, which shows a reduction of low clouds over tropical subsidence regions. The reduction of clouds occurs in the first 10 days in TAMIP when the marine boundary layer (MBL) is destabilized because of contrast between fast and slow warming in the MBL and aloft. Enhanced evaporation from the sea surface that should moisten the MBL through turbulent mixing is suppressed by a reduced surface wind speed associated with a slowdown of the Walker circulation. The sign of the low-cloud change over the subsidence regime is thus determined roughly by competition between convective drying and turbulent moistening of the MBL.
 With continuous development of general circulation models (GCMs) over the last few decades, their ability for simulating climate has been improved by increasing model resolution and using more sophisticated parameterization schemes [e.g., Gent et al., 2011; Martin et al., 2011; Watanabe et al., 2010]. However, large uncertainty remains in association with the equilibrium climate sensitivity (ECS), owing greatly to model dependence on cloud-radiative feedbacks to climate change [Andrews et al., 2012].
 Using results of Coupled Model Intercomparison Project phase 5 (CMIP5) models [Taylor et al., 2012], Andrews et al.  showed that change in the shortwave component of the cloud-radiative effect (SWcld) is the major contributor to uncertainty in ECS, as was found in an earlier phase of CMIP3 [e.g., Bony and Dufresne, 2005; Soden et al., 2008; Soden and Vecchi, 2011]. Zelinka et al.  analyzed cloud adjustments and feedbacks in five CMIP5 models. They showed that intermodel spread of net cloud-radiative feedbacks is dominated by varying responses of low clouds rather than high clouds, in which longwave and shortwave cloud feedbacks tend to compensate each other. In particular, the behavior of marine boundary layer (MBL) clouds in the tropical large-scale subsidence area is key for SWcld but is largely uncertain [Bony and Dufresne, 2005; Webb et al., 2006].
 Despite robust increases in lower tropospheric stability and estimated inversion strength that would be expected to cause increases in low-cloud amount [Klein and Hartmann, 1993; Wood and Bretherton, 2006], SWcld feedbacks in the subsidence regimes of CMIP3 models range widely from negative to positive [Webb et al., 2013]. Brient and Bony [2012, 2013] analyzed cloud feedbacks in the IPSL-CM5A climate model, showing that enhanced vertical advection of low moist static energy from the free troposphere to MBL reduces low-cloud amount and hence leads to a positive local SWcld feedback over the subsidence regime. Using a superparameterized climate model, Wyant et al.  found low-cloud increase in response to increase in sea surface temperature (SST). This was attributed to shallow cumulus convection enhanced by destabilization of the MBL caused by increased radiative cooling. Although positive low-cloud feedback has been suggested from observational analysis [Clement et al., 2009], further understanding of physical processes controlling the response of low clouds is required [e.g., Bony et al., 2011].
 To this end, we analyze tropical low-cloud responses within an atmospheric GCM (AGCM), in which a spatially uniform SST perturbation is introduced [Cess et al., 1990; Ringer et al., 2006]. A number of studies have shown that such experiments can be used to evaluate cloud feedbacks in coupled GCMs [Webb and Lock, 2012; Brient and Bony, 2013]. In addition to a long climate simulation, short weather hindcast ensemble experiments are carried out to identify fast cloud responses (“fast response” used here is different from responses to abrupt CO2 increase with surface temperature held fixed.). The latter experiments were originally proposed for improving parameterizations in GCMs [Phillips et al., 2004], but in the present study they are extensively used to understand mechanisms of low-cloud change with the SST increases. Although climate feedbacks, as defined by change in net radiation at the top of the atmosphere per 1 K change in surface air temperature [Gregory et al., 2004], occur over a wide range of time scales from months to centuries [Knutti and Hegerl, 2008], atmospheric processes controlling cloud response to SST increase are expected to have much shorter time scales. Short weather ensemble simulations should thus be useful to examine the fast responses in detail [cf. Kamae and Watanabe, 2012]. Here we propose a mechanism of low-cloud response that is roughly explained by competing local effects of cumulus convection and turbulent mixing, in addition to nonlocal advective effects. Our analysis is based on a single GCM only but could be extended to a multimodel ensemble for understanding the diversity of low-cloud response therein.
2 Model and Experiments
 We use the atmospheric component of the Model for Interdisciplinary Research on Climate version 5 (MIROC5) [Watanabe et al., 2010], which is one of the CMIP5 models. The model has a standard resolution of T85L40. Two sets of AGCM experiments were performed: an Atmospheric Model Intercomparison Project (AMIP) run [Gates et al., 1999] for 30 years from 1979 to 2008 with observed SST and sea-ice concentration and an ensemble of 10 day weather hindcasts extended from the Transpose AMIP II (TAMIP) experiment which calls for 5 day runs [Williams et al., 2012; Xie et al., 2012].
 TAMIP consists of 64 hindcasts, each initialized from the European Centre for Medium-Range Weather Forecasts (ECMWF) analysis of 2008–2009. Since initial conditions are chosen every 30 h from different seasons, the ensemble average can greatly smooth out diurnal and seasonal cycles, which are not the focus in this study. In addition to the above control experiments, both AMIP and TAMIP experiments were repeated with SST uniformly increased by 4 K. Deviations of these sensitivity experiments (called SST + 4 K runs) from the control runs, denoted as Δ, are analyzed. Another experiment performed with patterned SST anomalies added to the control run, based on a composite SST response from CMIP3 future scenario experiments [cf. Taylor et al., 2012], was also analyzed but showed little differences from the SST + 4 K run in terms of radiative feedbacks such as the changes in SWcld, so that results are shown for the SST + 4 K run only.
 The AMIP SST + 4 K experiment shows a positive change in liquid water path (ΔLWP) across the extratropics. In the tropics, ΔLWP is positive over the mean ascending region such as the western equatorial Pacific and negative over continents and subtropical oceans (Figure 1a). Of particular importance is the reduction in LWP across subsidence regions such as off Peru and California, indicating a positive local SWcld feedback (ΔSWcld) over the subtropical cool oceans within MIROC5 while the global mean of ΔSWcld is negative in the coupled atmosphere-ocean model [Watanabe et al., 2012]. Ensemble-mean response of ΔLWP in TAMIP, taken from the last 5 days, demonstrates a similar response at such a short time scale (Figure 1b). The patterns of ΔLWP are similar to those of ΔSWcld except that the sign is reversed, as confirmed by strong negative pattern correlations between the two variables (Figure 1c). This reveals that ΔSWcld is primarily determined by change in cloud liquid amount which roughly corresponds to change in cloud cover (see Figure S1 of the auxiliary material).
 Temporal evolution of ΔLWP can be examined from the TAMIP output sampled every 3 h. With computation of regional averages for the typical convective regime (10°S–10°N, 130°–160°E; red box in Figure 1b) and subsidence regime (170°–140°W, 15°–25°N and 110°–80°W, 15°–25°S; black boxes in Figure 1b), the sign of ΔLWP was initially positive for both regimes, but after day 2 ΔLWP increased over the ascent regime, while it decreased over the subsidence regime (Figure 1d). The tropical average of ΔLWP lies in between and is slightly positive. It is remarkable that ΔLWP in TAMIP on day 10 is very close to that of the AMIP 30 year average. This, however, does not imply that the entire atmosphere is equilibrated within 10 days, as will be discussed later. The ensemble-mean ΔLWP in TAMIP shows three distinct responses: an immediate substantial increase, rapid decrease in about 2 days, and a gradual increase and decrease after day 2 over the ascent and subsidence areas, respectively, the latter referred to as “transient response.” Hereafter, the decrease in LWP over the subsidence area during the transient response will be analyzed.
 Figure 2a shows vertical profiles of changes in relative humidity (ΔRH) and cloud fraction (ΔC) on day 10 across the subsidence areas shown in Figure 1b. The decrease of LWP in this area is associated with decreases of cloud fraction and RH in the lower troposphere roughly at 950 hPa > p > 800 hPa. The decrease of RH in the near-saturated lower troposphere reduces clouds considerably. Since low clouds in the area are formed primarily by large-scale condensation scheme depending on environmental RH, changes in RH tendency terms by each process, averaged over days 5–10 across the subsidence areas, are shown in Figure 2b in order to understand processes that decrease RH. The RH tendency was calculated from temperature (T) and water vapor specific humidity (q) tendencies. Cumulus convection decreases RH everywhere down to about 950 hPa. This is clearly demonstrated by Figure 2c (black bars), in which vertically averaged RH tendencies in the lower troposphere (950 hPa > p > 800 hPa) are shown. Turbulent mixing, which can be taken as proxy for evaporative moistening, acts to increase RH in the MBL in association with enhanced evaporation from the sea surface. Tendency due to cloud physics includes formation and dissipation of cloud such as large-scale condensation and cloud micro physics. This term is positive due to evaporation of cloud. Also, the cloud reduction suppresses radiative cooling, causing negative change of RH tendency by anomalous heating. Dynamics associated with a shoaling of MBL acts to increase (decrease) RH above (below) 900 hPa (Figure 2b), resulting in a net small positive ΔRH tendency (Figure 2c). Therefore, the sign of ΔRH is controlled roughly by the three terms of convective drying and turbulent moistening, and dynamical moistening, among which the first two terms are larger and compensating each other. Turbulent moistening, measured better by the Δq tendency, is gradually suppressed with longer lead time, while the convective drying is relatively maintained (light blue and red curves in Figure 2d). This time evolution explains why the positive ΔRH tendency cannot dominate at day 10 in Figure 2b.
 To clarify how cumulus convection dries the MBL, counteracting the moistening by surface fluxes and turbulence, change in temperature structure is examined further. Figure 3a shows a 10 day evolution of ΔT in the lower troposphere across the subsidence area in TAMIP. Timing of the warming is different between the MBL, the height of which is roughly 850 hPa, and free troposphere; temperature increases much faster in the former than in the latter. Figure 3b shows temperature profiles in the AMIP SST + 4 K run and TAMIP control and SST + 4 K experiments on days 1, 5, and 10. The MBL warms in the subsidence area and is equilibrated in 10 days (purple and black solid curves). However, the temperature profile in the free troposphere is similar to tropical-mean profiles (dashed curves) that roughly follow the moist adiabatic lapse rate, and it warms more slowly than in the MBL. Consequently, a stably stratified atmosphere accompanied by an inversion between the MBL and free troposphere is destabilized initially, and the destabilization in TAMIP SST + 4 K relative to the control run is sustained within 10 days.
 Figure 3c shows temporal evolution of the difference in T between 750 and 900 hPa levels (dashed lines in Figure 3b), which is a simple measure of inversion strength. The inversion weakens rapidly within about 2 days and is then reinforced gradually. The weakened inversion stimulates formation of cumulus convection and leads to the dry MBL over 10 days. The inversion in the AMIP SST + 4 K experiment (triangle in Figure 3c) is stronger than that in the control experiment (diamond in the figure) and does not stimulate additional cumulus convection, as discussed in the following section.
 Different timings of the MBL and free tropospheric warming may result from the disparity in warming processes. The MBL is warmed by enhanced latent and sensible heat fluxes through localized turbulent mixing, whereas temperature in the free troposphere increases adiabatically through large-scale circulation originating from the response in a remote deep convective regime, given a small horizontal temperature gradient in the tropics. In other words, warming above the MBL occurs on a tropical circulation time scale, which is roughly 2 weeks to a month [Jin and Hoskins, 1995].
 In addition to MBL drying from the enhanced cumulus convection, gradual weakening of the turbulent moistening (light blue curve in Figure 2d) is crucial. To understand causes of this gradual weakening, change in evaporation (ΔE) over the Pacific Ocean is depicted in Figure 4. In the AMIP SST + 4 K experiment, ΔE is positive almost everywhere but close to zero in the subsidence regions (Figure 4a). The ΔE in the TAMIP SST + 4 K experiment shows a large uniform increase over the ocean on day 1 (Figure 4b), whereas on day 10 the ΔE pattern becomes zonally nonuniform, similar to the AMIP SST + 4 K experiment (Figure 4c).
E is expressed as E ∝ |V|(qs(SST) − qa), where qs denotes saturation specific humidity at the sea surface as a function of SST, qa is specific humidity at the surface, and |V| is surface wind speed. Therefore, ΔE can be decomposed as , where the overbar signifies mean values in the control experiment and Δ is the difference between the SST + 4 K and control experiments. The first term on the right side, the contribution of change in qs − qa, increases the evaporation over the entire ocean (Figure 4d). The second term, the contribution of change in |V|, is negative over the central-eastern Pacific, including the subsidence regions (Figure 4e). The decrease in surface wind speed is associated with weakening of the tropical circulation [Held and Soden, 2006], which is also seen in the temporal evolution of change in vertical p-velocity at 500 hPa (ω500) averaged over the subsidence area (black curve in Figure 4f). Thus, in simulations in which SSTs are uniformly perturbed, weakening of the atmospheric tropical overturning circulation decreases surface wind speed, suppresses evaporation increase, and contributes to a decrease of MBL clouds in the subsidence regions.
4 Summary and Discussion
 To understand the mechanism of the tropical low-cloud decrease with surface warming over the subsidence regime within our long AGCM simulation, we investigated the fast response in detail using the short weather hindcast experiments based on TAMIP II. Although the 10 day integration of the TAMIP configuration was not sufficient to obtain equilibrium response, the LWP response in the TAMIP SST + 4 K runs can capture that identified in the AMIP SST + 4 K run. This indicates that the major part of the cloud response to increasing SST occurs on this time scale. In contrast to the uniform increases in mid-high latitude LWP, the LWP in the tropical subsidence area decreases in the SST + 4 K runs, resulting in a positive local SWcld feedback. Tendency terms of T, q, and RH from each physical process were analyzed to explore causes of the LWP decrease in the subsidence area. It was found that the decrease in low clouds could be attributable, in a simplified view, to the enhanced cumulus convection that dries the MBL, as well as to the suppression of turbulent moistening caused by tropical circulation weakening. This is shown schematically in Figure S2 of the auxiliary material.
 Since the TAMIP experiments have been initialized from the ECMWF analysis, an initial shock occurring as rapid adjustment processes from the analysis toward the model atmosphere might be included in the response by day 2. However, additional ensemble experiments initialized from the AMIP control experiment performed for 30 days show little difference from the TAMIP experiments (Figure S3). Therefore, artifacts produced by initialization from an alien analysis were not critical for the results presented. We confirmed that the 30 day runs continued to approach the AMIP equilibrium response.
 Given that the large-scale atmospheric circulation does not reach equilibrium within 10 days in the TAMIP SST + 4 K runs (Figure 4f), it is not clear why ΔLWP in TAMIP on day 10 is so close to that in AMIP (Figure 1d). The stronger inversion in the AMIP SST + 4 K run certainly weakens the cumulus convection; nevertheless, cumulus convection still acts to dry the MBL because of a steepened vertical gradient of specific humidity. Drying from cumulus convection is expressed as ∂ tq = M ∂ zq, where M is the cumulus mass flux. The change in specific humidity tendency can be decomposed into contributions of changes in cumulus mass flux and in vertical gradient of specific humidity, i.e., . In the AMIP SST + 4 K run, the second term dominates the first (Figure S4). This is consistent with Brient and Bony [2012, 2013] in that the enhanced vertical gradient of specific humidity tends to dry the MBL. However, we argue that subgrid scale cumulus convection controls the lower-troposphere drying rather than large-scale advection.
 Our model results suggest that, in response to a uniform SST increase, the sign and magnitude of low-cloud response in the subsidence region depend primarily on cumulus and turbulent processes, although the other processes are not negligible. Given that there is still a difference between responses of the TAMIP and AMIP experiments, the fast response does not quantitatively explain the equilibrium response in MIROC5. Also, daily weather disturbances are likely to decrease the signal-to-noise ratio in the TAMIP experiments (Figure 2a), suggesting that a larger ensemble size is desirable. Nonetheless, our results demonstrate that, by performing TAMIP-type experiments in each climate model, uncertainty of the cloud feedback in the multimodel ensemble may be partly attributed to differences in response of specific processes to the SST increase on a short time scale.
 We thank M. Kimoto, M. Satoh, M. Yoshimori, and two anonymous reviewers for their useful comments. This work was supported by the Program for Risk Information on Climate Change (SOUSEI) and Grants-in-Aid 23310014 and 23340137 from the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan.
 The Editor thanks the reviewers for their assistance in evaluating this paper.