Abstract

Uncertainty in tropical rainfall projections under increasing radiative forcing is studied by using 26 models from phase 5 of the Coupled Model Intercomparison Project. Intermodel spread in projected rainfall change generally increases with interactive sea surface temperature (SST) warming in coupled models compared to atmospheric models with a common pattern of prescribed SST increase. Moisture budget analyses reveal that much of the model uncertainty in tropical rainfall projections originates from intermodel discrepancies in the dynamical contribution due to atmospheric circulation change. Intermodel singular value decomposition (SVD) analyses further show a tight coupling between the intermodel variations in SST warming pattern and circulation change in the tropics. In the zonal mean, the first SVD mode features an anomalous interhemispheric Hadley circulation, while the second mode displays an SST peak near the equator. The asymmetric mode is accompanied by a coupled pattern of wind–evaporation–SST feedback in the tropics and is further tied to interhemispheric asymmetric change in extratropical shortwave radiative flux at the top of the atmosphere. Intermodel variability in the tropical circulation change exerts a strong control on the spread in tropical cloud cover change and cloud radiative effects among models. The results indicate that understanding the coupling between the anthropogenic changes in SST pattern and atmospheric circulation holds the key to reducing uncertainties in projections of future changes in tropical rainfall and clouds.

1. Introduction

Hydrological cycle change under global warming is vital to society because of its enormous impacts on, for example, freshwater availability, agriculture, the economy, human health, and biodiversity. Achieving a reliable precipitation projection is a major challenge for the climate modeling community since large uncertainties remain in both the sign and magnitude of regional rainfall change (IPCC 2013; Ma and Xie 2013). Reducing the uncertainty is key for adaptation planning, especially over regions threatened by rainfall change. Therefore, identifying sources that contribute to the intermodel spread in regional precipitation change and quantifying their relative roles is the first step.

Uncertainty in projected precipitation change by climate models is mainly from three sources: model uncertainty, internal variability, and radiative forcing (Tebaldi and Knutti 2007; Hawkins and Sutton 2009; Deser et al. 2012). The present study emphasizes the model uncertainty, which dominates long-term climate change at both global-mean and regional scales (Giorgi and Francisco 2000; Hawkins and Sutton 2009, 2011). The geographic focus is primarily on the tropics where the largest precipitation change is projected (Chou and Neelin 2004; Held and Soden 2006; Seager et al. 2010; Xie et al. 2010).

The model ensemble-mean results show that the spatial pattern of tropical rainfall projection under global warming is governed by the changes in both water vapor and atmospheric circulation. The atmospheric water vapor increases with global warming and results in a positive thermodynamical contribution that strengthens the present precipitation minus evaporation (PE) pattern in the tropics (“wet get wetter”) (Chou and Neelin 2004; Held and Soden 2006; Chou et al. 2009). However, the thermodynamical component alone is not enough to explain the projected rainfall change without considering the dynamical contribution due to the atmospheric circulation change. The slowdown of the tropical circulation under global warming (Held and Soden 2006; Vecchi et al. 2006; Vecchi and Soden 2007; Ma et al. 2012) causes a dynamical contribution that largely offsets the wet get wetter pattern (Seager et al. 2010; Chadwick et al. 2013), leaving the prominence of the SST warming pattern effect.

Beyond the ensemble-mean analyses, several studies have investigated intermodel spread of tropical rainfall change in a warmer climate. The intermodel variations in the SST warming pattern prove important for the differences in tropical precipitation and circulation changes among models from phase 3 of the Coupled Model Intercomparison Project (CMIP3) (Ma and Xie 2013). Moreover, two tropical Pacific SST warming pattern indices, the enhanced equatorial peak and the interhemispheric asymmetry, are found to be useful in describing the intermodel spread of tropical rainfall change (Grose et al. 2014). For regional rainfall change, intermodel spread is much larger in the dynamical component than the thermodynamical component in the tropics (Xie et al. 2015), suggesting that more attention should be given to the model uncertainty in the atmospheric circulation change.

The present study examines the sources and mechanisms of intermodel variability in precipitation projections under representative concentration pathway 4.5 (RCP4.5) among 26 CMIP5 models (Taylor et al. 2012) (Table 1). Model uncertainty, estimated as intermodel standard deviation, in projected rainfall change is large in the tropics. Based on a moisture budget method, changes in PE are decomposed into thermodynamical and dynamical components to quantify the uncertainty from moisture change and circulation change separately. We show that almost all the model uncertainty in tropical rainfall change can be traced back to the dynamical component due to atmospheric circulation change. Our intermodel singular value decomposition (SVD) analysis reveals that the intermodel variations in the SST warming pattern are tightly coupled with the spread of circulation change among models. The underlying mechanisms for the coupled modes of intermodel covariability between the SST warming and circulation change are investigated. We also show that the intermodel spread in tropical circulation change is important in explaining the differences in cloud cover change and cloud radiative effects among models.

Table 1.

Names and expansions of 26 CMIP5 coupled models and 12 available AMIP models (bold) used in this study. Note that near-surface humidity (huss), surface wind (sfcwind), surface wind speed (sfcspd), and clear sky upward shortwave radiation at TOA (rsutcs) outputs are unavailable in some models.

Names and expansions of 26 CMIP5 coupled models and 12 available AMIP models (bold) used in this study. Note that near-surface humidity (huss), surface wind (sfcwind), surface wind speed (sfcspd), and clear sky upward shortwave radiation at TOA (rsutcs) outputs are unavailable in some models.
Names and expansions of 26 CMIP5 coupled models and 12 available AMIP models (bold) used in this study. Note that near-surface humidity (huss), surface wind (sfcwind), surface wind speed (sfcspd), and clear sky upward shortwave radiation at TOA (rsutcs) outputs are unavailable in some models.

The rest of the paper is organized as follows. Section 2 describes the data and methods used in this study. Section 3 presents the CMIP5 multimodel ensemble-mean change under global warming. Section 4 quantifies sources of model uncertainty in projected precipitation change. Section 5 investigates the dominant modes of intermodel covariability between the SST warming pattern and circulation change in the zonal mean and on the equator. Section 6 further examines regional modes of intermodel variance in three tropical basins and mechanisms for the formation of the modes of intermodel variability in the SST warming pattern. Section 7 is a summary.

2. Data and methodology

a. CMIP5

The monthly outputs from 26 CMIP5 models in historical and RCP4.5 simulations are utilized to evaluate projections under increasing radiative forcing. The projected climate change (denoted as ) is calculated by subtracting the 1950–99 mean from the 2050–99 mean. Model uncertainty refers to intermodel spread due to model structure and representations of internal variability. We use the long simulations of preindustrial control runs to estimate the contribution to model uncertainty from internal variability (see methodology in Long and Xie 2015). In addition, differences between 30-year averages in two simulations from the Atmospheric Model Intercomparison Program (AMIP; 12 models available) are used to evaluate the model uncertainty in rainfall response to a prescribed SST increase: one is the AMIP control run (1979–2008) forced with the observed monthly mean SST and sea ice concentration, and the other is forced with the same boundary conditions as the AMIP control run plus a constant spatially uniform SST warming of 4 K (AMIP4K). All model outputs are interpolated onto a common grid of 2.5° × 2.5°, and only the first-member run (r1i1p1) of each model is analyzed to ensure equal weight in model uncertainty analyses. Note that some variables are not available in some models (Table 1). All changes in a warmer climate are normalized by the tropical mean (20°S–20°N) SST warming in each model to highlight the intermodel variations in spatial pattern.

b. Moisture budget

On monthly and longer time scales, we use the moisture budget equation (Trenberth and Guillemot 1995; Seager et al. 2010):

 
formula

where PE, , , , and are precipitation minus evaporation, the density of water, specific humidity, the horizontal vector wind, and pressure, respectively. The subscript denotes the surface value. The first term on the right-hand side is horizontal moisture advection by the monthly mean circulation, the second term is vertical moisture advection due to wind convergence, and the residual is largely a result of the transient eddy effect.

Based on Eq. (1), long-term precipitation change can be approximated as follows:

 
formula

The first three terms on the right-hand side are changes in evaporation, horizontal moisture advection, and vertical moisture advection, respectively. In the tropics, the vertical moisture advection term dominates precipitation change while the other three terms are small (Seager et al. 2010). In particular, the vertical moisture advection term can be further decomposed into a thermodynamical component due to changes in without changes in and a dynamical component due to changes in without changes in :

 
formula
 
formula

By making use of the continuity equation and boundary conditions that vertical velocity at the surface and the top of the atmosphere (TOA) is zero, we approximate Eqs. (3) and (4) as follows:

 
formula
 
formula

where is the pressure velocity at the 500-hPa pressure level and is surface specific humidity. The product of and is a good approximation for tropical precipitation (Held and Soden 2006; Huang et al. 2013; Huang 2014).

c. Cross-equatorial atmospheric energy transport

Cross-equatorial atmospheric energy transport is calculated by integrating the atmospheric energy budget (Frierson and Hwang 2012):

 
formula

Here , ϕ, , and are the cross-equatorial atmospheric energy transport, latitude, longitude, and radius of Earth, respectively. The term is net energy flux entering into the atmosphere and the sum of the downward net radiative flux at the TOA and the upward net heat flux at the surface. Particularly, the net radiative forcing can be decomposed into noncloud and cloud radiative components. The noncloud radiative component results from factors other than cloud (e.g., water vapor, CO2, ozone, and aerosols), while the cloud radiative component is due to cloud effect and defined as the radiative flux in all sky minus clear sky.

3. Multimodel ensemble-mean change

We first present the projected annual-mean precipitation change in RCP4.5 simulations (Fig. 1a) as a context for the model uncertainty analyses. The ensemble-mean precipitation change varies strongly in space and features increased precipitation in most of the tropics as well as intensified subtropical dryness and enhanced moistening in the mid- to high latitudes, similar to the result in CMIP3 (Seager et al. 2010). Here we focus on the tropics where the rainfall response to global warming is large and ocean–atmosphere interaction is important. The annual-mean precipitation change mainly follows the SST warming pattern (Xie et al. 2010; Chadwick et al. 2013), with positive (negative) rainfall change (Fig. 1a, color shading) over regions where the SST increase is larger (smaller) than the tropical mean warming (Fig. 1a, color contours). Indeed, the percentage precipitation change shows a high spatial correlation (0.56) with SST deviations from the tropical mean, but the correlation is nearly zero with the precipitation climatology in the tropics, confirming the importance of the SST warming pattern. Moisture budget analyses clearly show that vertical moisture advection change is dominant in the tropics (Fig. 1c), along with weak evaporation change, relatively small horizontal advection change that intensifies the subtropical drying, and negligible transient eddy effect. The vertical moisture advection term is further decomposed into thermodynamical and dynamical components based on Eqs. (3) and (4) to diagnose contributions from circulation change and moisture change separately. It is evident that the thermodynamical component shows a wet get wetter pattern (Fig. 2a), consistent with the theoretical prediction from the Clapeyron–Clausius equation (Held and Soden 2006). However, the dynamical component causes precipitation to decrease in the climatological ascent region that largely cancels the wet get wetter effect (Fig. 2b). As a result, the tropical rainfall change mainly follows the “warmer get wetter” pattern (Chadwick et al. 2013). Based on Eqs. (5) and (6), () can be further estimated with (). Figures 2a–d show that the spatial distributions of and resemble the patterns of and in the tropics, respectively. This is because surface specific humidity and its change are of broad spatial distribution over the tropical oceans (Figs. 2e,f). Here we multiply the pressure velocity by −1 to make it positive upward, for easy comparison with precipitation change. In particular, the pattern of shows a clear weakening of the tropical circulation (Figs. 2c,d), causing the large cancellation between and (Chadwick et al. 2013) and raising the importance of the atmospheric circulation change for the tropical rainfall change pattern.

Fig. 1.

CMIP5 ensemble-annual-mean change in (a) precipitation (color shading; mm day−1) along with SST deviations from the tropical mean [red and green contours; °C; contour interval is 0.1°C], (b) evaporation , (c) vertical moisture advection , (d) horizontal moisture advection , and (e) residual in RCP4.5 simulations. The black contours indicate climatological precipitation value at 4 mm day−1 in all panels.

Fig. 1.

CMIP5 ensemble-annual-mean change in (a) precipitation (color shading; mm day−1) along with SST deviations from the tropical mean [red and green contours; °C; contour interval is 0.1°C], (b) evaporation , (c) vertical moisture advection , (d) horizontal moisture advection , and (e) residual in RCP4.5 simulations. The black contours indicate climatological precipitation value at 4 mm day−1 in all panels.

Fig. 2.

CMIP5 ensemble-annual-mean (a) thermodynamical component and (b) dynamical component of vertical moisture advection change. Climatology and future change in (c),(d) pressure velocity at 500 hPa (hPa day−1), and (e),(f) surface specific humidity (g kg−1). The black contours indicate climatological precipitation value at 4 mm day−1.

Fig. 2.

CMIP5 ensemble-annual-mean (a) thermodynamical component and (b) dynamical component of vertical moisture advection change. Climatology and future change in (c),(d) pressure velocity at 500 hPa (hPa day−1), and (e),(f) surface specific humidity (g kg−1). The black contours indicate climatological precipitation value at 4 mm day−1.

4. Sources of model uncertainty in precipitation projections

Model uncertainty in precipitation change is estimated as intermodel standard deviation (here denotes the standard deviation, and the prime denotes the deviations from the ensemble-mean change). In atmospheric model simulations with prescribed spatial uniform warming of 4 K (AMIP4K), intermodel spread in rainfall change maximizes in the tropical oceans, especially over the climatological rainy regions (Fig. 3a). In coupled model simulations (RCP4.5), the ensemble-mean SST warming displays distinctive spatial structures (Fig. 1a, contours) and exerts a strong control on the rainfall change pattern. Therefore, intermodel spread in the SST warming pattern leads to substantial differences in rainfall change among models. This is equivalent to the idea that ocean–atmosphere feedbacks amplify the intermodel variability in rainfall change in the coupled models. As a result, model uncertainty in rainfall change increases in magnitude over most regions in RCP4.5 compared to AMIP4K, along with the broadening of the large uncertainty region as indicated by the 0.25 mm day−1 contour (Fig. 3b). To compare the uncertainty magnitudes between the two simulations, we calculate the RCP4.5-to-AMIP4K ratio of intermodel standard deviations. Figure 3c shows that the ratio is larger than one (larger uncertainty in RCP4.5 than AMIP4K) over most regions and is especially large over the subtropical regions in the Pacific and Atlantic Oceans because of large uncertainty in the cross-equatorial SST warming gradient in RCP4.5. Additional bands of large uncertainty ratio are found in the western boundary current regions, such as the Gulf Stream and Agulhas Current, where the intermodel spread in the local SST warming pattern is also important for rainfall change (Long and Xie 2015). In contrast, model uncertainty is significantly reduced in the coupled model simulations compared to the atmospheric model simulations over the northern Indian Ocean, western Pacific, eastern Pacific coastal regions, western Atlantic Ocean, and most regions in the midlatitudes. This result implies the importance of ocean–atmosphere coupling in rainfall projections.

Fig. 3.

Intermodel standard deviation of annual-mean rainfall change in (a) AMIP4K and (b) RCP4.5. The blue contours indicate at 0.25 mm day−1 in AMIP4K simulations. (c) The RCP4.5-to-AMIP4K ratio of , with black contours indicating a ratio equal to one. Note that rainfall change is normalized by the tropical mean (20°S–20°N) warming, which is 4 K for AMIP4K simulations, in each model.

Fig. 3.

Intermodel standard deviation of annual-mean rainfall change in (a) AMIP4K and (b) RCP4.5. The blue contours indicate at 0.25 mm day−1 in AMIP4K simulations. (c) The RCP4.5-to-AMIP4K ratio of , with black contours indicating a ratio equal to one. Note that rainfall change is normalized by the tropical mean (20°S–20°N) warming, which is 4 K for AMIP4K simulations, in each model.

To illustrate the role of model uncertainty in rainfall projections, we calculate the signal-to-noise ratio of annual-mean precipitation change (Fig. 4), defined as the absolute value of ensemble-mean change divided by the intermodel standard deviation . Ratios larger (smaller) than one indicate relatively small (large) intermodel spread compared to the ensemble-mean change and hence leave high (low) confidence in the rainfall projections. In the tropics, this ratio is small despite relatively high model agreement in the sign of rainfall change, falling below 0.2 over regions where the sign consistency is low. Indeed, the area-averaged signal-to-noise ratio is only 0.6 in the tropics, indicating the low confidence in rainfall projection due to the large model uncertainty. Note that both the signal-to-noise ratio and the sign consistency in rainfall change among models are high in the midlatitudes despite large internal variability (Deser et al. 2012), especially over the Northern Hemisphere (NH) continents and the Southern Ocean.

Fig. 4.

The signal-to-noise ratio in RCP4.5 simulations, with black contours indicating a ratio equal to one. Grid points are dotted where at least 18 models agree on the sign of the ensemble-mean change.

Fig. 4.

The signal-to-noise ratio in RCP4.5 simulations, with black contours indicating a ratio equal to one. Grid points are dotted where at least 18 models agree on the sign of the ensemble-mean change.

Figure 5 displays intermodel standard deviations of annual-mean rainfall change and individual contributions based on Eq. (2) to quantify the sources of the model uncertainty. Intermodel spread in rainfall change (Fig. 5a) is large in the tropics and small over most regions in the mid- to high latitudes but locally enhanced over western boundary current regions where evaporation change and transient eddy effects are important (Long and Xie 2015). In the tropics, model uncertainty in rainfall change is dominated by the vertical moisture advection term (Fig. 5c) with small contributions from the other three terms. Furthermore, intermodel spread is small in but large in (Figs. 6a,b), in contrast to the case of the ensemble-mean change in which they are comparable in magnitude. According to Eqs. (5) and (6), uncertainty in the thermodynamical and dynamical components of rainfall change can be caused by uncertainty in either pressure velocity or surface specific humidity, or both of them. Figures 6c and 6d show that and patterns dominate the spatial structures of and , respectively, in the model uncertainty analyses because of small spatial variations in intermodel standard deviations of and (Figs. 6e,f). The large difference between the magnitude of uncertainty in and can be explained by the relatively high consistency in climatological circulation because all models are designed to simulate the observations, but the atmospheric circulation change is poorly constrained (Xie et al. 2015). From the above step-by-step analyses, we conclude that almost all of the model uncertainty in annual-mean rainfall change originates from intermodel spread in the change of vertical moisture advection, which can be traced back to the differences in the atmospheric circulation change among models. Therefore, investigations into the intermodel variability in circulation change are essential to achieve a more credible tropical rainfall projection.

Fig. 5.

Plot of intermodel standard deviation of change in (a) precipitation, (b) evaporation, (c) vertical moisture advection, (d) horizontal moisture advection, and (e) residual. Black contours indicate the intermodel standard deviation of rainfall change at 0.25 mm day−1.

Fig. 5.

Plot of intermodel standard deviation of change in (a) precipitation, (b) evaporation, (c) vertical moisture advection, (d) horizontal moisture advection, and (e) residual. Black contours indicate the intermodel standard deviation of rainfall change at 0.25 mm day−1.

Fig. 6.

Plot of intermodel standard deviation in (a) thermodynamical component and (b) dynamical component of vertical moisture advection change; climatological and projected future change in (c),(d) pressure velocity at 500 hPa (hPa day−1) and (e),(f) surface specific humidity . Black contours indicate the intermodel standard deviation of rainfall change at 0.25 mm day−1 (Fig. 5a). Stippling indicates where the model uncertainty in circulation change is at least twice the magnitude of the uncertainty due to internal variability.

Fig. 6.

Plot of intermodel standard deviation in (a) thermodynamical component and (b) dynamical component of vertical moisture advection change; climatological and projected future change in (c),(d) pressure velocity at 500 hPa (hPa day−1) and (e),(f) surface specific humidity . Black contours indicate the intermodel standard deviation of rainfall change at 0.25 mm day−1 (Fig. 5a). Stippling indicates where the model uncertainty in circulation change is at least twice the magnitude of the uncertainty due to internal variability.

Intermodel spread in simulating internal variability is also a part of the model uncertainty in addition to intermodel differences in modeling structure. In most of the tropics, the intermodel variance in circulation change is at least twice of that resulting from internal variability (Fig. 6d, dotted grid points). Thus the contribution from internal variability is small, consistent with analyses of precipitation change for the annual mean (Rowell 2012) and seasonal mean (Kent et al. 2015). The effect of the intermodel spread in internal variability is only important in the mid- to high latitudes where atmospheric internal variability results in large uncertainty (Deser et al. 2012).

5. Zonal-mean and equatorial modes of intermodel variance

The above analyses identify the atmospheric circulation change as the dominant source of model uncertainty in tropical rainfall projections. Therefore, it is essential to understand the underlying physical mechanisms that drive the intermodel spread in tropical circulation change. Given the important role of the SST warming pattern uncertainty in intermodel variations in tropical circulation change in CMIP3 (Ma and Xie 2013), we examine the dominant modes of intermodel covariability between and by conducting the intermodel SVD analyses in CMIP5 models. The prime indicates the intermodel differences from the ensemble-mean change.

a. Zonal-mean modes of intermodel variance

For zonal-mean projections, the first two SVD modes of and show a meridional dipole mode with interhemispheric asymmetry and an equatorial peak mode with the largest response near the equator (Figs. 7a,b). These two modes explain 54.1% and 86% of the intermodel variances in and , respectively, indicating that not all of the intermodel variations in circulation change are due to those in SST warming pattern. The and values display high spatial correlations for the first two SVD modes at 0.7 and 0.74, respectively, confirming the physical significance of their relationship. Figures 7c and 7d show regressions of the Hadley cell change, represented by zonal-mean streamfunction change, against the principal components (PCs) of the SVD modes. The first regressed pattern displays a significant northward shift in the Hadley circulation as part of the tropical asymmetric mode, resulting in the strengthening (weakening) of the southern (northern) cell, while the second regressed pattern shows an intensified equatorial part and weakened polarward part of the Hadley cell. These results illustrate a tight coupling between the intermodel variations in SST warming pattern and the Hadley circulation change.

Fig. 7.

(a),(b) The first two modes of intermodel SVD analysis between tropical zonal-mean and in RCP4.5 simulations, with the explained variances and spatial correlations between patterns of and marked in each panel. (c),(d) Regressions of zonal-mean streamfunction change (1010 kg s−1) against the PCs of corresponding SVD modes along with the climatological streamfunction (contours; contour interval is 2 × 1010 kg s−1). Stippling indicates where the regression passes a t test at the 95% significance level.

Fig. 7.

(a),(b) The first two modes of intermodel SVD analysis between tropical zonal-mean and in RCP4.5 simulations, with the explained variances and spatial correlations between patterns of and marked in each panel. (c),(d) Regressions of zonal-mean streamfunction change (1010 kg s−1) against the PCs of corresponding SVD modes along with the climatological streamfunction (contours; contour interval is 2 × 1010 kg s−1). Stippling indicates where the regression passes a t test at the 95% significance level.

Based on an energy constraint argument, tropical precipitation shifts to the hemisphere receiving more heat at the TOA or surface to reach a balance with the anomalous atmospheric energy transport (Kang et al. 2008, 2009). Therefore, it is conceivable that intermodel differences in interhemispheric atmosphere energy flux will lead to intermodel spread of tropical asymmetry in precipitation and circulation changes. Figure 8 shows the regressions of NH minus Southern Hemisphere (SH) differences in energy fluxes entering into the atmosphere (i.e., the downward TOA radiative fluxes and upward surface heat fluxes) against the PC of the tropical asymmetric mode. The regressed downward net radiative flux change at the TOA shows a broad positive asymmetric pattern in 0°–60°, while the regression of upward net surface heat flux change is not significant at most latitudes (Fig. 8a). This suggests that much of the intermodel spread in the atmospheric energy transport change originates from the TOA radiation change. Therefore, we further examine the shortwave and longwave components of to trace the sources of the TOA radiation uncertainty. In the tropics, the regressed TOA longwave radiative flux change (Fig. 8b, blue line) displays significant positive asymmetry, implying a positive longwave radiation feedback to the tropical asymmetric mode. This positive feedback may be viewed as a combination of the longwave cloud feedback to the positive interhemispheric asymmetry in (Su et al. 2014) and the water vapor feedback (Soden and Held 2006) due to larger SST warming in the northern tropics than the southern tropics. However, the negative regressed TOA shortwave radiative flux change (Fig. 8b, red line) to some extent offsets the longwave feedback effect, reducing the positive asymmetric response in . In contrast, shows large positive regressions in 20°–60° that overwhelm the negative contribution from , leaving a broad positive asymmetry in in the extratropics. Note that the interhemispheric asymmetry in the regressed is not significant in 60°–90° because of the large cancellation between the cloud and noncloud shortwave components (not shown).

Fig. 8.

Regressions of (a) downward net radiative flux change at the TOA (red line) and upward net heat flux change at the surface (blue line), (b) TOA shortwave radiative flux change (red line), and TOA longwave radiative flux change (blue line) against the PC of tropical asymmetric mode in Fig. 7a. Cross marks indicate where the regression passes a t test at the 95% significance level.

Fig. 8.

Regressions of (a) downward net radiative flux change at the TOA (red line) and upward net heat flux change at the surface (blue line), (b) TOA shortwave radiative flux change (red line), and TOA longwave radiative flux change (blue line) against the PC of tropical asymmetric mode in Fig. 7a. Cross marks indicate where the regression passes a t test at the 95% significance level.

To better illustrate the underlying mechanisms for the intermodel spread of the tropical asymmetric pattern, we investigate the relationships between asymmetry indices of different variables (Fig. 9). The tropical asymmetry index is calculated as the area-weighted differences between the northern tropics (0°–20°N) and southern tropics (0°–20°S), while the extratropical asymmetry index is computed by the area-weighted differences between the NH extratropics (20°–60°N) and SH extratropics (20°–60°S). The tight coupling between the interhemispheric SST warming gradient and circulation change in the tropics can be seen from the significant positive correlation between their asymmetric indices among models (Fig. 9a; R = 0.75).

Fig. 9.

Scatterplot of indices in 26 models for (a) tropical asymmetry in (°C) vs that in (hPa day−1), (b) tropical asymmetry in vs cross-eq (W m−2), (c) latitude change of the Hadley cell center vs cross-eq , (d) tropical asymmetry in vs that in WES feedback effect (W m−2), (e) extratropical asymmetry in (W m−2) vs cross-eq , and (f) extratropical asymmetry in vs that in total cloud cover change (%). Correlation coefficient between indices among models is marked in each panel. Particularly, the correlation coefficients of cross-eq with and are also marked in (e). Each point is colored according to the magnitude of tropical asymmetry in SST warming.

Fig. 9.

Scatterplot of indices in 26 models for (a) tropical asymmetry in (°C) vs that in (hPa day−1), (b) tropical asymmetry in vs cross-eq (W m−2), (c) latitude change of the Hadley cell center vs cross-eq , (d) tropical asymmetry in vs that in WES feedback effect (W m−2), (e) extratropical asymmetry in (W m−2) vs cross-eq , and (f) extratropical asymmetry in vs that in total cloud cover change (%). Correlation coefficient between indices among models is marked in each panel. Particularly, the correlation coefficients of cross-eq with and are also marked in (e). Each point is colored according to the magnitude of tropical asymmetry in SST warming.

Indeed, the intermodel spread of the tropical asymmetric SST warming is negatively correlated with differences in cross-equatorial atmospheric energy transport change (cross-eq ) among models (Fig. 9c; R = −0.73), with models that have larger tropical asymmetry in SST warming showing a stronger southward cross-eq . Furthermore, the spread in cross-eq among models is tied to the intermodel discrepancies in the Hadley cell shift (Fig. 9d; R = −0.55) because the upper branches of the Hadley cells play an important role in the atmospheric energy transport (Held 2001; Hwang and Frierson 2013). Here the Hadley cell center is defined as the latitude of the zero crossing of the zonal-mean streamfunction at the 500-hPa level between two Hadley cells. The shift of the Hadley cell influences the trade wind change, which triggers a wind–evaporation–SST (WES) feedback that is important for the tropical asymmetric warming pattern (Xie et al. 2010). Consequently, intermodel differences in the WES feedback effect (Xie and Philander 1994) are closely related to the tropical asymmetry in SST warming among models (Fig. 9b; R = 0.86). Models showing larger asymmetry in SST warming tend to have less latent heat release from the surface in the northern tropics than the southern tropics (larger asymmetry in ). Here the WES feedback is estimated as the latent heat change due to wind speed change [, where , , and are climatological latent heat, climatological wind speed, and wind speed change, respectively].

Figure 8 reveals that the intermodel spread of interhemispheric asymmetry in atmospheric energy flux mostly comes from the extratropical TOA , especially in 20°–60°. For the annual mean, energy fluctuations in the atmosphere from the TOA or surface are balanced by anomalous atmospheric energy transport (Donohoe and Battisti 2012; Zelinka and Hartmann 2012). As a result, models that have larger positive extratropical asymmetry in TOA tend to display a stronger southward cross-eq (Fig. 9e), with a negative intermodel correlation (−0.7) between their indices. This negative correlation with at the equator is also evident in both the cloud and noncloud components of TOA . Particularly, intermodel spread of the extratropical asymmetry in total cloud cover change explains most of the TOA variability (Fig. 9f; R = −0.63).

The above analyses reveal that the extratropical TOA is closely correlated with the coupled pattern between the interhemispheric SST warming gradient and the Hadley circulation change, suggesting that intermodel spread in extratropical TOA remotely affects that in the tropical asymmetric pattern via anomalous atmospheric energy transport. The WES feedback and Hadley circulation change help mediate the tropical asymmetric adjustments to the extratropical radiative forcing, consistent with a previous study of model biases in climatological precipitation distribution (Li and Xie 2014).

b. Equatorial modes of intermodel variance

The Walker circulation is an important part of the tropical circulation and the ocean–atmosphere coupling system. Here we use equatorial (5°S–5°N mean) as an index of the Walker circulation change and focus on the Indo-Pacific sector to show intermodel covariability between and in the zonal direction. Modes of intermodel covariance in equatorial and are examined by conducting the intermodel SVD analysis between them. The first SVD mode (SVD1) shows an east–west dipole-like SST pattern associated with a circulation slowdown in the Indian Ocean and a central Pacific El Niño–like SST pattern tied to an eastward shift of the Walker circulation in the Pacific Ocean (Fig. 10a). The second SVD mode (SVD2) is more complicated, strengthening (weakening) the Walker circulation west (east) of 120°E (Fig. 10b). The first two SVD modes combined explain 63% of the intermodel variance in circulation change and show high spatial correlations (0.65 and 0.6, respectively) between and , confirming the importance of the SST warming pattern in circulation change.

Fig. 10.

(a),(b)The first two modes of intermodel SVD analysis between equatorial (5°S–5°N) (red line) and (blue line) in RCP4.5 simulations, with the explained variances and spatial correlations between patterns of and marked in each panel. The climatological pressure velocity at 500 hPa is also plotted for comparison. (c),(d) Regressions of changes in equatorial total cloud cover (%; blue line) and TOA cloud shortwave (red solid line) and longwave (red dashed line) radiative fluxes onto the PCs of SVD modes. Cross marks indicate grid points where the regression passes a t test at the 95% significance level.

Fig. 10.

(a),(b)The first two modes of intermodel SVD analysis between equatorial (5°S–5°N) (red line) and (blue line) in RCP4.5 simulations, with the explained variances and spatial correlations between patterns of and marked in each panel. The climatological pressure velocity at 500 hPa is also plotted for comparison. (c),(d) Regressions of changes in equatorial total cloud cover (%; blue line) and TOA cloud shortwave (red solid line) and longwave (red dashed line) radiative fluxes onto the PCs of SVD modes. Cross marks indicate grid points where the regression passes a t test at the 95% significance level.

The distributions of cloud are directly related to large-scale vertical circulation (Hartmann and Michelsen 1993; Bony et al. 2004; Bony and Dufresne 2005; Su et al. 2008) and hence the cloud radiative effects (Su et al. 2014). To investigate the influence of intermodel variability in circulation change on the cloud distribution and cloud radiative effects, we regress changes in equatorial total cloud cover and cloud radiative fluxes against the PCs of equatorial SVD modes (Figs. 10c,d). The regressed patterns of (blue line) resemble the patterns of for both SVD1 and SVD2, with high spatial correlations between them (Table 2). Moreover, the regressions of and are opposite in most longitudes as a direct effect of the cloud cover change. Therefore, intermodel differences in the coupling between the SST warming and circulation change exert a strong control on the spread in cloud cover change and cloud radiative effects among models. This is of great importance because the cloud feedback is one of the leading sources of the uncertainty in the equilibrium climate sensitivity (IPCC 2013).

Table 2.

Spatial correlations between the SVD modes of and the regressed patterns of the total cloud cover change and changes in cloud shortwave radiative flux and cloud longwave radiative flux in the intermodel variability analyses conducted on the equator and in three tropical basins.

Spatial correlations between the SVD modes of  and the regressed patterns of the total cloud cover change  and changes in cloud shortwave radiative flux  and cloud longwave radiative flux  in the intermodel variability analyses conducted on the equator and in three tropical basins.
Spatial correlations between the SVD modes of  and the regressed patterns of the total cloud cover change  and changes in cloud shortwave radiative flux  and cloud longwave radiative flux  in the intermodel variability analyses conducted on the equator and in three tropical basins.

6. Regional modes of intermodel variance

Now we further investigate the intermodel covariability between and in three tropical basins separately. In SVD1, a pattern of reduced SST warming is dominant in the Indian Ocean and associated with a broad band of negative in the SH. In addition, the enhanced equatorial warming is strong in the Pacific Ocean but weak in the Atlantic Ocean (Fig. 11a). For the SST pattern of SVD2, both the Indian Ocean and Pacific Ocean show prominent east–west dipole-like structures associated with the slowdown of the Walker circulation (Fig. 11b). Moreover, the interhemispheric asymmetric pattern is evident in the Atlantic Ocean in SVD2. The intermodel variances in circulation change explained by the first two SVD modes combined are 53.1%, 34.8%, and 38.2% in the Indian, Pacific, and Atlantic Oceans, respectively, illustrating the interbasin differences of the spread in the SST and circulation coupling among models. Spatial correlations between and patterns (Figs. 11c–e) in the first 10 SVD modes are very high in a few modes in the Indian and Atlantic Oceans. This suggests that uncertainty is tightly coupled between the SST warming pattern and circulation change, and the coupling can be well represented by a few leading modes in these two basins. However, the spatial correlations are relatively low for the Pacific Ocean modes, indicating diverse sources of SST pattern uncertainty. Most of the SST patterns in the first two SVD modes resemble the spatial structures of the ensemble-mean SST warming, such as the equatorial peak, the east–west dipole-like SST pattern, and the interhemispheric asymmetric mode. The regional modes of intermodel variance exert strong control on cloud and cloud radiative effects, with displaying high spatial correlations with the regressions of and the cloud radiative effects against the PCs of basin SVD modes (Table 2). As an exception, the regressed and display nearly zero correlation with the pattern in SVD2 in the Atlantic Ocean.

Fig. 11.

(a),(b) The first two modes of intermodel SVD analysis in three ocean basins between (color shading) and (contours; contour interval is 0.025), with the explained variances for (red) and (black) marked on the corresponding continent. (c)–(e) Spatial correlations between the patterns of and (blue line) and the explained intermodel variances for (red line) in the first 10 modes in the Indian, Pacific, and Atlantic Oceans, respectively.

Fig. 11.

(a),(b) The first two modes of intermodel SVD analysis in three ocean basins between (color shading) and (contours; contour interval is 0.025), with the explained variances for (red) and (black) marked on the corresponding continent. (c)–(e) Spatial correlations between the patterns of and (blue line) and the explained intermodel variances for (red line) in the first 10 modes in the Indian, Pacific, and Atlantic Oceans, respectively.

Intermodel differences in the SST warming pattern are a key issue in improving the model agreement in tropical circulation change because of their tight coupling. Therefore, underlying mechanisms for the modes of intermodel variance in SST warming pattern need to be investigated. Here we focus on the roles of the WES feedback effect () and ocean heat transport (, where is the surface downward net heat flux into the ocean), which are important for the ensemble-mean SST warming pattern (Xie et al. 2010; Lu and Zhao 2012; Long et al. 2014; Liu et al. 2015). Figure 12 shows the regressions of and against PCs of the SST modes in the corresponding basins. In SVD1, the regressed pattern of shows a positive contribution to the SST pattern in most regions, with increased (decreased) wind speed corresponding to a reduced (enhanced) SST warming (Fig. 12a). The effect of the ocean heat transport is only evident in the Pacific Ocean but negligible or even opposite to the SST patterns in the other two basins. In SVD2, the WES feedback effect only contributes to the SST pattern near the Sumatra coast, while is important for the basinwide east–west dipole SST pattern in the Indian Ocean (Figs. 12c,d). Moreover, both and contribute to the zonal dipole pattern in the Pacific Ocean, while the SST pattern in the Atlantic Ocean is dominated by . The above results suggest that it is imperative to improve the model agreement in the simulations of the surface wind change and ocean dynamical processes toward reducing the regional SST pattern uncertainty.

Fig. 12.

The first two SVD modes of (color shading) along with regressions of (a),(c) WES feedback effect (contours; contour interval is 1 W m−2) and surface wind change (green vectors) and (b),(d) ocean heat transport (; contour interval is 1 W m−2) onto PCs of the corresponding SVD modes in Fig. 10. Zero contours omitted for clarity. Stippling indicates where the regression passes a t test at the 95% significance level.

Fig. 12.

The first two SVD modes of (color shading) along with regressions of (a),(c) WES feedback effect (contours; contour interval is 1 W m−2) and surface wind change (green vectors) and (b),(d) ocean heat transport (; contour interval is 1 W m−2) onto PCs of the corresponding SVD modes in Fig. 10. Zero contours omitted for clarity. Stippling indicates where the regression passes a t test at the 95% significance level.

7. Summary

We have investigated the dominant sources and underlying mechanisms for uncertainty in projected rainfall change in a warmer climate based on 26 CMIP5 models. In the tropics, intermodel standard deviation of precipitation change is considerably large, comparable to or even exceeding the ensemble-mean change. This causes low confidence in rainfall projections from climate models. The vertical moisture advection change is dominant in both the ensemble-mean and intermodel spread of rainfall projections. We further decompose this term into thermodynamical and dynamical components to isolate the effects of moisture change and circulation change. Intermodel spread in the dynamical component is much larger than that in the thermodynamical component, although they are comparable in magnitude for the ensemble-mean change. Furthermore, model uncertainty in the dynamical contribution can be traced back to the intermodel differences in atmospheric circulation change because of high consistency in moisture change among models in the tropical oceans. Therefore, understanding the effect of the atmospheric circulation change is an important step toward improving consistency in tropical rainfall projections among models (Xie et al. 2015). Moreover, the intermodel variability in circulation change exerts a strong control on the spread of cloud cover change and cloud radiative effects among models in the tropics. One important implication is that improving the model representation of key processes to the tropical circulation change may significantly narrow the intermodel spread in tropical cloud change and hence the estimation of the equilibrium climate sensitivity.

Model uncertainty in projected rainfall change generally increases in the coupled model simulations with variations in the SST warming pattern compared to atmospheric model simulations with prescribed SST increase, suggesting the important role of the SST warming pattern uncertainty in the spread of rainfall projections among models (Chadwick 2016). The intermodel SVD analyses reveal two dominant modes of intermodel covariance in zonal-mean SST warming and circulation change. One is an interhemispheric dipole mode associated with the meridional shift of the Hadley cell, and the other is an equatorial peak mode along with the strengthening (weakening) in the equatorial (poleward) part of both the northern and southern Hadley cells. This is similar to the results from the analyses in CMIP3 models (Ma and Xie 2013). Moreover, the analyses between equatorial and also confirm the tight coupling between the tropical SST warming pattern and circulation change.

The energy analyses reveal that intermodel spread in extratropical interhemispheric asymmetry in TOA shortwave radiative flux change is important for the tropical asymmetric mode of intermodel covariability between SST warming and circulation change. The intermodel differences in extratropical radiative forcing are suggested to remotely affect the tropical asymmetric patterns through anomalous atmospheric energy transport. The WES feedback and Hadley circulation change are important processes involved in the tropical asymmetric adjustments. This is consistent with the theoretical prediction that extratropical thermal forcing is important for the shift of the intertropical convergence zone (Kang et al. 2008, 2009). At regional scales, the WES feedback effect and to a lesser extent the ocean heat transport are important for the SST modes in the intermodel SVD analyses, similar to the results of ensemble-mean change.

Our results imply that reducing the uncertainty in the atmospheric circulation change holds the promise of improving the confidence in rainfall projections. For that, further systematic investigations are needed into the mechanisms for intermodel spread in the tropical SST warming pattern and extratropical radiative flux change. Sensitivity experiments are also needed to quantify other factors contributing to the uncertainty in rainfall projections, including model biases in the climatology (Zhou and Xie 2015), convective schemes, and ocean dynamical processes.

Acknowledgments

We wish to thank X.-T. Zheng and Y.-T. Hwang for helpful discussions. This work is supported by the National Basic Research Program of China (2012CB955602), the NSFC-Shandong Joint Fund for Marine Science Research Centers (U1406401), the China Scholarship Council (201406330004), and the U.S. National Science Foundation. We acknowledge the WCRP Working Group on Coupled Modelling, which is responsible for CMIP, and the climate modeling groups for producing and making available the model outputs.

REFERENCES

REFERENCES
Bony
,
S.
, and
J. L.
Dufresne
,
2005
:
Marine boundary layer clouds at the heart of tropical cloud feedback uncertainties in climate models
.
Geophys. Res. Lett.
,
32
,
L20806
, doi:.
Bony
,
S.
,
J. L.
Dufresne
,
H.
Le Treut
,
J. J.
Morcrette
, and
C.
Senior
,
2004
:
On dynamic and thermodynamic components of cloud changes
.
Climate Dyn.
,
22
,
71
86
, doi:.
Chadwick
,
R.
,
2016
:
Which aspects of CO2 forcing and SST warming cause the most uncertainty in projections of tropical rainfall change over land and ocean?
J. Climate
, doi:,
in press
.
Chadwick
,
R.
,
I.
Boutle
, and
G.
Martin
,
2013
:
Spatial patterns of precipitation change in CMIP5: Why the rich do not get richer in the tropics
.
J. Climate
,
26
,
3803
3822
, doi:.
Chou
,
C.
, and
J. D.
Neelin
,
2004
:
Mechanisms of global warming impacts on regional tropical precipitation
.
J. Climate
,
17
,
2688
2701
, doi:.
Chou
,
C.
,
J. D.
Neelin
,
C. A.
Chen
, and
J. Y.
Tu
,
2009
:
Evaluating the “rich-get-richer” mechanism in tropical precipitation change under global warming
.
J. Climate
,
22
,
1982
2005
, doi:.
Deser
,
C.
,
A.
Phillips
,
V.
Bourdette
, and
H. Y.
Teng
,
2012
:
Uncertainty in climate change projections: The role of internal variability
.
Climate Dyn.
,
38
,
527
546
, doi:.
Donohoe
,
A.
, and
D. S.
Battisti
,
2012
:
What determines meridional heat transport in climate models?
J. Climate
,
25
,
3832
3850
, doi:.
Frierson
,
D. M. W.
, and
Y. T.
Hwang
,
2012
:
Extratropical influence on ITCZ shifts in slab ocean simulations of global warming
.
J. Climate
,
25
,
720
733
, doi:.
Giorgi
,
F.
, and
R.
Francisco
,
2000
:
Evaluating uncertainties in the prediction of regional climate change
.
Geophys. Res. Lett.
,
27
,
1295
1298
, doi:.
Grose
,
M. R.
,
J.
Bhend
,
S.
Narsey
,
A.
Sen Gupta
, and
J. R.
Brown
,
2014
:
Can we constrain CMIP5 rainfall projections in the tropical Pacific based on surface warming patterns?
J. Climate
,
27
,
9123
9138
, doi:.
Hartmann
,
D. L.
, and
M. L.
Michelsen
,
1993
:
Large-scale effects on the regulation of tropical sea surface temperature
.
J. Climate
,
6
,
2049
2062
, doi:.
Hawkins
,
E.
, and
R.
Sutton
,
2009
:
The potential to narrow uncertainty in regional climate predictions
.
Bull. Amer. Meteor. Soc.
,
90
,
1095
1107
, doi:.
Hawkins
,
E.
, and
R.
Sutton
,
2011
:
The potential to narrow uncertainty in projections of regional precipitation change
.
Climate Dyn.
,
37
,
407
418
, doi:.
Held
,
I. M.
,
2001
:
The partitioning of the poleward energy transport between the tropical ocean and atmosphere
.
J. Atmos. Sci.
,
58
,
943
948
, doi:.
Held
,
I. M.
, and
B. J.
Soden
,
2006
:
Robust responses of the hydrological cycle to global warming
.
J. Climate
,
19
,
5686
5699
, doi:.
Huang
,
P.
,
2014
:
Regional response of annual- mean tropical rainfall to global warming
.
Atmos. Sci. Lett.
,
15
,
103
109
, doi:.
Huang
,
P.
,
S.-P.
Xie
,
K. M.
Hu
,
G.
Huang
, and
R. H.
Huang
,
2013
:
Patterns of the seasonal response of tropical rainfall to global warming
.
Nat. Geosci.
,
6
,
357
361
, doi:.
Hwang
,
Y. T.
, and
D. M. W.
Frierson
,
2013
:
Link between the double-intertropical convergence zone problem and cloud biases over the Southern Ocean
.
Proc. Natl. Acad. Sci. USA
,
110
,
4935
4940
, doi:.
IPCC
,
2013
:
Summary for policymakers
. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 1–29.
Kang
,
S. M.
,
I. M.
Held
,
D. M. W.
Frierson
, and
M.
Zhao
,
2008
:
The response of the ITCZ to extratropical thermal forcing: Idealized slab-ocean experiments with a GCM
.
J. Climate
,
21
,
3521
3532
, doi:.
Kang
,
S. M.
,
D. M. W.
Frierson
, and
I. M.
Held
,
2009
:
The tropical response to extratropical thermal forcing in an idealized GCM: The importance of radiative feedbacks and convective parameterization
.
J. Atmos. Sci.
,
66
,
2812
2827
, doi:.
Kent
,
C.
,
R.
Chadwick
, and
D. P.
Rowell
,
2015
:
Understanding uncertainties in future projections of seasonal tropical precipitation
.
J. Climate
,
28
,
4390
4413
, doi:.
Li
,
G.
, and
S.-P.
Xie
,
2014
:
Tropical biases in CMIP5 multimodel ensemble: The excessive equatorial Pacific cold tongue and double ITCZ problems
.
J. Climate
,
27
,
1765
1780
, doi:.
Liu
,
W.
,
J.
Lu
, and
S.-P.
Xie
,
2015
:
Understanding the Indian Ocean response to double CO2 forcing in a coupled model
.
Ocean Dyn.
,
65
,
1037
1046
, doi:.
Long
,
S.-M.
, and
S.-P.
Xie
,
2015
:
Intermodel variations in projected precipitation change over the North Atlantic: Sea surface temperature effect
.
Geophys. Res. Lett.
,
42
,
4158
4165
, doi:.
Long
,
S.-M.
,
S.-P.
Xie
,
X.-T.
Zheng
, and
Q.
Liu
,
2014
:
Fast and slow responses to global warming: Sea surface temperature and precipitation patterns
.
J. Climate
,
27
,
285
299
, doi:.
Lu
,
J.
, and
B.
Zhao
,
2012
:
The role of oceanic feedback in the climate response to doubling CO2
.
J. Climate
,
25
,
7544
7563
, doi:.
Ma
,
J.
, and
S.-P.
Xie
,
2013
:
Regional patterns of sea surface temperature change: A source of uncertainty in future projections of precipitation and atmospheric circulation
.
J. Climate
,
26
,
2482
2501
, doi:.
Ma
,
J.
,
S.-P.
Xie
, and
Y.
Kosaka
,
2012
:
Mechanisms for tropical tropospheric circulation change in response to global warming
.
J. Climate
,
25
,
2979
2994
, doi:.
Rowell
,
D. P.
,
2012
:
Sources of uncertainty in future changes in local precipitation
.
Climate Dyn.
,
39
,
1929
1950
, doi:.
Seager
,
R.
,
N.
Naik
, and
G. A.
Vecchi
,
2010
:
Thermodynamic and dynamic mechanisms for large-scale changes in the hydrological cycle in response to global warming
.
J. Climate
,
23
,
4651
4668
, doi:.
Soden
,
B. J.
, and
I. M.
Held
,
2006
:
An assessment of climate feedbacks in coupled ocean–atmosphere models
.
J. Climate
,
19
,
3354
3360
, doi:.
Su
,
H.
,
J. H.
Jiang
,
D. G.
Vane
, and
G. L.
Stephens
,
2008
:
Observed vertical structure of tropical oceanic clouds sorted in large-scale regimes
.
Geophys. Res. Lett.
,
35
,
L24704
, doi:.
Su
,
H.
,
J. H.
Jiang
,
C. X.
Zhai
,
T. J.
Shen
,
J. D.
Neelin
,
G. L.
Stephens
, and
Y. L.
Yung
,
2014
:
Weakening and strengthening structures in the Hadley circulation change under global warming and implications for cloud response and climate sensitivity
.
J. Geophys. Res.
,
119
,
5787
5805
, doi:.
Taylor
,
K. E.
,
R. J.
Stouffer
, and
G. A.
Meehl
,
2012
:
An overview of CMIP5 and the experiment design
.
Bull. Amer. Meteor. Soc.
,
93
,
485
498
, doi:.
Tebaldi
,
C.
, and
R.
Knutti
,
2007
:
The use of the multi-model ensemble in probabilistic climate projections
.
Philos. Trans. Roy. Soc. London
,
365A
,
2053
2075
, doi:.
Trenberth
,
K. E.
, and
C. J.
Guillemot
,
1995
:
Evaluation of the global atmospheric moisture budget as seen from analyses
.
J. Climate
,
8
,
2255
2272
, doi:.
Vecchi
,
G. A.
, and
B. J.
Soden
,
2007
:
Global warming and the weakening of the tropical circulation
.
J. Climate
,
20
,
4316
4340
, doi:.
Vecchi
,
G. A.
,
B. J.
Soden
,
A. T.
Wittenberg
,
I. M.
Held
,
A.
Leetmaa
, and
M. J.
Harrison
,
2006
:
Weakening of tropical Pacific atmospheric circulation due to anthropogenic forcing
.
Nature
,
441
,
73
76
, doi:.
Xie
,
S.-P.
, and
S. G. H.
Philander
,
1994
:
A coupled ocean-atmosphere model of relevance to the ITCZ in the eastern Pacific
.
Tellus
,
46A
,
340
350
, doi:.
Xie
,
S.-P.
,
C.
Deser
,
G. A.
Vecchi
,
J.
Ma
,
H. Y.
Teng
, and
A. T.
Wittenberg
,
2010
:
Global warming pattern formation: Sea surface temperature and rainfall
.
J. Climate
,
23
,
966
986
, doi:.
Xie
,
S.-P.
, and Coauthors
,
2015
:
Towards predictive understanding of regional climate change
.
Nat. Climate Change
,
5
,
921
930
, doi:.
Zelinka
,
M. D.
, and
D. L.
Hartmann
,
2012
:
Climate feedbacks and their implications for poleward energy flux changes in a warming climate
.
J. Climate
,
25
,
608
624
, doi:.
Zhou
,
Z.-Q.
, and
S.-P.
Xie
,
2015
:
Effects of climatological model biases on the projection of tropical climate change
.
J. Climate
,
28
,
9909
9917
, doi:.