Abstract

The sources of intermodel uncertainty in regional tropical rainfall projections are examined using a framework of atmosphere-only experiments. Uncertainty is dominated by model disagreement on shifts in convective regions, but the drivers of this uncertainty differ between land and ocean. Over the tropical oceans SST pattern uncertainty plays a substantial role, although it is not the only cause of uncertainty. Over land SST pattern uncertainty appears to be much less influential, and the largest source of uncertainty comes from the response to uniform SST warming, with a secondary contribution from the response to direct CO2 forcing. This may be because a larger number of processes can cause rainfall change in response to uniform SST warming than direct CO2 forcing, and so there is more potential for models to disagree. However, new experiments designed to more accurately decompose the regional climate responses of coupled models, combined with results from high-resolution climate modeling, are needed before these results can be considered robust. The pattern of present-day rainfall does not in general provide emergent constraints on future regional rainfall change. Correlations between relative humidity (RH) change and spatial shifts in convection over many land regions suggest that a proposed causal influence of RH change on dynamical rainfall change is plausible, although causality is not demonstrated here.

1. Introduction

Climate change is likely to bring substantial changes to annual and seasonal mean rainfall in many tropical regions (Neelin et al. 2006; Seager et al. 2010; Chadwick et al. 2015), with increases in some areas and decreases in others. Such changes could potentially have large impacts, as many tropical countries are particularly vulnerable to any change in the frequency of drought or flood years (IPCC 2014). The total area of tropical land affected by large annual mean changes in rainfall is projected to scale with global mean warming (Chadwick et al. 2015), but the location of changes remains very uncertain across climate models (Collins et al. 2013; McSweeney and Jones 2013; Kent et al. 2015). This impedes the planning and implementation of climate change adaptation in tropical countries, and the ability to attribute current trends in regional rainfall to greenhouse gas, aerosol, or natural forcing, or internal variability.

To narrow this uncertainty, it is first necessary to understand the physical mechanisms that lead to regional rainfall change in the tropics. Improved physical understanding could lead to the identification of those climate models with the most realistic representation of key processes for any given region, and hence more credible rainfall projections for that region (e.g., Brown et al. 2015). In the longer term, improved understanding should allow focused model development on the processes most important for rainfall change, and improved confidence in climate model projections.

The mechanisms that control rainfall change are dependent on the spatial scale that is considered. In the global mean, rainfall changes are governed by the tropospheric energy balance between changes in net radiative emission and changes in latent heat flux (Allen and Ingram 2002). One step down from this, for zonal means or other large spatial averages, the effect of greater atmospheric moisture transport in a warmer atmosphere can be seen with increased gradients of precipitation minus evaporation (PE), in both models (Held and Soden 2006) and observed changes (Durack et al. 2012; Liu and Allan 2013). This is known as the “wet-get-wetter, dry-get-drier” effect (Chou and Neelin 2004; Held and Soden 2006) and can be thought of as the thermodynamic component of rainfall change. However, this thermodynamic scaling fails to accurately predict rainfall change at smaller spatial scales (Chadwick et al. 2013; Roderick et al. 2014; Greve 2014; Byrne and O’Gorman 2015). At the regional and country scales relevant to climate change impacts, tropical rainfall projections are dominated by shifts in the position of wet and dry regions associated with circulation change (Muller and O’Gorman 2011; Chadwick et al. 2013; Kent et al. 2015). Therefore, accurate simulation of dynamical changes in regional climate is crucial for producing credible projections of regional tropical rainfall change (Shepherd 2014).

A large number of mechanisms have been proposed that could potentially drive regional changes in precipitation under warming, and it is likely that many of them contribute to the pattern of change seen in ensemble mean climate projections (He and Soden 2015). In this study, a set of idealized atmosphere-only GCM (AGCM) experiments are used to examine which aspects of CO2 forcing and sea surface temperature (SST) warming, and associated physical mechanisms, are most influential in causing intermodel uncertainty in future regional rainfall projections in the tropics. The individual contributions to uncertainty of uniform SST warming, SST pattern change, and the direct effects of CO2 forcing are all considered. In particular, how the balance of these mechanisms differs between land and ocean is examined.

Over the tropical oceans, dynamical shifts in convective regions have been linked to SST pattern change under warming (Xie et al. 2010; Ma and Xie 2013; Huang et al. 2013; Chadwick et al. 2014; Long et al. 2016). Ma and Xie (2013) found that the first two empirical orthogonal function (EOF) modes of intermodel variability in SST pattern change account for around one-third of intermodel variance in rainfall pattern change over the tropical oceans. This suggests that coupled ocean–atmosphere processes are important for determining the pattern of rainfall change over the tropical oceans, and may also influence rainfall change over land via atmospheric teleconnections. This has implications for the response of large-scale modes of ocean–atmosphere variability such as El Niño–Southern Oscillation (ENSO) to warming, with future changes in ENSO rainfall variability influenced by the mean-state pattern change of SST and rainfall (Power et al. 2013). It has also been suggested that rainfall could decrease on the edges of some convective regions, resulting from moisture transport into these regions not increasing as much as free-tropospheric temperature, therefore leading to increased moist stability (Chou and Neelin 2004).

The mechanisms that control regional rainfall change over tropical land are not as well established and may be less influenced by SST pattern changes. He et al. (2014) examined the ensemble mean response of regional rainfall change in the same set of atmosphere-only experiments as are used here, and found that the pattern of SST change appears not to be a major influence outside of the tropics, and is less important for most tropical land regions than it is over the tropical oceans. A slightly contrasting result is that of Anderson et al. (2015), who used a “principal uncertainty pattern” approach to link SST and precipitation change in coupled GCM projections and found that around 25% of the intermodel variance in regional precipitation change over land can be explained by uncertainty in the patterns of SST change.

Land regions warm more than the oceans (Joshi et al. 2008) and this enhanced temperature gradient is likely to drive circulation and rainfall change (Bayr and Dommenget 2013). The enhanced land warming is coupled to reductions in relative humidity (RH) over land (Joshi et al. 2008; Lambert et al. 2011; Byrne and O’Gorman 2013), which itself could drive rainfall change both through direct reductions in available moisture (Chadwick et al. 2013; Byrne and O’Gorman 2015) and through its effect on cloud-base height (Fasullo 2012) and circulation. Stability change over land can also be viewed as a balance between 1) remote ocean warming leading to free-tropospheric warming over land, which then increases stability and reduces rainfall, and 2) local CO2 radiative heating of the surface, which decreases stability and increases rainfall (Giannini 2010).

The effect of CO2-driven heating on rainfall change is generally partitioned into two parts, corresponding to different time scales of response. The first is associated with SST and ocean warming, and therefore acts on the time scales of years (mixed-layer heating) to centuries (warming of the deep ocean) (Held et al. 2010). The second is known as the fast response, or CO2 direct effect, and includes both direct atmospheric heating associated with increased CO2 and the part of land warming that is a direct response to increased CO2 (not mediated by SST increases). The CO2 direct atmospheric heating drives a tropics-wide increase in stability (Dong et al. 2009; Andrews et al. 2010; Cao et al. 2011; Bony et al. 2013) and a consequent global mean decrease in rainfall, but does not appear to directly play a major role in rainfall pattern change (Chadwick et al. 2014)—although it may do so indirectly by partly driving SST pattern change through circulation, RH, or cloud (Voigt and Shaw 2015) changes. In contrast, the land-heating aspect of the fast CO2 response does appear to be influential in driving regional rainfall change (Cao et al. 2012; Biasutti 2013; Chadwick et al. 2014; Richardson et al. 2016).

CO2 can also influence rainfall change by causing changes in plant stomata and transpiration (known as the plant physiological effect; Dong et al. 2009; Cao et al. 2012). This is a major driver of rainfall change in tropical forest regions in at least one climate model (Betts et al. 2008; Cao et al. 2012). Some Earth-system models now allow vegetation type and coverage to respond and feedback on CO2 forcing and warming, and this also has the potential to cause regional rainfall change (Martin and Levine 2012). These processes are present in many of the coupled GCMs examined here, but not in the corresponding atmosphere-only experiments, and so are possible sources of discrepancies between the two.

Anthropogenic aerosol emissions have been implicated in some of the largest regional rainfall changes of the past 50 years (Held et al. 2005; Rotstayn et al. 2007; Ackerley et al. 2011), and projected future reductions in aerosol concentrations are likely to be influential in low-to-medium greenhouse gas emission scenarios such as representative concentration pathways (RCPs) 2.6 and 4.5 (Rotstayn et al. 2015). In this study, for simplicity, only regional precipitation change mechanisms associated with CO2 forcing are considered. In fact, as will be shown in section 4, regional tropical rainfall change and uncertainty in a high-emissions scenario (RCP8.5) appears to be dominated by the response to greenhouse gas forcing rather than aerosols.

On a larger scale, latitudinal shifts of the zonal-mean tropical interconvergence zone (ITCZ) have been linked to changes in interhemispheric energy balance (Hwang et al. 2013; Seo et al. 2014). However, the tropical precipitation response of GCMs in high greenhouse-gas emissions scenarios does not resemble a zonal-mean latitudinal shift (Chadwick et al. 2013), possibly because the radiative perturbation induced by greenhouse gas forcing is much more hemispherically symmetrical than that associated with twentieth-century aerosol forcing.

The manuscript is structured as follows: Section 2 describes the GCM and AGCM experiments used in this study, and is followed in section 3 by a description of the precipitation change decomposition method used here. Section 4 is a discussion of how uncertainty in the AGCM experiments can be interpreted in relation to the full GCM experiments. Section 5 shows results on which mechanisms appear to be most influential in causing regional rainfall uncertainty, and in section 6 this is related to uncertainty in the GCM representation of the present-day rainfall climate. Section 7 investigates a possible connection between RH change and shifts in convection over tropical land, and finally section 8 contains a summary and some conclusions.

2. Experiments

A number of experiments from phase 5 of the Coupled Model Intercomparison Project (CMIP5) were used in this analysis (Taylor et al. 2012): 1) piControl: a coupled ocean–atmosphere control run with preindustrial forcing; 2) abrupt4xCO2: as in piControl, but with CO2 concentrations abruptly quadrupled at the start of the experiment; 3) Historical: a coupled atmosphere–ocean run for the period 1860–2005 with historical forcings applied; 4) RCP8.5: a coupled atmosphere–ocean run for the period 2006–2100 with forcings applied from a future higher emissions scenario with no explicit greenhouse gas mitigation policy; 5) amip (named for the Atmosphere Model Intercomparison Project): an atmosphere-only experiment, with forcings and SST boundary conditions taken from observations from 1979–2008; 6) amip4xCO2: as in amip, but with CO2 quadrupled throughout the experiment, although SST boundary conditions are the same as for amip; 7) amip4K: as in amip, but with a global uniform +4K SST perturbation; and 8) amipFuture: as in amip, but with an imposed climatological SST pattern anomaly taken from the ensemble mean of CMIP3 1% yr−1 CO2 increase runs at quadrupling, scaled to have a global mean of +4 K. The applied SST pattern anomaly is the same for all models. Models were only used in the majority of this analysis if all required variables were available for all seven experiments, and these 10 models are listed in Table 1. All models were regridded to a 2.5° grid for comparison with one another.

Table 1.

List of CMIP5 models with all necessary variables available for the idealized amip experiments. Spatial correlations of JJA precipitation change between RCP8.5, abrupt4xCO2, and amipTot are also shown for individual models.

List of CMIP5 models with all necessary variables available for the idealized amip experiments. Spatial correlations of JJA precipitation change between RCP8.5, abrupt4xCO2, and amipTot are also shown for individual models.
List of CMIP5 models with all necessary variables available for the idealized amip experiments. Spatial correlations of JJA precipitation change between RCP8.5, abrupt4xCO2, and amipTot are also shown for individual models.

To examine the response of precipitation to forcing, 30-yr mean differences between experiments are used. For amip4xCO2, amip4K and amipFuture, the difference with amip over the period 1979–2008 is used. For abrupt4xCO2, the difference with piControl over years 91–120 is used. For RCP8.5, the difference with Historical is taken, with years 2071–2100 and 1971–2000 used respectively. One further difference is taken, between amipFuture and amip4K, in order to isolate the impact of SST pattern change without mean SST warming (assuming that the two mechanisms combine linearly)—this difference is referred to as amipPattern.

All differences except for amip4xCO2-amip were scaled by a factor of (tropical mean SST warming)/4 K for each model, so that the tropical mean change was equivalent to that of amip4K in each case. This scaling produces better agreement between the ensemble mean rainfall change in the coupled and amip experiments, but would not be expected to make much of a difference to the intermodel uncertainty, as the contribution of uncertainty in global mean temperature change to uncertainty in regional precipitation change is small (Kent et al. 2015).

For analysis of correlations between climatological precipitation, ΔP and RH change in RCP8.5 (see sections 6 and 7), a larger number of models (35) was available. For calculation of model biases, monthly Global Precipitation Climatology Project (GPCP) rainfall observations were used (Adler 2003).

3. Decomposition of precipitation change

The mean rainfall change in each experiment can be decomposed into components corresponding to different driving mechanisms (Chadwick et al. 2013):

 
formula

where is the thermodynamic change in precipitation P due to Clausius–Clapeyron (CC)-related increases in specific humidity (ΔqCC) at constant ; is a proxy for convective mass flux from the boundary layer to the free troposphere, defined as = P/q, and q is 2-m specific humidity. Note that is closely correlated with the actual model convective mass flux (Chadwick et al. 2013) and is the “dynamical” variable in this decomposition. Also, ΔPRH = is the change due to near-surface relative humidity changes, where ΔqRH is the residual from the expected CC increase in q. Furthermore, ΔPweak = is the change at constant q due to the weakening tropical circulation; ΔPshift = is the change due to spatial shifts in the pattern of mass flux, driven by, for instance, SST pattern change, land–sea temperature contrast changes or local aerosol effects. Finally, ΔPcross = is the “cross term” component of precipitation change associated with the interaction of thermodynamic and dynamic change.

The decomposition here uses convective mass flux (or, in fact, a proxy for convective mass flux) as the basis for its dynamical terms ΔPweak and ΔPshift, and so in some cases may need to be interpreted differently to other decompositions that use large-scale vertical winds (ω) as the basis for their dynamical components (e.g., Seager et al. 2010; Bony et al. 2013). The terms ΔPweak, ΔPshift, ΔPcross, , and are denoted ΔPdiv, ΔPspat, ΔPNL, , and respectively in Chadwick et al. (2013), but their derivation is identical. For convenience, all terms except ΔPshift can be condensed into a single term,

 
formula

4. Using idealized AMIP experiments to understand uncertainty in coupled projections

The June–August (JJA) 30-yr mean surface temperature and precipitation changes for the ensemble mean of each experiment are shown in Figs. 1 and 2, respectively. The RCP8.5 (Fig. 2a) and abrupt4xCO2 (Fig. 2b) precipitation changes are very similar to one another (see Table 1 for spatial correlation coefficients for each model), indicating that greenhouse gas forcing is likely to dominate the response in RCP8.5, and that the simpler abrupt4xCO2 experiment is relevant for understanding regional projections under higher emissions scenarios. The similarity of responses in the two experiments also indicates that for 30-yr means the forced response to CO2 is generally much larger than internal variability even on regional scales, which is consistent with previous results (Rowell 2012; Kent et al. 2015). For brevity only the JJA means are shown in this and several other figures, but results from other seasons are also consistent with the analysis presented here.

Fig. 1.

CMIP5 JJA ensemble-mean surface temperature change (K) for (a) RCP8.5, (b) abrupt4xCO2, (c) amipTot, (d) amip4K, (e) amip4xCO2, and (f) amipPattern.

Fig. 1.

CMIP5 JJA ensemble-mean surface temperature change (K) for (a) RCP8.5, (b) abrupt4xCO2, (c) amipTot, (d) amip4K, (e) amip4xCO2, and (f) amipPattern.

Fig. 2.

As in Fig. 1, but for precipitation change (mm day−1).

Fig. 2.

As in Fig. 1, but for precipitation change (mm day−1).

Figure 2c shows the sum of the responses of amip4xCO2 and amipFuture (hereafter referred to as amipTot); amipTot has the same tropical-mean SST warming and CO2 concentration change as abrupt4xCO2 (after scaling), but with different baseline SSTs and atmospheric constituents, and with a different pattern of SST change. The ensemble-mean pattern of rainfall change in amipTot is on large scales similar to that of abrupt4xCO2, with a large increase in rainfall over the equatorial Pacific, increased rainfall over the South and East Asian monsoon regions with a corresponding decrease over the southern tropical Indian Ocean, increased rainfall just north of the equator in the Atlantic with decreases to the north and south, and decreased rainfall over Central America. However at smaller regional scales there are significant discrepancies between the two, for example over the West African monsoon region, the southwest Pacific, the southern Maritime Continent, and southwest India. When the responses of individual models to abrupt4xCO2 and amipTot are compared, these regional differences become more marked; low to moderate spatial correlations between the two experiments for each model are shown in Table 1.

These differences are likely to be mainly due to the different patterns of SST change between abrupt4xCO2 (different for each model) and amipFuture (the same imposed pattern for each model), the use of different baseline SSTs and atmospheric constituents, and processes that are represented in abrupt4xCO2 but not in amipTot such as the plant stomatal response to CO2. The absence of ocean–atmosphere coupling from the amip experiments could also potentially lead to discrepancies with the coupled models. However recent work on the use of AGCM time-slice experiments suggests that when forced with appropriate SST boundary conditions they can reproduce the same mean regional pattern of climate change as seen in coupled GCMs, despite their lack of coupling (Skinner et al. 2012; He and Soden 2016).

In section 5 the intermodel uncertainty in the regional rainfall response to forcing of each of the amip experiments is used to understand the main drivers of uncertainty in the abrupt4xCO2 experiment. Ideally, the abrupt4xCO2 rainfall pattern response would be identical to that of amipTot for each model, meaning that the rainfall response of abrupt4xCO2 at each grid point could be separated into the responses of amip4K, amip4xCO2, and amipPattern. Clearly this condition is not met in many regions (Table 1), so the use of this separation here is an approximation, and should only be taken as indicative of the proportion of regional uncertainty associated with each aspect of forcing and warming. Despite this necessary caveat, the CMIP5 amip experiments should provide useful information about the mechanisms that drive uncertainty in projections of rainfall change, particularly on larger scales.

The experiments amip4K (Fig. 2d), amip4xCO2 (Fig. 2e), and amipPattern (Fig. 2f) all contribute toward the total ensemble mean pattern of change in amipTot (see also He and Soden 2015). Of these, amipPattern is particularly influential over the tropical oceans, with amip4xCO2 acting to oppose and mitigate the effect of amip4K in many regions. When rainfall change in each experiment is decomposed using the method in section 3 it can be seen that shifts in convection (ΔPshift) are the most influential mechanism driving the pattern of rainfall change (cf. Figs. 2 and 3) (Chadwick et al. 2013; Kent et al. 2015). These shifts can be local or remote, and could be caused by any mechanism that alters the spatial distribution of convection.

Fig. 3.

As in Fig. 2, but for the component of precipitation change associated with spatial shifts in convection (ΔPshift).

Fig. 3.

As in Fig. 2, but for the component of precipitation change associated with spatial shifts in convection (ΔPshift).

The sum of the other components (ΔPother, Fig. 4) for abrupt4xCO2 (Fig. 4b) and amipTot (Fig. 4c) provides a large-scale weak increase in rainfall across tropical wet regions (in this case for JJA), associated with the residual of thermodynamic moisture increases and dynamical rainfall decreases due to the large-scale weakening of the tropical circulation. In the case of amip4xCO2, the weak signal of ΔPother is consistent with the results of Chadwick et al. (2014), who found that rainfall pattern change in amip4xCO2 is likely to be largely driven by changes in land–sea temperature gradient rather than the large-scale stabilization of the atmosphere caused by direct CO2-driven atmospheric heating. For amip4K the magnitudes of ΔPshift and ΔPother are more comparable, and the ΔPshift component here may also be largely driven by changes in land–sea temperature gradient; over land this can be thought of as the remote effect of SST warming suppressing land-based convection (Giannini 2010).

Fig. 4.

As in Fig. 2, but for the sum of all components of precipitation change except spatial shifts in convection (ΔPother).

Fig. 4.

As in Fig. 2, but for the sum of all components of precipitation change except spatial shifts in convection (ΔPother).

5. Sources of uncertainty in future rainfall projections

The intermodel standard deviation (SD) of precipitation change at each grid point is shown in Fig. 5. The patterns of uncertainty are very similar between RCP8.5 and abrupt4xCO2, suggesting that abrupt4xCO2 is useful for understanding uncertainty in model projections under higher emissions scenarios, as well as the ensemble mean change. The dominant source of uncertainty in all experiments is intermodel disagreement on the position and magnitude of shifts in convective regions (Figs. 6 and 7), a result that has been previously shown for RCP8.5 (Kent et al. 2015), but which is seen here to also apply for abrupt4xCO2 and the amip experiments. This is due to anticorrelations across models between several of the other terms in the rainfall change decomposition, particularly ΔPT and ΔPweak, and ΔPT and ΔPRH over land, resulting in only a small uncertainty in the combined term ΔPother (Kent et al. 2015).

Fig. 5.

CMIP5 JJA intermodel standard deviation of precipitation change (mm day−1) for (a) RCP8.5, (b) abrupt4xCO2, (c) amip4K+amip4xCO2, (d) amip4K, (e) amip4xCO2, and (f) amipPattern.

Fig. 5.

CMIP5 JJA intermodel standard deviation of precipitation change (mm day−1) for (a) RCP8.5, (b) abrupt4xCO2, (c) amip4K+amip4xCO2, (d) amip4K, (e) amip4xCO2, and (f) amipPattern.

Fig. 6.

As in Fig. 5, but for the component of precipitation change associated with spatial shifts in convection (ΔPshift).

Fig. 6.

As in Fig. 5, but for the component of precipitation change associated with spatial shifts in convection (ΔPshift).

Fig. 7.

As in Fig. 5, but for the sum of all components of precipitation change except spatial shifts in convection (ΔPother).

Fig. 7.

As in Fig. 5, but for the sum of all components of precipitation change except spatial shifts in convection (ΔPother).

As explained in section 4, the amip experiments do not combine to be identical to abrupt4xCO2 for each model, so care must be taken with the comparison of intermodel uncertainty between the various experiments. In particular differences between uncertainties in the experiments (e.g., SD of abrupt4xCO2 − SD of amip4K) are not plotted, as on a grid point by grid point basis this comparison is likely to be misleading. Instead, more general and larger-scale differences between the intermodel uncertainties are examined.

a. Ocean

Figure 5c shows the SD of the sum of amip4K and amip4xCO2. Unlike in the ensemble mean analysis (Fig. 2) we do not use amipTot here (i.e., amipPattern is not also included), because the use of a single pattern of SST change across models in amipTot leads to difficulties in interpretation of its intermodel uncertainty compared to that of the coupled experiments (see section 4). Instead, by comparing the SD of amip4K+amip4xCO2 (Fig. 5c) to the SD of abrupt4xCO2 (5b), any differences should be due to sources of uncertainty that are present in abrupt4xCO2 but not in amip4K+amip4xCO2. These include intermodel uncertainty in mean SST pattern change, uncertainty due to ocean–atmosphere coupling, and uncertainty due to Earth-system processes such as the plant physiological effect. Over the oceans there are large regions in the tropical Pacific, and southern tropical Indian and Atlantic Oceans where uncertainty in abrupt4xCO2 is substantially larger than in amip4K+amip4xCO2. This “missing” uncertainty is likely to be associated with variations in the pattern of SST change across models (Xie et al. 2010; Ma and Xie 2013), and the atmospheric response to SST pattern change, and this appears to be particularly important over much of the tropical Pacific.

However this is not the only source of uncertainty over the tropical oceans, as amip4K+amip4xCO2 has significant uncertainty even without the effect of SST pattern change–particularly over the tropical Atlantic, east Pacific, Maritime Continent, and northern Indian Ocean regions. In the northern Indian Ocean and the region around the Philippines, uncertainty in amip4K+amip4xCO2 is actually larger than in abrupt4xCO2, which could only be physical if there is an anticorrelation across models between the rainfall response to some mechanisms represented in amip4K+amip4xCO2, and the response to some processes that are not (e.g., the circulation response to, say, land–sea temperature gradient change in each model might be partly opposed by the response to, say, SST pattern change in each model). Alternatively this could be an artifact of using the amip experiments to interpret abrupt4xCO2. Internal atmospheric variability may also contribute to differences between the experiments, but this has been shown to be relatively small in GCM simulations of tropical rainfall when 30-yr means are used (Kent et al. 2015).

Of the uncertainty in amip4K+amip4xCO2, the contribution from amip4K is larger. There is also some negative covariance between the two terms, meaning that the variance of amip4K+amip4xCO2 is smaller than the sum of the individual variances. This covariance may be because amip4K warms the SSTs, whereas amip4xCO2 directly warms the land surface, and the circulation responses over the ocean oppose each other to some extent. The relative contribution of amip4xCO2 to the combined SD is largest in the tropical Atlantic, east and west Pacific regions, and northern Indian Ocean.

The SD of amipPattern (Fig. 5f) is a measure of the intermodel uncertainty of the atmospheric and land surface response to a given pattern of SST pattern change (in this case the CMIP3 ensemble-mean change). This does not include any uncertainty due to differences in SST pattern change between models, and as such is perhaps surprisingly large. It is possible that some of this uncertainty is artificial, due to the imposition on each model of an ensemble mean pattern of SST change that is not actually produced by any individual model in response to CO2 forcing. Nevertheless it highlights that different models can produce a wide range of local and remote rainfall response even when the SST pattern change is specified. It is possible that some of this uncertainty could be constrained by examining which models are best at representing present-day teleconnections in response to SST pattern variability (e.g., ENSO).

b. Land

To examine rainfall uncertainty over land in more detail, Figs. 8 and 9 show the intermodel SD for each experiment with ocean regions masked out, for JJA and December–February (DJF) respectively. For the majority of land regions, the magnitude of the uncertainty in amip4K+amip4xCO2 (Figs. 8c and 9c) is comparable to that of abrupt4xCO2 (Figs. 8b and 9b). This is somewhat surprising, as it indicates that SST pattern uncertainty and other Earth-system process uncertainty may not be important contributors to intermodel rainfall change uncertainty over the majority of tropical land (see also He et al. 2014). There are exceptions in several regions, and here SST pattern uncertainty may well be important (Ma and Xie 2013; Anderson et al. 2015), as well as the response of tropical forests to the plant physiological effect (Betts et al. 2008; Cao et al. 2012).

Fig. 8.

As in Fig. 5, but for land regions only.

Fig. 8.

As in Fig. 5, but for land regions only.

Fig. 9.

As Fig. 8, but for DJF.

Fig. 9.

As Fig. 8, but for DJF.

Both amip4K (Figs. 8d and 9d) and amip4xCO2 (Figs. 8e and 9e) contribute to the uncertainty in amip4K+amip4xCO2, but the contribution from amip4K is larger, as it is over the oceans. This is consistent with the results of Richardson et al. (2016), who found relatively similar regional rainfall responses across CMIP5 models in the amip4xCO2 experiments. Again there is negative covariance between the amip4K and amip4xCO2 experiments, which can be explained by the opposing effects of remote SST forcing and local CO2-induced land warming on atmospheric stability and rainfall (Giannini 2010). In each model the magnitudes of the responses over land in amip4K and amip4xCO2 are comparable (as is seen for the ensemble mean; Figs. 2d,e), so the greater SD in amip4K than in amip4xCO2 is not simply explained by a difference in the magnitude of response between the experiments. Instead, the reason may be that there are more rainfall change mechanisms at work in amip4K than in amip4xCO2, and therefore more potential for different models to represent the response to these mechanisms in different ways and cause uncertainty. In amip4xCO2, rainfall change over land is driven by both direct stabilization of the atmosphere (Sugi and Yoshimura 2004) and an increased land–sea temperature gradient (Cao et al. 2012; Bayr and Dommenget 2013), of which the land heating may be dominant (Chadwick et al. 2014). In amip4K, potential processes of rainfall change include remote suppression of rainfall over land by SST warming (Giannini 2010), an increased land–sea temperature gradient, impacts of decreased RH (relative humidity) on cloud-base height (Fasullo 2012), and changes in local moisture (Chadwick et al. 2013) and moisture gradients (Byrne and O’Gorman 2015).

In several regions, particularly the Maritime Continent, South America, India, and Southeast Asia, amipPattern (Figs. 8f and 9f) shows that the atmospheric response to a given pattern of SST change is also a source of uncertainty. In fact, if this uncertainty is added to that of amip4K+amip4xCO2 (Figs. 8c and 9c) then it is larger than the abrupt4xCO2 uncertainty in several regions (e.g., India JJA). Again, this can only be physical if there is some anticorrelation across models between processes represented in the amip experiments, and those not represented in them but included in abrupt4xCO2, or between amip4K+amip4xCO2 and amipPattern. Otherwise it must be an artifact of the use of amip experiments to examine uncertainty in coupled experiments. Related to this, it seems unlikely that the rainfall uncertainty in a given region could be affected by the atmospheric response to SST pattern change, but not uncertainty in the pattern of SST change itself, unless intermodel uncertainty in future SST pattern change happens to be small at the source of the relevant teleconnection.

6. Relationship between uncertainty in future projections and uncertainty in the simulation of current climate

In this section, relationships between the pattern of present-day rainfall in CMIP5, the intermodel uncertainty in this pattern, and the intermodel uncertainty in future projections are investigated. Figure 10 shows the intermodel standard deviation of climatological rainfall in the Historical, piControl, and amip experiments. The pattern is similar among all three experiments, with most uncertainty in the seasonal rainy regions as might be expected. The intermodel uncertainty in the amip experiment is smaller than in the coupled experiments in much of the tropical Pacific, as would be expected in regions where the influence of model-dependent SST patterns is strong. However in many other regions the amip precipitation uncertainty is actually larger than in the coupled experiments, despite all models having the same SST boundary conditions in amip. One possible reason for this is the presence of compensating errors—a negative feedback between SST and precipitation biases—in coupled models (Inness and Slingo 2003; Martin et al. 2006) that serve to reduce precipitation differences between models but that are not present in the atmosphere-only amip experiment.

Fig. 10.

CMIP5 JJA intermodel standard deviation of climatological precipitation (mm day−1) for (a) Historical, (b) preindustrial control, and (c) amip.

Fig. 10.

CMIP5 JJA intermodel standard deviation of climatological precipitation (mm day−1) for (a) Historical, (b) preindustrial control, and (c) amip.

The pattern of intermodel uncertainty in climatological rainfall appears at least somewhat similar to the pattern of uncertainty in future rainfall change, so this is examined in more detail for the Historical and RCP8.5 experiments (Figs. 11 and 12) by taking advantage of the much larger set of models (35) available for these runs than for the idealized amip experiments. The pattern of future rainfall uncertainty calculated with 35 models (Fig. 11b) is quite similar to (although in many regions smaller than) the pattern calculated with 10 models (Fig. 5a), giving some confidence that the results of intermodel uncertainty in the amip experiments is representative of uncertainty in the wider population of CMIP5 models. In fact the standard deviation in the larger set of models would be expected to be smaller, as it contains a larger proportion of models from the same modeling center, which in general are less independent from one another than models from different modeling centers (Pennell and Reichler 2011).

Fig. 11.

(a) CMIP5 JJA ensemble mean Historical 1971–2000 precipitation (mm day−1), (b) JJA intermodel standard deviation of 1971–2000 precipitation in the Historical experiment (mm day−1), (c) JJA intermodel standard deviation of RCP8.5 precipitation change (mm day−1), and (d) intermodel correlation at each grid point between climatological precipitation and precipitation change. Correlations which are statistically significant at the 95% level are stippled.

Fig. 11.

(a) CMIP5 JJA ensemble mean Historical 1971–2000 precipitation (mm day−1), (b) JJA intermodel standard deviation of 1971–2000 precipitation in the Historical experiment (mm day−1), (c) JJA intermodel standard deviation of RCP8.5 precipitation change (mm day−1), and (d) intermodel correlation at each grid point between climatological precipitation and precipitation change. Correlations which are statistically significant at the 95% level are stippled.

Fig. 12.

As Fig. 11, but for DJF.

Fig. 12.

As Fig. 11, but for DJF.

Spatial correlation coefficients over the tropics between the intermodel uncertainty in future rainfall change, uncertainty in present-day rainfall, and ensemble mean present-day rainfall are shown in Table 2 (first and second columns) and are relatively high (e.g., for JJA the value is 0.81 for land and ocean, and 0.88 for land only). However, it is possible that much of this correlation occurs simply because rainfall is generally limited to certain broad regions of the tropics in each season, and so it is natural for both the ensemble mean present-day rainfall, the uncertainty in present-day rainfall, and uncertainty in future change to all be collocated within the overall regions of the tropics where rainfall actually occurs. To test this, the spatial correlations were repeated, omitting seasonally dry regions of the tropics where mean rainfall is <3 mm day−1 (Table 2) (correlations are not sensitive to the exact value of this threshold). In this case the correlations between the different patterns are only moderate (for JJA, 0.58 for land and ocean and 0.65 for land only), suggesting that much of the original high correlation may indeed be because of this trivial reason. The presence of these moderate correlations does suggest some influence of the pattern of present-day rainfall (and/or intermodel uncertainty in this pattern) on the pattern of uncertainty in future rainfall change, although not a dominant one.

Table 2.

Spatial correlation coefficients (cc) between various precipitation fields, for land and ocean, wet regions of land and ocean, land only, and wet regions of land only, annually (ANN) and for each season. Columns show 1) spatial cc between the intermodel standard deviation of precipitation change between the Historical and RCP8.5 experiments (ΔP), and the intermodel standard deviation of precipitation in the Historical (1971–2000) experiment (Pclim); 2) spatial cc between the intermodel standard deviation of ΔP, and the ensemble mean Pclim; 3) ensemble mean of each model’s spatial cc between ΔP and Pclim; and 4) ensemble mean of each model’s spatial cc between ΔP and its present-day bias compared to GPCP observations (Pbias). Here, the em dash means not applicable.

Spatial correlation coefficients (cc) between various precipitation fields, for land and ocean, wet regions of land and ocean, land only, and wet regions of land only, annually (ANN) and for each season. Columns show 1) spatial cc between the intermodel standard deviation of precipitation change between the Historical and RCP8.5 experiments (ΔP), and the intermodel standard deviation of precipitation in the Historical (1971–2000) experiment (Pclim); 2) spatial cc between the intermodel standard deviation of ΔP, and the ensemble mean Pclim; 3) ensemble mean of each model’s spatial cc between ΔP and Pclim; and 4) ensemble mean of each model’s spatial cc between ΔP and its present-day bias compared to GPCP observations (Pbias). Here, the em dash means not applicable.
Spatial correlation coefficients (cc) between various precipitation fields, for land and ocean, wet regions of land and ocean, land only, and wet regions of land only, annually (ANN) and for each season. Columns show 1) spatial cc between the intermodel standard deviation of precipitation change between the Historical and RCP8.5 experiments (ΔP), and the intermodel standard deviation of precipitation in the Historical (1971–2000) experiment (Pclim); 2) spatial cc between the intermodel standard deviation of ΔP, and the ensemble mean Pclim; 3) ensemble mean of each model’s spatial cc between ΔP and Pclim; and 4) ensemble mean of each model’s spatial cc between ΔP and its present-day bias compared to GPCP observations (Pbias). Here, the em dash means not applicable.

If there is a simple relationship across models between present-day rainfall and future rainfall change for any region, this could be used as an emergent constraint. Figures 11d and 12d show intermodel correlations at each grid point between these two variables, and values are generally low and often not statistically significant within rainy seasons and regions, particularly over land. Some dry season correlations are relatively large (e.g., the western Sahara in DJF and southwest Africa in JJA), but the magnitudes of the rainfall changes involved are likely to be small. Many semiarid regions experience a dry-season drying response to climate change (Solomon et al. 2009), so these negative correlations may be because models that start with higher dry-season rainfall in these regions are able to experience larger reductions in rainfall under warming. In general a simple constraint with climatological precipitation does not appear to be a promising approach for narrowing the range of projections of future regional precipitation change, although there may be regional exceptions. This is also true when fractional changes in rainfall (ΔP/Pclim) are considered.

Finally, the relationships between the pattern of future rainfall change in each model, that model’s pattern of present-day rainfall, and its biases (compared to GPCP observations) were examined by calculating spatial correlations between each field. Both of these correlations are very weak (Table 2, columns 3 and 4), suggesting that local change in rainfall in any particular model is not related to the magnitude of its local present-day rainfall (see also Chadwick et al. 2013) or to the magnitude of its local model biases (Chadwick et al. 2015).

7. Connection between RH change and land rainfall shifts

One possible cause of some of the shifts in convection that dominate regional uncertainty in tropical precipitation projections is reductions in relative humidity over many land regions. RH reductions over land appear may be a fundamental response of the climate system to warming (Joshi et al. 2008; Dong et al. 2009; Byrne and O’Gorman 2013). As well as directly affecting rainfall through changes in available moisture (Chadwick et al. 2013; Byrne and O’Gorman 2015), reduced RH could potentially inhibit convection by increasing the height of the lifting condensation level (Fasullo 2012). This possible mechanism is investigated here by examining correlations across models at each grid point between ΔPshift and ΔRH. Here, 27 CMIP5 models with all necessary variables were available for the RCP8.5-Historical simulations.

Figure 13 shows that relatively high, statistically significant correlations between ΔPshift and ΔRH are present over a number of land regions, including Australia, the drier regions of Africa, much of South and Central America, and India during its monsoon season. Although this does not establish anything about causality, and precipitation changes could equally be responsible for RH changes, the existence of these correlations indicates that further study of this mechanism may be worthwhile. Correlations between ΔP and ΔRH (not shown) have a very similar pattern but are in general slightly stronger, which is expected due to the additional relationship between ΔRH and ΔPRH that is included in this correlation. It should be noted that the number of degrees of freedom used in these significance tests is probably overestimated, as many CMIP5 GCMs are not truly independent of one another (Pennell and Reichler 2011), so the threshold value of R taken as the 95% significance level is likely to be underestimated (both here and section 6).

Fig. 13.

CMIP5 RCP8.5 intermodel correlation at each grid point between ΔRH and ΔPshift for (a) DJF, (b) March–May (MAM), (c) JJA, and (d) September–November (SON). Correlations that are statistically significant at the 95% level are stippled.

Fig. 13.

CMIP5 RCP8.5 intermodel correlation at each grid point between ΔRH and ΔPshift for (a) DJF, (b) March–May (MAM), (c) JJA, and (d) September–November (SON). Correlations that are statistically significant at the 95% level are stippled.

8. Summary and conclusions

The idealized amip experiments examined here suggest that regional tropical rainfall uncertainty in coupled climate projections is dominated by model disagreement on shifts in convective regions, but that the source of this uncertainty differs between land and ocean. Over the tropical oceans SST pattern uncertainty plays a substantial role, particularly in the Pacific, although it is not the only cause of uncertainty. Over land SST pattern uncertainty appears, somewhat surprisingly, to be much less influential (see also He et al. 2014).

The largest source of intermodel uncertainty over land comes from models’ varying responses to uniform SST warming, with a secondary contribution from their responses to direct CO2 forcing—much of which is likely to be associated with land warming. This may be because a larger number of processes can cause rainfall change in response to uniform warming than direct CO2 forcing, and so there is more potential for models to disagree. Alternatively, one or more of the processes associated with the influence of uniform SST warming over land may be particularly uncertain across GCMs. The amip4K and amip4xCO2 experiments produce opposing responses over many regions of land and ocean, but in most cases the patterns are not identical, and the total pattern of rainfall change over land is often determined by the residual of the two responses (Figs. 2d,e). Understanding why these two mechanisms have similar but slightly different patterns could increase our understanding of regional rainfall change.

If the largest cause of tropical land rainfall uncertainty in future projections really is the response to uniform SST warming, this would substantially simplify the search for detailed causes of this uncertainty. Mechanisms could be examined in atmosphere-only experiments, without the need to consider coupled atmosphere–ocean processes or the pattern of mean SST change. This would also provide significant encouragement that (global and possibly regional) high-resolution atmosphere-only time-slice experiments can accurately represent the regional land rainfall response to climate change. The implication would be that changing patterns of land–ocean teleconnections are not crucial to determining local land rainfall responses, and perhaps the land response can be thought of more as a combination of responses to remote uniform SST warming and local CO2 forcing (Giannini 2010). Of course, changes in large-scale land–land teleconnections could still be crucial.

Although this would be encouraging for the prospects of understanding and narrowing uncertainty in future rainfall change over tropical land, more work is needed before this result can be considered robust. The idealized amip experiments used here were not designed to be used to decompose the response of coupled climate models on a regional scale. The use of observed SSTs and atmospheric constituents as a base state in the amip experiment, as well as the application of a single mean pattern of SST change in amipFuture, makes it difficult to interpret coupled model regional responses as a sum of the amip responses. This has motivated a new set of experiments, sstPi (SST preindustrial), which will form part of the Cloud Feedbacks Model Intercomparison Project (CFMIP) contribution to CMIP6, and formally decompose the coupled abrupt4xCO2 response in each model into the various aspects of SST warming and CO2 forcing. These experiments will build on the results of the CMIP5 amip experiments shown here by providing a much more reliable and detailed estimate of the processes that drive regional climate change and intermodel uncertainty.

It is also unclear whether the current generation of GCMs is able to represent land–ocean teleconnections sufficiently well to be able to capture any future changes in these teleconnections. Many models represent observed present-day teleconnections poorly, and even the best models do not capture all teleconnections (Rowell 2013). This may be partly due to inadequate horizontal and/or vertical resolution in GCMs, and there is evidence that land rainfall trends in response to SST pattern change can be represented much better at higher resolution (Vellinga et al. 2016). Therefore even if the idealized amip experiments do correctly represent the balance of processes in current coupled projections, the coupled models themselves may be underestimating the effect of SST pattern change on land rainfall changes. A continued focus on understanding the effects of resolution in climate modeling will be needed to address this question.

There appears to be a moderate spatial correlation between the spatial patterns of present-day rainfall in GCMs, uncertainty in present-day rainfall position, and uncertainty in future rainfall changes. However, in general the pattern of present-day rainfall does not provide emergent constraints on the future regional rainfall change simulated by GCMs. Correlations between RH change and spatial shifts in convection over many land regions suggest that a proposed causal influence of RH change on dynamical rainfall change (Fasullo 2012) is plausible, although causality is not demonstrated here.

Acknowledgments

Many thanks to Chris Kent for writing the code that was adapted for use in this analysis. Gill Martin and Peter Good also contributed valuable comments and insights. The author was supported by the Joint UK DECC/Defra Met Office Hadley Centre Climate Programme (GA01101). I acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is responsible for CMIP, and thank the climate modelling groups for producing and making available their model output. For CMIP the U.S. Department of Energy’s Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. I thank Geert Jan van Oldenborgh for making the CMIP5 RCP8.5 and Historical data easily available via the KNMI Climate Explorer tool, and Jamie Kettleborough, Ian Edmond, and Emma Hibling for developing the software used to download the other CMIP5 data used here.

REFERENCES

REFERENCES
Ackerley
,
D.
,
B.
Booth
,
S.
Knight
,
E.
Highwood
,
D.
Frame
,
M.
Allen
, and
D.
Rowell
,
2011
:
Sensitivity of twentieth-century Sahel rainfall to sulfate aerosol and CO2 forcing
.
J. Climate
,
24
,
4999
5014
, doi:.
Adler
,
R.
,
2003
:
The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present)
.
J. Hydrometeor.
,
4
,
1147
1167
, doi:.
Allen
,
M.
, and
W.
Ingram
,
2002
:
Constraints on future changes in climate and the hydrologic cycle
.
Nature
,
419
,
224
232
, doi:.
Anderson
,
B.
,
B.
Lintner
,
B.
Langenbrunner
,
J.
Neelin
,
E.
Hawkins
, and
J.
Syktus
,
2015
:
Sensitivity of terrestrial precipitation trends to the structural evolution of sea surface temperatures
.
Geophys. Res. Lett.
,
42
,
1190
1196
, doi:.
Andrews
,
T.
,
P. M.
Forster
,
O.
Boucher
,
N.
Bellouin
, and
A.
Jones
,
2010
:
Precipitation, radiative forcing and global temperature change
.
Geophys. Res. Lett.
,
37
,
L14701
, doi:.
Bayr
,
T.
, and
D.
Dommenget
,
2013
:
The tropospheric land–sea warming contrast as the driver of tropical sea level pressure changes
.
J. Climate
,
26
,
1387
1402
, doi:.
Betts
,
R.
,
P.
Cox
,
M.
Collins
,
P.
Harris
,
C.
Huntingford
, and
C.
Jones
,
2008
:
The role of ecosystem–atmosphere interactions in simulated Amazonian precipitation decrease and forest dieback under global climate warming
.
Theor. Appl. Climatol.
,
78
,
157
175
, doi:.
Biasutti
,
M.
,
2013
:
Forced Sahel rainfall trends in the CMIP5 archive
.
J. Geophys. Res. Atmos.
,
118
,
1613
1623
, doi:.
Bony
,
S.
,
G.
Bellon
,
D.
Klocke
,
S.
Sherwood
,
S.
Fermepin
, and
S.
Denvil
,
2013
:
Robust direct effect of carbon dioxide on tropical circulation and regional precipitation
.
Nat. Geosci.
,
6
,
447
451
, doi:.
Brown
,
J.
,
R.
Matear
,
J.
Brown
, and
J.
Katzfey
,
2015
:
Precipitation projections in the tropical Pacific are sensitive to different types of SST bias adjustment
.
Geophys. Res. Lett.
,
42
,
10 856
10 866
, doi:.
Byrne
,
M. P.
, and
P. A.
O’Gorman
,
2013
:
Link between land–ocean warming contrast and surface relative humidities in simulations with coupled climate models
.
Geophys. Res. Lett.
,
40
,
5223
5227
, doi:.
Byrne
,
M. P.
, and
P. A.
O’Gorman
,
2015
:
The response of precipitation minus evapotranspiration to climate warming: Why the “wet-get-wetter, dry-get-drier” scaling does not hold over land
.
J. Climate
,
28
,
8078
8092
, doi:.
Cao
,
L.
,
G.
Bala
, and
K.
Caldeira
,
2011
:
Why is there a short-term increase in global precipitation in response to diminished CO2 forcing?
Geophys. Res. Lett.
,
38
,
L06703
, doi:.
Cao
,
L.
,
G.
Bala
, and
K.
Caldeira
,
2012
:
Climate response to changes in atmospheric carbon dioxide and solar irradiance on the time scale of days to weeks
.
Environ. Res. Lett.
,
7
, 034015, doi:.
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:.
Chadwick
,
R.
,
P.
Good
,
T.
Andrews
, and
G.
Martin
,
2014
:
Surface warming patterns drive tropical rainfall pattern responses to CO2 forcing on all timescales
.
Geophys. Res. Lett.
,
41
,
610
615
, doi:.
Chadwick
,
R.
,
P.
Good
,
G.
Martin
, and
D.
Rowell
,
2015
:
Large rainfall changes consistently projected over substantial areas of tropical land
.
Nat. Climate Change
,
6
,
177
181
, doi:.
Chou
,
C.
, and
J.
Neelin
,
2004
:
Mechanisms of global warming impacts on regional tropical precipitation
.
J. Climate
,
17
,
2688
2701
, doi:.
Collins
,
M.
, and Coauthors
,
2013
: Long-term climate change: Projections, commitments and irreversibility. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 1029–1136.
Dong
,
B.
,
J. M.
Gregory
, and
R. T.
Sutton
,
2009
:
Understanding land–sea warming contrast in response to increasing greenhouse gases. Part I: Transient adjustment
.
J. Climate
,
22
,
3079
3097
, doi:.
Durack
,
P.
,
S.
Wijffels
, and
R.
Matear
,
2012
:
Ocean salinities reveal strong global water cycle intensification during 1950 to 2000
.
Science
,
336
,
455
458
, doi:.
Fasullo
,
J.
,
2012
:
A mechanism for land–ocean contrasts in global monsoon trends in a warming climate
.
Climate Dyn.
,
39
,
1137
1147
, doi:.
Giannini
,
A.
,
2010
:
Mechanisms of climate change in the semiarid African Sahel: The local view
.
J. Climate
,
23
,
743
756
, doi:.
Greve
,
P. A.
,
2014
:
Global assessment of trends in wetting and drying over land
.
Nat. Geosci.
,
7
,
716
721
, doi:.
He
,
J.
, and
B.
Soden
,
2015
:
Anthropogenic weakening of the tropical circulation: The relative roles of direct CO2 forcing and sea surface temperature change
.
J. Climate
,
28
,
8728
8742
, doi:.
He
,
J.
, and
B.
Soden
,
2016
:
Does the lack of coupling in SST-forced atmosphere-only models limit their usefulness for climate change studies?
J. Climate
, in press.
He
,
J.
,
B.
Soden
, and
B.
Kirtman
,
2014
:
The robustness of the atmospheric circulation and precipitation response to future anthropogenic surface warming
.
Geophys. Res. Lett.
,
41
,
2614
2622
, doi:.
Held
,
I. M.
, and
B. J.
Soden
,
2006
:
Robust responses of the hydrological cycle to global warming
.
J. Climate
,
19
,
5686
5699
, doi:.
Held
,
I. M.
,
T.
Delworth
,
J.
Lu
,
K.
Findell
, and
T.
Knutson
,
2005
:
Simulation of Sahel drought in the 20th and 21st centuries
.
Proc. Natl. Acad. Sci. USA
,
102
,
17 891
17 896
, doi:.
Held
,
I. M.
,
M.
Winton
,
K.
Takahashi
,
T.
Delworth
,
F.
Zeng
, and
G. K.
Vallis
,
2010
:
Probing the fast and slow components of global warming by returning abruptly to preindustrial forcing
.
J. Climate
,
23
,
2418
2427
, doi:.
Huang
,
P.
,
S.-P.
Xie
,
K.
Hu
,
G.
Huang
, and
R.
Huang
,
2013
:
Patterns of the seasonal response of tropical rainfall to global warming
.
Nat. Geosci.
,
6
,
357
361
, doi:.
Hwang
,
Y.-T.
,
D. M. W.
Frierson
, and
S. M.
Kang
,
2013
:
Anthropogenic sulphate aerosol and the southward shift of tropical precipitation in the late 20th century
.
Geophys. Res. Lett.
,
40
,
2845
2850
, doi:.
Inness
,
P.
, and
J.
Slingo
,
2003
:
Simulation of the Madden–Julian oscillation in a coupled general circulation model. Part I: Comparison with observations and an atmosphere-only GCM
.
J. Climate
,
16
,
345
364
, doi:.
IPCC
,
2014
: Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. C. B. Field et al., Eds., Cambridge University Press, 1132 pp. [Available online at http://www.ipcc.ch/pdf/assessment-report/ar5/wg2/WGIIAR5-PartA_FINAL.pdf.]
Joshi
,
M.
,
J.
Gregory
,
M.
Webb
,
D.
Sexton
, and
T.
Johns
,
2008
:
Mechanisms for the land/sea warming contrast exhibited by simulations of climate change
.
Climate Dyn.
,
30
,
455
465
, doi:.
Kent
,
C.
,
R.
Chadwick
, and
D.
Rowell
,
2015
:
Understanding uncertainties in future projections of seasonal tropical precipitation
.
J. Climate
,
28
,
4390
4413
, doi:.
Lambert
,
F. H.
,
M. J.
Webb
, and
M. M.
Joshi
,
2011
:
The relationship between land–ocean surface temperature contrast and radiative forcing
.
J. Climate
,
24
,
3239
3256
, doi:.
Liu
,
C.
, and
R. P.
Allan
,
2013
:
Observed and simulated precipitation responses in wet and dry regions
.
Environ. Res. Lett.
,
8
, 034002, doi:.
Long, S.-M., S.-P. Xie, and W. Liu
,
2016
:
Uncertainty in tropical rainfall projections: Atmospheric circulation effect and the ocean coupling
.
J. Climate
, doi:,
in press
.
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:.
Martin
,
G.
, and
R.
Levine
,
2012
:
The influence of dynamic vegetation on the present-day simulation and future projections of the South Asian summer monsoon in the HadGEM2 family
.
Earth Syst. Dyn.
,
3
,
245
261
, doi:.
Martin
,
G.
,
M.
Ringer
,
V.
Pope
,
A.
Jones
,
C.
Dearden
, and
T.
Hinton
,
2006
:
The physical properties of the atmosphere in the new Hadley Centre Global Environmental Model (HadGEM1). Part I: Model description and global climatology
.
J. Climate
,
19
,
1274
1301
, doi:.
McSweeney
,
C. F.
, and
R. G.
Jones
,
2013
:
No consensus on consensus: The challenge of finding a universal approach to measuring and mapping ensemble consistency in GCM projections
.
Climatic Change
,
119
,
617
629
, doi:.
Muller
,
C. J.
, and
P. A.
O’Gorman
,
2011
:
An energetic perspective on the regional response of precipitation to climate change
.
Nat. Climate Change
,
1
,
266
271
, doi:.
Neelin
,
J.
,
M.
Münnich
,
H.
Su
,
J.
Meyerson
, and
C.
Holloway
,
2006
:
Tropical drying trends in global warming models and observations
.
Proc. Natl. Acad. Sci. USA
,
103
,
6110
6115
, doi:.
Pennell
,
C.
, and
T.
Reichler
,
2011
:
On the effective number of climate models
.
J. Climate
,
24
,
2358
2367
, doi:.
Power
,
S.
,
F.
Delage
,
C.
Chung
,
G.
Kociuba
, and
K.
Keay
,
2013
:
Robust twenty-first-century projections of El Niño and related precipitation variability
.
Nature
,
502
,
541
545
, doi:.
Richardson
,
T.
,
P.
Forster
,
T.
Andrews
, and
D.
Parker
,
2016
:
Understanding the rapid precipitation response to CO2 and aerosol forcing on a regional scale
.
J. Climate
,
29
,
583
594
, doi:.
Roderick
,
M.
,
F.
Sun
,
W.
Lim
, and
G.
Farquhar
,
2014
:
A general framework for understanding the response of the water cycle to global warming over land and ocean
.
Hydrol. Earth Syst. Sci.
,
18
,
1575
1589
, doi:.
Rotstayn
,
L.
, and Coauthors
,
2007
:
Have Australian rainfall and cloudiness increased due to the remote effects of Asian anthropogenic aerosols?
J. Geophys. Res.
,
112
,
D09202
, doi:.
Rotstayn
,
L.
,
M.
Collier
, and
J.-J.
Luo
,
2015
:
Effects of declining aerosols on projections of zonally averaged tropical precipitation
.
Environ. Res. Lett.
,
10
, 044018, doi:.
Rowell
,
D.
,
2012
:
Sources of uncertainty in future changes in local precipitation
.
Climate Dyn.
,
39
,
1929
1950
, doi:.
Rowell
,
D.
,
2013
:
Simulating SST teleconnections to Africa: What is the state of the art?
J. Climate
,
26
,
5397
5418
, 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:.
Seo
,
J.
,
S. M.
Kamg
, and
D. M. W.
Frierson
,
2014
:
Sensitivity of intertropical convergence zone movement to the latitudinal position of thermal forcing
.
J. Climate
,
27
,
3035
3042
, doi:.
Shepherd
,
T.
,
2014
:
Atmospheric circulation as a source of uncertainty in climate change projections
.
Nat. Geosci.
,
7
,
703
708
, doi:.
Skinner
,
C.
,
M.
Ashfaq
, and
N.
Diffenbaugh
,
2012
:
Influence of twenty-first-century atmospheric and sea surface temperature forcing on West African climate
.
J. Climate
,
25
,
527
542
, doi:.
Solomon
,
S.
,
G.-K.
Plattner
,
R.
Knutti
, and
P.
Friedlingstein
,
2009
:
Irreversible climate change due to carbon dioxide emissions
.
Proc. Natl. Acad. Sci. USA
,
106
,
1704
1709
, doi:.
Sugi
,
M.
, and
J.
Yoshimura
,
2004
:
A mechanism of tropical precipitation change due to CO2 increase
.
J. Climate
,
17
,
238
243
, 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:.
Vellinga
,
M.
,
M.
Roberts
,
P.-L.
Vidale
,
M.
Mizielinski
,
M.-E.
Demory
,
R.
Schiemann
,
J.
Strachan
, and
C.
Bain
,
2016
:
Sahel decadal rainfall variability and the role of model horizontal resolution
.
Geophys. Res. Lett.
,
43
,
326
333
, doi:.
Voigt
,
A.
, and
T.
Shaw
,
2015
:
Circulation response to warming shaped by radiative changes of clouds and water vapour
.
Nat. Geosci.
,
8
,
102
106
, doi:.
Xie
,
S.-P.
,
C.
Deser
,
G. A.
Vecchi
,
J.
Ma
,
H.
Teng
, and
A. T.
Wittenberg
,
2010
:
Global warming pattern formation: Sea surface temperature and rainfall
.
J. Climate
,
23
,
966
986
, doi:.